[feat]: standalone QAD 5090 benchmark harness (multi-run stats)

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

.agents/skills/reseed-performance-baseline/SKILL.md

@@ -69,7 +69,7 @@ approval, then upload reviewed accepted baseline records.
 | `model_id` | Yes | Benchmark id, e.g. `wan-t2v-1.3b-2gpu`. This maps to the HF subdirectory after `sanitize(model_id)`. |
 | `gpu_type` | Yes | Exact GPU device string from the performance record, e.g. the L40S device name emitted by CI. Baselines are GPU-specific. |
 | `source_results` | Yes | One or more local paths or Buildkite artifact URLs for accepted shifted performance JSONs. Prefer normalized `normalized_perf_*.json` artifacts emitted by `compare_baseline.py`. Accept `source_result` as an alias only for a single JSON. |
-| `max_intra_batch_regression` | No | Maximum allowed regression of any source JSON against the source batch median. Default: `PERF_MAX_REGRESSION` if set, otherwise `0.05` (5%). |
+| `max_intra_batch_regression` | No | Maximum allowed regression of any source JSON against the source batch median. Default: `0.05` (5%). |
 | `intent_rationale` | Yes | One-line explanation for why the baseline shift is legitimate. This is written into provenance and should be reused in the PR. |
 
 Hardcoded defaults:
@@ -148,8 +148,7 @@ For each metric with at least two non-null source values:
 4. Stop if any source record regresses against the batch median by more than
    `max_intra_batch_regression`.
 
-Default `max_intra_batch_regression` to `PERF_MAX_REGRESSION` when set,
-otherwise `0.05`. Print a table with per-source values, batch median, and
+Default `max_intra_batch_regression` to `0.05`. Print a table with per-source values, batch median, and
 worst intra-batch regression.
 
 This check prevents uploading a mixed batch where one JSON is materially
@@ -183,7 +182,7 @@ present, that run is not a valid source for baseline reseeding.
 
 ### 2. Sync and back up existing HF records under /tmp
 
-Use `fastvideo/tests/performance/hf_store.py` helpers directly. Do **not** use
+Use `fastvideo/performance/hf_store.py` helpers directly. Do **not** use
 `compare_baseline.py` as a sync shortcut; on full main runs it can persist
 records, while this step must only fetch and back up existing history.
 
@@ -192,7 +191,7 @@ The sync command pattern is:
 ```bash
 export PERFORMANCE_TRACKING_ROOT="${PERFORMANCE_TRACKING_ROOT:-/tmp/perf-tracking}"
 export HF_REPO_ID="${HF_REPO_ID:-FastVideo/performance-tracking}"
-PYTHONPATH=fastvideo/tests/performance python -c 'from hf_store import sync_from_hf; import os; sync_from_hf(os.environ["PERFORMANCE_TRACKING_ROOT"], strict=True)'
+python -c 'from fastvideo.performance.hf_store import sync_from_hf; import os; sync_from_hf(os.environ["PERFORMANCE_TRACKING_ROOT"], strict=True)'
 ```
 
 Then back up only the sanitized model directory under `/tmp`:
@@ -200,8 +199,8 @@ Then back up only the sanitized model directory under `/tmp`:
 ```bash
 SHORT_COMMIT=$(git rev-parse --short=12 HEAD)
 TIMESTAMP=$(date -u +%Y%m%d_%H%M%S)
-MODEL_SAFE=$(PYTHONPATH=fastvideo/tests/performance python - <<'PY'
-from hf_store import sanitize
+MODEL_SAFE=$(python - <<'PY'
+from fastvideo.performance.hf_store import sanitize
 print(sanitize("<model_id>"))
 PY
 )
@@ -235,7 +234,7 @@ first baseline seed. Continue, but report that baseline history was empty.
 Load the last 5 successful records for the target:
 
 ```python
-from hf_store import load_records_for_model
+from fastvideo.performance.hf_store import load_records_for_model
 
 records = load_records_for_model(
     "/tmp/perf-tracking",
@@ -372,7 +371,7 @@ prepared records plus backup on disk.
 Use the shared storage helper so the path and repo type match CI:
 
 ```python
-from hf_store import upload_record
+from fastvideo.performance.hf_store import upload_record
 
 upload_record("<local_record_path>", record, strict=True)
 ```
@@ -460,7 +459,7 @@ directories created for this reseed. Never remove unrelated `/tmp` contents.
   intentional baseline replacement.
 - `fastvideo/tests/performance/compare_baseline.py` — normalization, rolling
   median comparison, and persistence rules.
-- `fastvideo/tests/performance/hf_store.py` — HF sync, record loading,
+- `fastvideo/performance/hf_store.py` — HF sync, record loading,
   `sanitize()`, and `upload_record()`.
 - `fastvideo/tests/performance/test_inference_performance.py` — source result
   JSON schema.

.buildkite/performance-benchmarks/tests/wan-t2v-1.3b.json

@@ -1,9 +1,9 @@
 {
   "benchmark_id": "wan-t2v-1.3b-2gpu",
   "config_schema_version": 2,
-  "workload_id": "wan-t2v-1.3b",
-  "variant_id": "canonical",
-  "benchmark_version": 1,
+  "workload_id": "wan-t2v",
+  "variant_id": "1.3b-sp2",
+  "benchmark_version": 2,
   "description": "Wan2.1 T2V 1.3B inference performance",
   "model": {
     "model_path": "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",

.buildkite/pipeline.yml

@@ -436,7 +436,7 @@ steps:
                 - "pyproject.toml"
                 - "docker/Dockerfile"
               config:
-                command: "timeout 15m .buildkite/scripts/pr_test.sh"
+                command: "timeout 25m .buildkite/scripts/pr_test.sh"
                 label: ":test_tube: LoRA Training Tests"
                 env:
                   - TEST_TYPE=training_lora
@@ -481,6 +481,7 @@ steps:
                 - "fastvideo/layers/**"
                 - "fastvideo/worker/**"
                 - "fastvideo/entrypoints/**"
+                - "fastvideo/performance/**"
                 - "fastvideo/tests/performance/**"
                 - ".buildkite/performance-benchmarks/**"
                 - "pyproject.toml"

.buildkite/scripts/pr_test.sh

@@ -76,10 +76,27 @@ EFFECTIVE_PR=${BUILDKITE_PULL_REQUEST:-false}
 if [ "$EFFECTIVE_PR" = "false" ] && [ -n "${PR_NUMBER:-}" ]; then
     EFFECTIVE_PR=$PR_NUMBER
 fi
-MODAL_ENV="BUILDKITE_REPO=$BUILDKITE_REPO BUILDKITE_COMMIT=$BUILDKITE_COMMIT BUILDKITE_PULL_REQUEST=$EFFECTIVE_PR BUILDKITE_BRANCH=${BUILDKITE_BRANCH:-} TEST_SCOPE=${TEST_SCOPE:-} BUILDKITE_BUILD_URL=${BUILDKITE_BUILD_URL:-} BUILDKITE_BUILD_ID=${BUILDKITE_BUILD_ID:-} BUILDKITE_JOB_ID=${BUILDKITE_JOB_ID:-} IMAGE_VERSION=$IMAGE_VERSION"
+MODAL_ENV="BUILDKITE_REPO=$BUILDKITE_REPO BUILDKITE_COMMIT=$BUILDKITE_COMMIT BUILDKITE_PULL_REQUEST=$EFFECTIVE_PR BUILDKITE_BRANCH=${BUILDKITE_BRANCH:-} BUILDKITE_SOURCE=${BUILDKITE_SOURCE:-} TEST_SCOPE=${TEST_SCOPE:-} BUILDKITE_BUILD_URL=${BUILDKITE_BUILD_URL:-} BUILDKITE_BUILD_ID=${BUILDKITE_BUILD_ID:-} BUILDKITE_JOB_ID=${BUILDKITE_JOB_ID:-} IMAGE_VERSION=$IMAGE_VERSION"
 
 POST_RUN_HOOK=""
 
+is_truthy() {
+    case "${1:-}" in
+        1|true|TRUE|yes|YES|on|ON) return 0 ;;
+        *) return 1 ;;
+    esac
+}
+
+ssim_bootstrap_args() {
+    local title="${PR_TITLE:-}"
+    local message="${BUILDKITE_MESSAGE:-}"
+    if is_truthy "${FASTVIDEO_SSIM_BOOTSTRAP_MODE:-}" \
+        || [[ "$title" == *"[new-model]"* ]] \
+        || [[ "$message" == *"[new-model]"* ]]; then
+        printf ' --bootstrap-mode'
+    fi
+}
+
 upload_performance_artifacts() {
     SHORT_SHA=${BUILDKITE_COMMIT:0:7}
     LOCAL_DIR="downloaded_reports"
@@ -172,7 +189,12 @@ case "$TEST_TYPE" in
         ;;
     "ssim")
         log "Running SSIM tests..."
-        MODAL_COMMAND="$MODAL_ENV HF_API_KEY=$HF_API_KEY python3 -m modal run $MODAL_SSIM_TEST_FILE::run_ssim_tests"
+        SSIM_BOOTSTRAP_ARGS=$(ssim_bootstrap_args)
+        if [ -n "$SSIM_BOOTSTRAP_ARGS" ]; then
+            log "SSIM bootstrap mode enabled for new-model reference draft generation"
+        fi
+        MODAL_COMMAND="$MODAL_ENV HF_API_KEY=$HF_API_KEY python3 -m modal run "
+        MODAL_COMMAND+="$MODAL_SSIM_TEST_FILE::run_ssim_tests$SSIM_BOOTSTRAP_ARGS"
         ;;
     "training")
         log "Running training tests..."

.github/scripts/gate_full_suite.sh

@@ -0,0 +1,133 @@
+#!/usr/bin/env bash
+# Gate the expensive Buildkite full suite on the cheap GitHub checks.
+#
+# Polls the workflow runs for the PR head commit and only exits 0 once the
+# watched cheap workflows (pre-commit, docs build) have succeeded, so the
+# 'ready' label cannot burn ~20 GPU lanes on a head that a cheap check has
+# already doomed.
+#
+# Semantics:
+#   - watched run completed with a bad conclusion  -> exit 1 (fail CLOSED:
+#     no full suite; the next push re-arms via the 'synchronize' trigger)
+#   - watched run cancelled                         -> still pending: the docs
+#     workflow's repo-global 'pages' concurrency group cancels runs superseded
+#     by unrelated pushes, so 'cancelled' is not a verdict on this PR
+#   - watched runs pending                          -> poll until done
+#   - docs run absent                               -> not applicable after a
+#     short grace period ('Deploy Documentation' is path-filtered on PRs)
+#   - pre-commit run absent                         -> keep polling: pre-commit
+#     is never path-filtered, so its absence is always anomalous
+#   - 'ready' label removed while waiting           -> exit 1 (fail CLOSED:
+#     un-labeling is a deliberate maintainer action)
+#   - GitHub API unreachable or timeout             -> exit 0 (fail OPEN,
+#     loud warning: never brick CI on a GitHub outage)
+#
+# Required env: PR_SHA (PR head commit), PR_NUMBER, GITHUB_REPOSITORY, GH_TOKEN.
+set -euo pipefail
+
+: "${PR_SHA:?PR_SHA (PR head commit) is required}"
+: "${PR_NUMBER:?PR_NUMBER (pull request number) is required}"
+: "${GITHUB_REPOSITORY:?GITHUB_REPOSITORY is required}"
+
+# Workflow-level `name:` values that must be green before the full suite
+# may start. "Deploy Documentation" is path-filtered on PRs, so its run may
+# legitimately never exist; pre-commit always runs, so it must appear.
+WATCHED_NAMES='["pre-commit", "Deploy Documentation"]'
+WATCHED_REGEX='^(pre-commit|Deploy Documentation)$'
+POLL_SECS="${POLL_SECS:-20}"
+GRACE_SECS="${GRACE_SECS:-60}"
+MAX_WAIT_SECS="${MAX_WAIT_SECS:-1500}"
+
+# Bound each API call so a hung connection hits the 3-strike fail-open path
+# instead of pinning the loop until the job timeout (which would fail closed
+# on exactly the GitHub-outage case this script is meant to survive).
+if command -v timeout >/dev/null 2>&1; then
+  gh_api() { timeout 30 gh api "$@"; }
+else
+  gh_api() { gh api "$@"; }  # macOS dev boxes; CI always has coreutils timeout
+fi
+
+# The workflow checked the label before starting the gate, but the wait can
+# last ~25 min: re-check once before any exit 0 and fail closed if 'ready'
+# was removed in the meantime. An API error here proceeds (the label was
+# present when the gate started; never brick CI on an outage).
+recheck_ready_label() {
+  local pr_json
+  if pr_json=$(gh_api "repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}" 2>/dev/null); then
+    if ! jq -e '[.labels[]?.name] | index("ready")' <<<"$pr_json" >/dev/null 2>&1; then
+      echo "::error::PR #${PR_NUMBER} no longer has the 'ready' label —" \
+        "NOT triggering the Buildkite full suite. Re-add the label to re-arm."
+      exit 1
+    fi
+  else
+    echo "::warning::Could not re-check the 'ready' label on PR #${PR_NUMBER}; proceeding (it was present when the gate started)."
+  fi
+}
+
+start=$(date +%s)
+api_fails=0
+missing=""
+
+while true; do
+  elapsed=$(( $(date +%s) - start ))
+
+  if runs_json=$(gh_api "repos/${GITHUB_REPOSITORY}/actions/runs?head_sha=${PR_SHA}&per_page=100" 2>/dev/null) \
+     && state=$(jq --arg re "$WATCHED_REGEX" '
+          [.workflow_runs[]? | select(.name // "" | test($re))]
+          | group_by(.name) | map(max_by(.id))
+          | map({name, status, conclusion})' <<<"$runs_json" 2>/dev/null); then
+    api_fails=0
+    echo "t+${elapsed}s watched checks: $(jq -c . <<<"$state")"
+
+    failed=$(jq -r '[.[] | select(.status == "completed"
+          and (.conclusion | IN("success", "skipped", "neutral", "cancelled") | not))]
+        | map(.name) | join(", ")' <<<"$state")
+    if [ -n "$failed" ]; then
+      echo "::error::Cheap check(s) failed on ${PR_SHA}: ${failed}." \
+        "NOT triggering the Buildkite full suite. Push a fix (the 'ready'" \
+        "label re-arms on every push), or re-run the failed check and then" \
+        "re-run this workflow."
+      exit 1
+    fi
+
+    # 'cancelled' counts as pending: wait for a re-run to reach a real verdict
+    # (bounded by MAX_WAIT, then the fail-open below).
+    pending=$(jq '[.[] | select(.status != "completed" or .conclusion == "cancelled")] | length' <<<"$state")
+    missing=$(jq -r --argjson watched "$WATCHED_NAMES" '($watched - map(.name)) | join(", ")' <<<"$state")
+    if [ "$pending" -eq 0 ]; then
+      if [ -z "$missing" ]; then
+        recheck_ready_label
+        echo "All watched cheap checks are green — full suite may proceed."
+        exit 0
+      fi
+      case "$missing" in
+        *pre-commit*)
+          echo "pre-commit run not found for ${PR_SHA} yet; waiting (pre-commit is never path-filtered, so its absence is anomalous)."
+          ;;
+        *)
+          if [ "$elapsed" -ge "$GRACE_SECS" ]; then
+            recheck_ready_label
+            echo "::warning::Watched run(s) never appeared for ${PR_SHA}: ${missing} (path-filtered, likely not applicable). Proceeding on the checks that did run."
+            exit 0
+          fi
+          echo "Waiting up to ${GRACE_SECS}s grace for path-filtered run(s) to appear: ${missing}."
+          ;;
+      esac
+    fi
+  else
+    api_fails=$(( api_fails + 1 ))
+    echo "::warning::GitHub API error querying workflow runs for ${PR_SHA} (attempt ${api_fails}/3)."
+    if [ "$api_fails" -ge 3 ]; then
+      recheck_ready_label
+      echo "::warning::FAILING OPEN: cannot query GitHub check status — triggering the full suite WITHOUT the cheap-check gate."
+      exit 0
+    fi
+  fi
+
+  if [ "$elapsed" -ge "$MAX_WAIT_SECS" ]; then
+    recheck_ready_label
+    echo "::warning::FAILING OPEN: watched checks still pending after $(( MAX_WAIT_SECS / 60 )) min${missing:+ (never appeared: ${missing})} — triggering the full suite anyway."
+    exit 0
+  fi
+  sleep "$POLL_SECS"
+done

.github/scripts/test_gate_full_suite.sh

@@ -0,0 +1,122 @@
+#!/usr/bin/env bash
+# Self-test for gate_full_suite.sh using a mocked `gh`. No network, runs on
+# any dev box: bash .github/scripts/test_gate_full_suite.sh
+set -u
+here=$(cd "$(dirname "$0")" && pwd)
+tmp=$(mktemp -d)
+trap 'rm -rf "$tmp"' EXIT
+
+# Mock gh. Asserts the exact endpoint (including head_sha) it is called
+# with — an endpoint typo in the gate script fails the test rather than
+# silently serving canned data. On the runs endpoint it serves
+# $MOCK_DIR/response_<call#>.json, sticking on the highest existing file,
+# and exits 1 if none exist (simulates a GitHub API outage). On the pulls
+# endpoint it serves $MOCK_DIR/pr.json, defaulting to a 'ready'-labeled PR.
+cat > "$tmp/gh" <<'EOF'
+#!/usr/bin/env bash
+if [ "${1:-}" != "api" ]; then
+  echo "unexpected gh invocation: $*" >> "$MOCK_DIR/endpoint_error"
+  exit 2
+fi
+case "${2:-}" in
+  "repos/o/r/actions/runs?head_sha=deadbeef&per_page=100")
+    n=$(( $(cat "$MOCK_DIR/count" 2>/dev/null || echo 0) + 1 ))
+    echo "$n" > "$MOCK_DIR/count"
+    while [ "$n" -gt 0 ]; do
+      if [ -f "$MOCK_DIR/response_$n.json" ]; then
+        cat "$MOCK_DIR/response_$n.json"
+        exit 0
+      fi
+      n=$(( n - 1 ))
+    done
+    echo "api outage" >&2
+    exit 1
+    ;;
+  "repos/o/r/pulls/42")
+    if [ -f "$MOCK_DIR/pr.json" ]; then
+      cat "$MOCK_DIR/pr.json"
+    else
+      echo '{"labels": [{"name": "ready"}]}'
+    fi
+    ;;
+  *)
+    echo "unexpected gh endpoint: $2" >> "$MOCK_DIR/endpoint_error"
+    exit 2
+    ;;
+esac
+EOF
+chmod +x "$tmp/gh"
+
+PC_OK='{"name": "pre-commit", "id": 1, "status": "completed", "conclusion": "success"}'
+PC_BAD='{"name": "pre-commit", "id": 1, "status": "completed", "conclusion": "failure"}'
+PC_PENDING='{"name": "pre-commit", "id": 1, "status": "in_progress", "conclusion": null}'
+DOCS_OK='{"name": "Deploy Documentation", "id": 2, "status": "completed", "conclusion": "success"}'
+DOCS_BAD='{"name": "Deploy Documentation", "id": 2, "status": "completed", "conclusion": "failure"}'
+DOCS_CANCELLED='{"name": "Deploy Documentation", "id": 2, "status": "completed", "conclusion": "cancelled"}'
+OTHER='{"name": "Trigger Full Suite", "id": 3, "status": "in_progress", "conclusion": null}'
+NULL_NAME='{"name": null, "id": 4, "status": "completed", "conclusion": "failure"}'
+PC_OK_RERUN='{"name": "pre-commit", "id": 5, "status": "completed", "conclusion": "success"}'
+
+fails=0
+want_log=""  # optional: expect() also greps out.log for this regex, then resets
+pr_json=""   # optional: served for the pulls (label re-check) endpoint, then resets
+raw_body=""  # optional: serve responses verbatim instead of wrapping in workflow_runs
+expect() { # <name> <expected-exit> <response json>...
+  local name=$1 want=$2 dir i=1
+  shift 2
+  dir=$(mktemp -d "$tmp/test_XXXXXX")
+  for body in "$@"; do
+    if [ -n "$raw_body" ]; then
+      printf '%s' "$body" > "$dir/response_$i.json"
+    else
+      printf '{"workflow_runs": [%s]}' "$body" > "$dir/response_$i.json"
+    fi
+    i=$(( i + 1 ))
+  done
+  [ -n "$pr_json" ] && printf '%s' "$pr_json" > "$dir/pr.json"
+  ( export PATH="$tmp:$PATH" MOCK_DIR="$dir" PR_SHA=deadbeef PR_NUMBER=42 \
+      GITHUB_REPOSITORY=o/r POLL_SECS=0 GRACE_SECS=1 MAX_WAIT_SECS=3
+    bash "$here/gate_full_suite.sh" > "$dir/out.log" 2>&1 )
+  local rc=$?
+  if [ "$rc" -ne "$want" ]; then
+    echo "FAIL: $name (exit $rc, want $want)"
+    cat "$dir/out.log"
+    fails=1
+  elif [ -f "$dir/endpoint_error" ]; then
+    echo "FAIL: $name (mock gh got an unexpected call)"
+    cat "$dir/endpoint_error"
+    fails=1
+  elif [ -n "$want_log" ] && ! grep -Eq "$want_log" "$dir/out.log"; then
+    echo "FAIL: $name (log does not match: $want_log)"
+    cat "$dir/out.log"
+    fails=1
+  else
+    echo "ok: $name"
+  fi
+  want_log="" pr_json="" raw_body=""
+}
+
+expect "both green -> proceed" 0 "$PC_OK, $DOCS_OK, $OTHER, $NULL_NAME"
+expect "docs build failed -> blocked" 1 "$PC_OK, $DOCS_BAD"
+expect "pre-commit failed -> blocked" 1 "$PC_BAD"
+expect "pending then green -> proceed" 0 "$PC_PENDING" "$PC_OK, $DOCS_OK"
+want_log="never appeared.*Deploy Documentation"
+expect "docs run absent (path-filtered) -> proceed after grace" 0 "$PC_OK"
+expect "API outage -> fail open" 0
+want_log="FAILING OPEN"
+expect "pending past MAX_WAIT -> fail open" 0 "$PC_PENDING"
+want_log="FAILING OPEN"
+expect "unrelated runs only -> no grace, fail open at MAX_WAIT" 0 "$OTHER"
+expect "cancelled docs then green -> proceed" 0 \
+  "$PC_OK, $DOCS_CANCELLED" "$PC_OK, $DOCS_OK"
+want_log="FAILING OPEN"
+expect "cancelled docs forever -> fail open at MAX_WAIT" 0 "$PC_OK, $DOCS_CANCELLED"
+want_log="FAILING OPEN"
+expect "pre-commit absent -> no grace, fail open at MAX_WAIT" 0 "$DOCS_OK"
+expect "duplicate run names -> latest wins" 0 "$PC_BAD, $PC_OK_RERUN, $DOCS_OK"
+raw_body=1
+expect "garbage response body -> fail open" 0 "this is not json"
+pr_json='{"labels": [{"name": "other"}]}'
+expect "ready label removed mid-gate -> blocked" 1 "$PC_OK, $DOCS_OK"
+
+exit "$fails"

.github/workflows/ci-precommit.yml

@@ -1,7 +1,11 @@
 name: pre-commit
 
 on:
-  pull_request:
+  # pull_request_target instead of pull_request: the workflow definition and
+  # the hook config are always taken from the BASE branch, so fork /
+  # first-time-contributor PRs run immediately without a maintainer clicking
+  # "Approve and run". The PR head is checked out as data only.
+  pull_request_target:
     branches: [main]
   workflow_call:
     inputs:
@@ -15,12 +19,25 @@ permissions:
 
 jobs:
   pre-commit:
-    if: github.event_name == 'workflow_call' || github.event.pull_request.draft != true
+    if: github.event.pull_request.draft != true
     runs-on: ubuntu-latest
     steps:
     - uses: actions/checkout@v4
       with:
         ref: ${{ inputs.ref || '' }}
+    # For PR events, lint the PR head — but keep the hook definitions from
+    # the base branch so an untrusted PR cannot alter what gets executed.
+    - name: Save trusted hook config
+      if: github.event_name == 'pull_request_target'
+      run: cp .pre-commit-config.yaml "$RUNNER_TEMP/trusted-pre-commit-config.yaml"
+    - uses: actions/checkout@v4
+      if: github.event_name == 'pull_request_target'
+      with:
+        ref: ${{ github.event.pull_request.head.sha }}
+        persist-credentials: false
+    - name: Restore trusted hook config
+      if: github.event_name == 'pull_request_target'
+      run: cp "$RUNNER_TEMP/trusted-pre-commit-config.yaml" .pre-commit-config.yaml
     - uses: actions/setup-python@v5
       with:
         python-version: "3.12"
@@ -30,3 +47,6 @@ jobs:
     - uses: pre-commit/action@v3.0.1
       with:
         extra_args: --all-files --hook-stage manual
+    # After pre-commit so a self-test failure cannot mask lint failures.
+    - name: Full-suite gate self-test
+      run: bash .github/scripts/test_gate_full_suite.sh

.github/workflows/ci-slash-commands.yml

@@ -52,6 +52,7 @@ jobs:
             core.setOutput('pr_sha', pr.head.sha);
             core.setOutput('pr_branch', pr.head.ref);
             core.setOutput('pr_number', String(prNumber));
+            core.setOutput('pr_title', pr.title);
 
       - name: Trigger Full Suite
         if: steps.perm.outputs.has_write == 'true'
@@ -60,6 +61,7 @@ jobs:
           PR_SHA: ${{ steps.label.outputs.pr_sha }}
           PR_BRANCH: ${{ steps.label.outputs.pr_branch }}
           PR_NUMBER: ${{ steps.label.outputs.pr_number }}
+          PR_TITLE: ${{ steps.label.outputs.pr_title }}
           BK_ORG: ${{ vars.BUILDKITE_ORG_SLUG }}
           BK_PIPELINE: ${{ vars.BUILDKITE_PIPELINE_SLUG }}
         run: |
@@ -71,6 +73,7 @@ jobs:
               --arg commit "$PR_SHA" \
               --arg branch "$PR_BRANCH" \
               --arg message "Full Suite for PR #${PR_NUMBER} (via /merge)" \
+              --arg pr_title "$PR_TITLE" \
               --argjson pr_id "$PR_NUMBER" \
               '{
                 commit: $commit,
@@ -80,11 +83,12 @@ jobs:
                 pull_request_id: $pr_id,
                 pull_request_base_branch: "main",
                 env: {
-                  TEST_SCOPE: "full",
-                  FULL_SUITE: "true",
-                  PR_NUMBER: ($pr_id | tostring)
-                }
-              }')"
+                    TEST_SCOPE: "full",
+                    FULL_SUITE: "true",
+                    PR_NUMBER: ($pr_id | tostring),
+                    PR_TITLE: $pr_title
+                  }
+                }')"
 
   parse-command:
     if: >-
@@ -241,6 +245,7 @@ jobs:
           TEST_SCOPE: ${{ needs.parse-command.outputs.test_scope }}
           FULL_SUITE: ${{ needs.parse-command.outputs.full_suite }}
           TEST_TYPE: ${{ needs.parse-command.outputs.test_type }}
+          PR_TITLE: ${{ github.event.issue.title }}
           BK_ORG: ${{ vars.BUILDKITE_ORG_SLUG }}
           BK_PIPELINE: ${{ vars.BUILDKITE_PIPELINE_SLUG }}
         run: |
@@ -257,6 +262,7 @@ jobs:
               --arg full_suite "$FULL_SUITE" \
               --arg test_type "$TEST_TYPE" \
               --arg pr_number "$PR_NUMBER" \
+              --arg pr_title "$PR_TITLE" \
               '{
                 commit: $commit,
                 branch: $branch,
@@ -266,8 +272,9 @@ jobs:
                 pull_request_base_branch: "main",
                 env: {
                   TEST_SCOPE: $test_scope,
-                  FULL_SUITE: $full_suite,
-                  TEST_TYPE: $test_type,
-                  PR_NUMBER: $pr_number
-                }
-              }')"
+                    FULL_SUITE: $full_suite,
+                    TEST_TYPE: $test_type,
+                    PR_NUMBER: $pr_number,
+                    PR_TITLE: $pr_title
+                  }
+                }')"

.github/workflows/ci-trigger-full-suite.yml

@@ -7,6 +7,7 @@ on:
 permissions:
   contents: read
   pull-requests: read
+  actions: read
 
 concurrency:
   group: full-suite-${{ github.event.pull_request.number }}
@@ -18,6 +19,8 @@ jobs:
       (github.event.action == 'labeled' && github.event.label.name == 'ready')
       || github.event.action == 'synchronize'
     runs-on: ubuntu-latest
+    # Gate below may wait for cheap checks (up to MAX_WAIT_SECS = 25 min).
+    timeout-minutes: 35
     steps:
       - name: Check ready label
         id: check
@@ -49,6 +52,20 @@ jobs:
               "https://api.buildkite.com/v2/organizations/${{ vars.BUILDKITE_ORG_SLUG }}/pipelines/${{ vars.BUILDKITE_PIPELINE_SLUG }}/builds/${build_num}/cancel"
           done
 
+      # Checks out the BASE branch (default for pull_request_target), so PR
+      # authors cannot tamper with the gate script.
+      - name: Checkout gate script
+        if: steps.check.outputs.has_ready == 'true'
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+
+      - name: Wait for pre-commit and docs build
+        if: steps.check.outputs.has_ready == 'true'
+        env:
+          GH_TOKEN: ${{ github.token }}
+          PR_SHA: ${{ github.event.pull_request.head.sha }}
+          PR_NUMBER: ${{ github.event.pull_request.number }}
+        run: bash .github/scripts/gate_full_suite.sh
+
       - name: Trigger Buildkite Full Suite
         if: steps.check.outputs.has_ready == 'true'
         env:
@@ -56,6 +73,7 @@ jobs:
           PR_SHA: ${{ github.event.pull_request.head.sha }}
           PR_BRANCH: ${{ github.event.pull_request.head.ref }}
           PR_NUMBER: ${{ github.event.pull_request.number }}
+          PR_TITLE: ${{ github.event.pull_request.title }}
           BK_ORG: ${{ vars.BUILDKITE_ORG_SLUG }}
           BK_PIPELINE: ${{ vars.BUILDKITE_PIPELINE_SLUG }}
         run: |
@@ -67,6 +85,7 @@ jobs:
               --arg commit "$PR_SHA" \
               --arg branch "$PR_BRANCH" \
               --arg message "Full Suite for PR #${PR_NUMBER}" \
+              --arg pr_title "$PR_TITLE" \
               --argjson pr_id "$PR_NUMBER" \
               '{
                 commit: $commit,
@@ -78,6 +97,7 @@ jobs:
                 env: {
                   TEST_SCOPE: "full",
                   FULL_SUITE: "true",
-                  PR_NUMBER: ($pr_id | tostring)
+                  PR_NUMBER: ($pr_id | tostring),
+                  PR_TITLE: $pr_title
                 }
               }')"

.gitignore

@@ -72,8 +72,7 @@ docs/distillation/examples/
 # Python pickle files
 *.pkl
 
-# Reference videos
-!fastvideo/tests/ssim/reference_videos/**/*.mp4
+# Reference videos (negations must come after the catch-all on line below)
 
 # Static images
 !docs/assets/images/**/*.png
@@ -127,6 +126,8 @@ apps/dreamverse/web/.env.production.local
 .sisyphus/
 openspec/
 fastvideo/tests/ssim/reference_videos/**
+!fastvideo/tests/ssim/reference_videos/**/*.mp4
+!fastvideo/tests/ssim/reference_videos/**/*.png
 
 # Editor logs and local Python version pins (accidentally committed)
 *.nvimlog

apps/performance_dashboard/frontend/src/App.tsx

@@ -1,6 +1,7 @@
 import { useEffect, useMemo, useState } from "react";
 
-import { fetchSummary, fetchTrends, refreshData, RunSource, SummaryResponse, TrendGroup, TrendPoint } from "./api";
+import { fetchSummary, fetchTrends, refreshData } from "./api";
+import type { CohortValue, RunSource, SummaryResponse, TrendGroup, TrendPoint } from "./api";
 
 const METRIC_KEYS = ["latency", "throughput", "memory", "text_encoder_time_s", "dit_time_s", "vae_decode_time_s"];
 const RUN_SOURCES: Array<{ value: "" | RunSource; label: string }> = [
@@ -109,6 +110,61 @@ function metricLabel(metricKey: string) {
   return METRIC_DEFINITIONS[metricKey]?.label ?? metricKey;
 }
 
+type CohortFields = {
+  model_id: string;
+  gpu_type: string;
+  workload_id: CohortValue;
+  variant_id: CohortValue;
+  benchmark_version: CohortValue;
+  recipe_fingerprint: CohortValue;
+  hardware_profile_id: CohortValue;
+  software_profile_id: CohortValue;
+};
+
+function cohortValue(value: CohortValue) {
+  if (value === null || value === undefined || value === "") {
+    return "legacy";
+  }
+  return String(value);
+}
+
+function shortCohortValue(value: CohortValue) {
+  const text = cohortValue(value);
+  if (text === "legacy" || text.length <= 14) {
+    return text;
+  }
+  return text.slice(0, 12);
+}
+
+function cohortKey(cohort: CohortFields) {
+  return [
+    cohort.model_id,
+    cohort.gpu_type,
+    cohortValue(cohort.workload_id),
+    cohortValue(cohort.variant_id),
+    cohortValue(cohort.benchmark_version),
+    cohortValue(cohort.recipe_fingerprint),
+    cohortValue(cohort.hardware_profile_id),
+    cohortValue(cohort.software_profile_id)
+  ].join("|");
+}
+
+function cohortTitle(cohort: CohortFields) {
+  const workload = cohortValue(cohort.workload_id);
+  const variant = cohortValue(cohort.variant_id);
+  const version = cohortValue(cohort.benchmark_version);
+  const versionLabel = version === "legacy" ? version : `v${version}`;
+  return `${workload} / ${variant} / ${versionLabel}`;
+}
+
+function cohortDetail(cohort: CohortFields) {
+  return [
+    `recipe ${shortCohortValue(cohort.recipe_fingerprint)}`,
+    shortCohortValue(cohort.hardware_profile_id),
+    shortCohortValue(cohort.software_profile_id)
+  ].join(" | ");
+}
+
 function formatMetricValue(metricKey: string, value: number | null | undefined, tooltip = false) {
   const definition = METRIC_DEFINITIONS[metricKey];
   if (!definition) {
@@ -171,7 +227,9 @@ function TrendChart({ group, metricKey }: { group: TrendGroup; metricKey: string
         top: `${(activePoint.y / height) * 100}%`
       }
     : undefined;
-  const ariaLabel = `${metricLabel(metricKey)} trend for ${group.model_id} on ${group.gpu_type}`;
+  const ariaLabel = `${metricLabel(metricKey)} trend for ${group.model_id} on ${group.gpu_type}, ${cohortTitle(
+    group
+  )}`;
 
   return (
     <div className="chart-shell">
@@ -419,7 +477,7 @@ export default function App() {
       <section className="panel">
         <div className="panel-header">
           <h2>Latest Status</h2>
-          <span>{latestRows.length} model/GPU groups</span>
+          <span>{latestRows.length} comparison cohorts</span>
         </div>
         {latestRows.length === 0 ? (
           <div className="empty">No records match the selected filters.</div>
@@ -432,6 +490,7 @@ export default function App() {
                   <th>Recomputed</th>
                   <th>Model</th>
                   <th>GPU</th>
+                  <th>Cohort</th>
                   <th>Commit</th>
                   <th>Source</th>
                   <th>Baseline</th>
@@ -446,7 +505,7 @@ export default function App() {
               </thead>
               <tbody>
                 {latestRows.map((row) => (
-                  <tr key={`${row.model_id}-${row.gpu_type}`}>
+                  <tr key={cohortKey(row)}>
                     <td>
                       <span className={`badge ${row.status}`}>{row.status}</span>
                     </td>
@@ -457,6 +516,12 @@ export default function App() {
                     </td>
                     <td>{row.model_id}</td>
                     <td>{row.gpu_type}</td>
+                    <td>
+                      <div className="cohort-cell">
+                        <strong>{cohortTitle(row)}</strong>
+                        <span>{cohortDetail(row)}</span>
+                      </div>
+                    </td>
                     <td>{shortSha(row.commit_sha)}</td>
                     <td>
                       <span className={`source-badge source-${row.run_source}`}>{runSourceLabel(row.run_source)}</span>
@@ -495,11 +560,13 @@ export default function App() {
           ) : (
             trends.map((group) =>
               METRIC_KEYS.map((metricKey) => (
-                <article className="trend-card" key={`${group.model_id}-${group.gpu_type}-${metricKey}`}>
+                <article className="trend-card" key={`${cohortKey(group)}-${metricKey}`}>
                   <div>
                     <h3>{metricLabel(metricKey)}</h3>
                     <p>
                       {group.model_id} | {group.gpu_type}
+                      <span>{cohortTitle(group)}</span>
+                      <span>{cohortDetail(group)}</span>
                     </p>
                   </div>
                   <TrendChart group={group} metricKey={metricKey} />

apps/performance_dashboard/frontend/src/api.ts

@@ -13,6 +13,17 @@ export type MetricValue = {
   precision: number;
 };
 
+export type CohortValue = string | number | null;
+
+export type ComparisonCohort = {
+  workload_id: CohortValue;
+  variant_id: CohortValue;
+  benchmark_version: CohortValue;
+  recipe_fingerprint: CohortValue;
+  hardware_profile_id: CohortValue;
+  software_profile_id: CohortValue;
+};
+
 export type SummaryRow = {
   model_id: string;
   gpu_type: string;
@@ -34,7 +45,7 @@ export type SummaryRow = {
   build_id: string;
   job_id: string;
   metrics: Record<string, MetricValue>;
-};
+} & ComparisonCohort;
 
 export type RunSource = "pr" | "local" | "scheduled_main" | "unknown";
 
@@ -68,13 +79,13 @@ export type TrendPoint = {
   build_id: string;
   job_id: string;
   metrics: Record<string, number | null>;
-};
+} & ComparisonCohort;
 
 export type TrendGroup = {
   model_id: string;
   gpu_type: string;
   points: TrendPoint[];
-};
+} & ComparisonCohort;
 
 export type TrendsResponse = {
   groups: TrendGroup[];

apps/performance_dashboard/frontend/src/styles.css

@@ -149,6 +149,11 @@ h3 {
   font-size: 0.82rem;
 }
 
+.trend-card p {
+  display: grid;
+  gap: 2px;
+}
+
 .stat strong {
   display: block;
   margin-top: 8px;
@@ -186,7 +191,7 @@ h3 {
 
 table {
   width: 100%;
-  min-width: 1120px;
+  min-width: 1260px;
   border-collapse: collapse;
 }
 
@@ -209,6 +214,25 @@ td {
   font-size: 0.9rem;
 }
 
+.cohort-cell {
+  display: grid;
+  gap: 2px;
+}
+
+.cohort-cell strong,
+.trend-card p span {
+  color: #1b2836;
+  font-size: 0.78rem;
+  font-weight: 700;
+}
+
+.cohort-cell span,
+.trend-card p span + span {
+  color: #607080;
+  font-family: ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", monospace;
+  font-size: 0.72rem;
+}
+
 .badge {
   display: inline-flex;
   align-items: center;

assets/eval/worldmodel_synthetic_flow_calibration.json

@@ -0,0 +1,22 @@
+{
+  "alpha_yaw": 0.08734091699186919,
+  "alpha_pitch": 0.08169667696275307,
+  "alpha_turn": 5.724587470723463e-17,
+  "beta_fwd": 0.02842768078408099,
+  "beta_strafe": 0.022531015077067108,
+  "focal_length": 457.0,
+  "frame_shape": [
+    352,
+    640
+  ],
+  "calibrated_from": [
+    "1_wasd_only",
+    "camera",
+    "camera4hold_alpha1",
+    "fully_random",
+    "wasdonly_alpha1",
+    "wasd4holdrandview_simple_1key1mouse1"
+  ],
+  "residual_rms": 15.890399609478676,
+  "n_equations": 4125232
+}

docker/Dockerfile

@@ -68,7 +68,7 @@ ARG FLASH_ATTN_WHEEL_RELEASE_ARM64=https://github.com/mjun0812/flash-attention-p
 # cutlass-4.4 `cute.core.ThrMma` API, which crashes on the cutlass-dsl 4.5 that
 # flashinfer/quack pull in. After the wheel install we overlay this cutlass-4.5-safe
 # upstream cute (flash-attn-4) so the image runs FA4 instead of the FA2 fallback.
-ARG FA4_CUTE_REF=940cd9680f3315f2f06b43ab5bea2c2cf2d96806
+ARG FA4_CUTE_REF=82d6441eec5d4dfec120153db2c0145ae855a083
 
 # Provided automatically by BuildKit/buildx (e.g. "amd64" / "arm64") and used to
 # select the prebuilt flash-attn wheel. Empty under a plain `docker build` without
@@ -161,7 +161,7 @@ RUN --mount=type=cache,target=/opt/uv/cache \
         uv pip install flash-attn==${FLASH_ATTN_VERSION} --no-build-isolation; \
     fi
 
-# Overlay the cutlass-4.5-safe upstream FA4 cute (FA4_CUTE_REF) over the
+# Overlay the CuTe-DSL-4.6-compatible upstream FA4 cute (FA4_CUTE_REF) over the
 # wheel/source one so the image runs FA4, not the FA2 fallback. This pulls the FA4
 # runtime stack (cutlass-dsl, quack-kernels, apache-tvm-ffi, torch-c-dlpack-ext) --
 # the same deps the [dreamverse] extra already installs in CI; the installed torch

docs/contributing/ci_architecture.md

@@ -99,6 +99,13 @@ status.
 Full Suite is also path-filtered. It validates broader behavior before Mergify
 can merge a PR.
 
+A `ready`-labeled PR does not hit Buildkite immediately:
+`ci-trigger-full-suite.yml` first runs `.github/scripts/gate_full_suite.sh`,
+which waits for the cheap Tier-1 checks (pre-commit, docs build) on the PR
+head. A red cheap check blocks the suite (fail closed; the next push re-arms
+it), while a GitHub outage or a >25 min wait lets it run anyway (fail open).
+`/test full` bypasses the gate.
+
 | Buildkite label | `TEST_TYPE` | Main watched paths |
 |---|---|---|
 | SSIM Tests | `ssim` | `fastvideo/**/*.py`, `pyproject.toml`, `docker/Dockerfile` |

docs/contributing/performance_benchmarks.md

@@ -79,10 +79,11 @@ fastvideo/performance/
 ```
 
 The HF dataset (`FastVideo/performance-tracking` by default) holds one
-normalized JSON per `(model_id, gpu_type, run)` tuple. The rolling baseline is
-the median of the last 5 successful, baseline-eligible records for that
-model+GPU. PR and local records are visible in the dashboard but are not
-baseline eligible.
+normalized JSON per run. For v2 records, the rolling baseline is the median of
+the last 5 successful, baseline-eligible records in the same comparison cohort:
+`model_id`, `gpu_type`, `workload_id`, `variant_id`, `benchmark_version`,
+`recipe_fingerprint`, `hardware_profile_id`, and `software_profile_id`. PR and
+local records are visible in the dashboard but are not baseline eligible.
 
 ## Planned Coverage
 
@@ -112,10 +113,11 @@ per-metric policy with direction, percent threshold, absolute threshold, and a
 while it runs so pipeline stage execution times are available in
 `generate_video(...).logging_info`. Stage logs use pipeline-unique keys such as
 `prompt_encoding_stage` so duplicate stage classes do not collide. For
-`PipelineStage` entries, the extractor maps the `stage_class` field:
-`TextEncodingStage` maps to `text_encoder_time_s`, `DenoisingStage` and
-`DmdDenoisingStage` map to `dit_time_s`, and `DecodingStage` maps to
-`vae_decode_time_s`, with a fallback for older logs that used the class name as
+`PipelineStage` entries, shared component stage bases emit a stable
+`component_metric`: text encoding stages map to `text_encoder_time_s`,
+denoising stages and subclasses map to `dit_time_s`, and decoding stages map to
+`vae_decode_time_s`. The extractor falls back to known `stage_class` names for
+older logs that do not include `component_metric` or that used the class name as
 the stage key. Generator-side timings such as `PostDecodeFrameProcessStage`,
 `VideoSaveStage`, and `AudioMuxStage` are intentionally ignored. If a pipeline
 does not report one of the mapped stages, that component metric is stored as
@@ -157,13 +159,16 @@ unrealistic memory growth, and optionally large component-specific slowdowns
 even when the rolling baseline is empty. They are hand-set with generous
 headroom and almost never need touching.
 
-### Rolling baseline (per `(model_id, gpu_type)`)
+### Rolling baseline (per comparison cohort)
 
 `compare_baseline.py` loads the last 5 successful, baseline-eligible records
-for the same `(model_id, gpu_type)` from the HF dataset, computes the median
-for each available metric, and evaluates the current run with the metric's
-rolling regression policy. For latency, memory, and component times, higher
-values are regressions. For throughput, lower values are regressions.
+for the same comparison cohort from the HF dataset, computes the median for
+each available metric, and evaluates the current run with the metric's
+rolling regression policy. For v2 records, that cohort is `model_id`,
+`gpu_type`, `workload_id`, `variant_id`, `benchmark_version`,
+`recipe_fingerprint`, `hardware_profile_id`, and `software_profile_id`. For
+latency, memory, and component times, higher values are regressions. For
+throughput, lower values are regressions.
 
 A metric exceeds its rolling threshold when both of these are true:
 
@@ -200,9 +205,9 @@ configs and remain loadable. New or migrated configs should use
 {
   "benchmark_id": "wan-t2v-1.3b-2gpu",
   "config_schema_version": 2,
-  "workload_id": "wan-t2v-1.3b",
-  "variant_id": "canonical",
-  "benchmark_version": 1
+  "workload_id": "wan-t2v",
+  "variant_id": "1.3b-sp2",
+  "benchmark_version": 2
 }
 ```
 
@@ -213,21 +218,27 @@ metadata that make the measured workload explicit:
 
 | Field | Purpose |
 |---|---|
-| `workload_id` | Stable benchmark family, such as `wan-t2v-1.3b`. |
-| `variant_id` | Intentional recipe family, such as `canonical`. |
+| `workload_id` | Stable benchmark family, such as `wan-t2v`. |
+| `variant_id` | Intentional recipe family, including model size and parallelism config, such as `1.3b-sp2`. |
 | `benchmark_version` | Version of the measurement protocol and comparison policy. |
 
 If a config declares `config_schema_version: 2`, loading fails clearly when any
 required v2 identity field is missing. If v2 identity or metadata fields are
 added without `config_schema_version: 2`, loading also fails so partial
 migrations do not silently run as v1 configs. Optional v2 metadata fields
-reserved for follow-up work, such as `recipe`, `metric_threshold_policy`, and
-`quality_metadata`, must be JSON objects when present.
+reserved for follow-up work, such as `metric_threshold_policy` and
+`quality_metadata`, must be JSON objects when present. (`recipe` is emitted
+by the harness and is not config-declarable.)
 
 Recipe fingerprinting, hardware/software profile IDs, exact-identity
-comparison, metric-specific threshold policy behavior, promoted baselines, and
-dashboard regrouping are separate follow-up changes. Until those land, rolling
-baseline comparison remains keyed by `(model_id, gpu_type)`.
+comparison, and dashboard cohort grouping land with this change: v2 records
+compare only within their identity cohort, and a record that opens a NEW
+cohort is marked `baseline_status: "initialized_new_cohort"` (regression
+gating starts once that cohort accumulates history). Legacy v1 configs still
+run and are normalized for reporting, but their records skip rolling-baseline
+comparison entirely (`baseline_status: "skipped_missing_identity"`, never
+baseline eligible); only static thresholds gate them. Metric-specific
+threshold policies and promoted baselines remain separate follow-ups.
 
 ### Raw record (`results/perf_*.json`)
 
@@ -236,10 +247,10 @@ Written by `test_inference_performance.py`. One file per benchmark run.
 ```jsonc
 {
   "benchmark_id": "wan-t2v-1.3b-2gpu",
-  "config_schema_version": 2,
-  "workload_id": "wan-t2v-1.3b",
-  "variant_id": "canonical",
-  "benchmark_version": 1,
+  "result_schema_version": 2,
+  "workload_id": "wan-t2v",
+  "variant_id": "1.3b-sp2",
+  "benchmark_version": 2,
   "model_short_name": "Wan2.1-T2V-1.3B-Diffusers",
   "device": "NVIDIA L40S",
   "num_gpus": 2,
@@ -265,11 +276,54 @@ Written by `test_inference_performance.py`. One file per benchmark run.
     }
   },
   "commit": "<full sha>",
+  "run_source": "pr",
+  "branch": "feature/perf-change",
   "pr_number": "1234",
+  "test_scope": "direct",
+  "build_url": "https://buildkite.example/build",
+  "build_id": "<buildkite-build-id>",
+  "job_id": "<buildkite-job-id>",
   "timestamp": "2026-05-08T22:00:00+00:00",
+  "quality_metadata": { "quality_status": "canonical" },
   "text_encoder_time_s": 2.141,
   "dit_time_s": 8.437,
-  "vae_decode_time_s": 3.208
+  "vae_decode_time_s": 3.208,
+  "recipe": {
+    "recipe_schema_version": 1,
+    "benchmark": {
+      "benchmark_id": "wan-t2v-1.3b-2gpu",
+      "workload_id": "wan-t2v",
+      "variant_id": "1.3b-sp2",
+      "benchmark_version": 2
+    },
+    "model": { "model_path": "Wan-AI/Wan2.1-T2V-1.3B-Diffusers" },
+    "init_kwargs": { "num_gpus": 2, "sp_size": 2, "tp_size": 1 },
+    "generation_kwargs": { "height": 480, "width": 832, "num_frames": 45 },
+    "inputs": { "prompt_count": 1, "prompt_sha256": ["<measured-prompt-sha256>"] },
+    "attention": { "requested_backend": "FLASH_ATTN", "resolved_backend": "FLASH_ATTN" }
+  },
+  "recipe_fingerprint": "<sha256>",
+  "hardware_profile": {
+    "device_type": "cuda",
+    "gpu_count": 2,
+    "gpus": [{ "name": "NVIDIA L40S", "memory_gb": 48, "compute_capability": "8.9" }],
+    "interconnect": "none_or_partial"
+  },
+  "hardware_profile_id": "hw-<sha256-prefix>",
+  "software_profile": {
+    "python": "3.12",
+    "pytorch": "2.12",
+    "cuda": "13.0",
+    "packages": {
+      "fastvideo_kernel": "0.3.2",
+      "flashinfer": "0.2.11",
+      "nvidia_cutlass_dsl": "4.5.0",
+      "triton": "3.4.1"
+    }
+  },
+  "software_profile_id": "sw-<sha256-prefix>",
+  "environment_metadata": { "env": { "IMAGE_VERSION": "py3.12-cuda13.0.0" } },
+  "environment_fingerprint": "env-<sha256-prefix>"
 }
 ```
 
@@ -281,6 +335,10 @@ result, used as the rolling-baseline source of truth.
 ```jsonc
 {
   "model_id": "wan-t2v-1.3b-2gpu",
+  "result_schema_version": 2,
+  "workload_id": "wan-t2v",
+  "variant_id": "1.3b-sp2",
+  "benchmark_version": 2,
   "timestamp": "2026-05-08T22:00:00+00:00",
   "commit_sha": "<full sha>",
   "gpu_type": "NVIDIA L40S",
@@ -297,17 +355,46 @@ result, used as the rolling-baseline source of truth.
       "gated": true
     }
   },
+  "recipe_fingerprint": "<sha256>",
+  "hardware_profile_id": "hw-<sha256-prefix>",
+  "software_profile_id": "sw-<sha256-prefix>",
+  "environment_fingerprint": "env-<sha256-prefix>",
+  "run_source": "pr",
+  "branch": "feature/perf-change",
+  "pr_number": "1234",
+  "test_scope": "direct",
+  "build_url": "https://buildkite.example/build",
+  "build_id": "<buildkite-build-id>",
+  "job_id": "<buildkite-job-id>",
+  "quality_metadata": { "quality_status": "canonical" },
   "success": true
 }
 ```
 
 ### Compatibility with legacy records
 
-Older records in the HF dataset may not have component timing fields. The
-comparator ignores missing or `null` metrics when computing a median, and the
-dashboard lists skipped plots for metric series that have no non-null values.
-Records missing both `run_source` and `baseline_eligible` are treated as legacy
-successful main/full-suite uploads and remain eligible for rolling baselines.
+Older records in the HF dataset may not have `result_schema_version`,
+component timing fields, or v2 identity/profile fields. Records without
+`result_schema_version` are treated as v1. The comparator ignores missing or
+`null` metrics when computing a median, and the dashboard lists skipped plots
+for metric series that have no non-null values. Records missing both
+`run_source` and `baseline_eligible` are treated as legacy successful
+main/full-suite uploads and remain eligible for rolling baselines.
+Current `perf_*.json` artifacts that lack the v2 comparison identity are
+normalized for reporting but skip rolling-baseline comparison and are not marked
+baseline eligible.
+
+New records compare only against the same `model_id`, `gpu_type`,
+`workload_id`, `variant_id`, `benchmark_version`, `recipe_fingerprint`,
+`hardware_profile_id`, and `software_profile_id` cohort.
+`environment_metadata` and `environment_fingerprint` are audit data and are not
+part of the comparison key.
+The recipe prompt digests describe the prompts actually measured by the
+benchmark run; extra configured prompts are ignored unless the benchmark runner
+executes them.
+Software profile package cohorts keep exact versions for relevant
+attention/kernel packages, including FastVideo kernels, FlashAttention,
+FlashInfer, Cutlass DSL, SageAttention, Triton, and xFormers when installed.
 
 ## Environment variable reference
 
@@ -317,7 +404,7 @@ successful main/full-suite uploads and remain eligible for rolling baselines.
 | `PERF_REPORTS_DIR` | `/root/data/perf_reports` | `compare_baseline.py`, `dashboard.py` | Where the Markdown summary and Plotly HTML get written for Buildkite to pick up. |
 | `HF_REPO_ID` | `FastVideo/performance-tracking` | `fastvideo/performance/hf_store.py` | HF dataset repo holding rolling-baseline records. |
 | `HF_API_KEY`, `HUGGINGFACE_HUB_TOKEN`, `HF_TOKEN` | unset | `fastvideo/performance/hf_store.py` | Required for upload or private dataset reads. |
-| `PERF_RUN_SOURCE` | inferred | `compare_baseline.py` | Source metadata for uploaded records: `pr`, `local`, `scheduled_main`, or `unknown`. |
+| `PERF_RUN_SOURCE` | inferred | `compare_baseline.py`, `test_inference_performance.py` | Source metadata for uploaded records: `pr`, `local`, `scheduled_main`, or `unknown`. |
 | `PERF_UPLOAD_POLICY` | `never` | `compare_baseline.py` | Upload policy: `never`, `pass`, or `always`. |
 | `PERF_PYTEST_RC` | unset | `compare_baseline.py` | Static-threshold pytest exit code, used so scheduled-main failures can be uploaded with `success=false`. |
 | `TEST_SCOPE` | unset | `compare_baseline.py` | CI context used to infer scheduled-main runs together with `BUILDKITE_BRANCH=main`. |
@@ -334,11 +421,14 @@ point is `fastvideo/tests/modal/pr_test.py:run_performance_tests` and the
 Buildkite artifact upload is in
 `.buildkite/scripts/pr_test.sh:upload_performance_artifacts`.
 
-Each performance build runs pytest first. If that fixed-threshold phase fails,
-PR/direct runs skip `compare_baseline.py` because they only upload passing
-records. Scheduled-main runs still execute `compare_baseline.py` with
-`PERF_PYTEST_RC` set so the failed canonical attempt is visible in normalized
-JSON and dashboard history. The dashboard runs best-effort for observability.
+Each performance build runs pytest first. PR and direct runs only continue to
+`compare_baseline.py` when that fixed-threshold phase passes; if pytest fails,
+Markdown summaries and normalized JSON artifacts are not emitted. Scheduled
+main runs set `PERF_UPLOAD_POLICY=always`, so they still run
+`compare_baseline.py` (with `PERF_PYTEST_RC` set) after a fixed-threshold
+failure. Those failed scheduled main runs emit summaries and normalized
+records, upload records with `success=false`, and are excluded from future
+rolling baselines. The dashboard still runs best-effort for observability.
 When the rolling-baseline phase runs, it emits:
 
 * **Markdown summary** — appended to `$GITHUB_STEP_SUMMARY` when that variable
@@ -346,7 +436,7 @@ When the rolling-baseline phase runs, it emits:
   per-benchmark row with current vs. baseline values for latency, throughput,
   memory, text encoder time, DiT time, and VAE decode time.
 * **Plotly dashboard** — `dashboard_<sha>_<ts>.html` showing time-series for
-  each metric grouped by `(model_id, gpu_type)`.
+  each metric grouped by comparison cohort.
 * **Normalized records** — `normalized_perf_*.json`, one per benchmark.
   Useful as input to the
   [`reseed-performance-baseline`](https://github.com/hao-ai-lab/FastVideo/blob/main/.agents/skills/reseed-performance-baseline/SKILL.md)
@@ -363,7 +453,7 @@ When the rolling-baseline phase runs, it emits:
      "benchmark_id": "<unique-id>",
      "config_schema_version": 2,
      "workload_id": "<stable-workload-id>",
-     "variant_id": "canonical",
+     "variant_id": "<variant, e.g. 1.3b-sp2>",
      "benchmark_version": 1,
      "model": { "model_path": "...", "model_short_name": "..." },
      "init_kwargs": { "num_gpus": 1, ... },
@@ -390,7 +480,9 @@ When the rolling-baseline phase runs, it emits:
 
    Legacy v1 configs without `config_schema_version` still load, but should not
    gain v2 identity or metadata fields until they are migrated to
-   `config_schema_version: 2`.
+   `config_schema_version: 2`. For v2 configs, `workload_id`, `variant_id`,
+   and `benchmark_version` are part of the comparison key; benchmark runs
+   fail if any of these identity fields are missing.
 
 2. The pytest test auto-discovers all configs — no test code needed. CI
    picks it up on the next `/test performance` run.
@@ -418,8 +510,8 @@ When the rolling-baseline phase runs, it emits:
 
 ## Troubleshooting
 
-**"No baseline for ... Initializing"** — first run for this `(model_id,
-gpu_type)`. Run will pass and (if persisting) seed the first record.
+**"No baseline for ... Initializing"** — first run for this comparison cohort.
+Run will pass and (if persisting) seed the first record.
 
 **Persistent failure right after a torch / kernel / image upgrade** —
 genuine regression *or* baseline drift. Compare the failing normalized record
@@ -432,5 +524,5 @@ pipelines that did not report a mapped component stage.
 
 **Component timing is `null`** — the generated result did not include a mapped
 stage in `logging_info.stages`. Check that the pipeline emits stage logging
-and that the stage name is listed in `STAGE_METRIC_MAP` in
-`test_inference_performance.py`.
+and that the stage emits `component_metric` or is covered by the legacy
+`STAGE_METRIC_MAP` fallback in `test_inference_performance.py`.

docs/contributing/testing.md

@@ -180,6 +180,30 @@ python fastvideo/tests/ssim/reference_videos_cli.py copy-local \
   --device-folder L40S_reference_videos
 ```
 
+### SSIM Bootstrap Mode
+
+Normal SSIM runs are strict: if a reference video or latent is missing, the
+test fails. For new-model PRs, CI can run SSIM in bootstrap mode so missing
+references are uploaded as draft artifacts for review instead of immediately
+blocking on a missing canonical reference.
+
+Buildkite enables SSIM bootstrap mode when either condition is true:
+
+- the PR title or Buildkite message contains `[new-model]`;
+- `FASTVIDEO_SSIM_BOOTSTRAP_MODE=1` is set for the Buildkite job.
+
+Bootstrap mode passes `--ssim-bootstrap-mode` to pytest. When a generated
+artifact is available, the test uploads it under the `drafts/...` namespace in
+the SSIM reference repo and marks that case as expected-failed. After reviewing
+the draft, promote it into the canonical reference layout:
+
+```bash
+python fastvideo/tests/ssim/reference_videos_cli.py promote-draft \
+  --quality-tier default \
+  --device-folder L40S_reference_videos \
+  --model-id <model_id>
+```
+
 ## CI Integration
 
 FastVideo CI tests are orchestrated by Buildkite and run on Modal GPU

docs/design/inference_schema_parity_inventory.yaml

@@ -41,11 +41,6 @@ surfaces:
       disable_autocast: generator.engine.disable_autocast
       enable_stage_verification: generator.engine.enable_stage_verification
       prompt_txt: request.inputs.prompt_path
-      batching_mode: generator.engine.batching.mode
-      batching_max_size: generator.engine.batching.max_size
-      batching_delay_ms: generator.engine.batching.delay_ms
-      batching_config: generator.engine.batching.config_path
-      enable_batching_metrics: generator.engine.batching.enable_metrics
       override_text_encoder_safetensors: generator.pipeline.components.text_encoder_weights
       override_text_encoder_quant: generator.engine.quantization.text_encoder_quant
       transformer_quant: generator.engine.quantization.transformer_quant
@@ -196,6 +191,9 @@ surfaces:
         sources:
           - fastvideo.configs.pipelines.gen3c.Gen3CConfig
           - fastvideo.configs.pipelines.gen3c.Gen3CInferenceConfig
+      color_correction_strength:
+        sources:
+          - fastvideo.configs.pipelines.dreamx_world.DreamXWorld5BARPipelineConfig
       default_camera_rotation:
         sources:
           - fastvideo.configs.pipelines.gen3c.Gen3CConfig
@@ -460,6 +458,8 @@ surfaces:
       guidance_scale: request.sampling.guidance_scale
       guidance_scale_2: request.sampling.guidance_scale_2
       guidance_rescale: request.sampling.guidance_rescale
+      use_embedded_guidance: request.sampling.use_embedded_guidance
+      true_cfg_scale: request.sampling.true_cfg_scale
       boundary_ratio: request.sampling.boundary_ratio
       sigmas: request.sampling.sigmas
       enable_teacache: request.runtime.enable_teacache

docs/distillation/anyflow.md

@@ -0,0 +1,116 @@
+# 🌊 AnyFlow Any-Step Video Distillation
+
+**AnyFlow** ([paper](https://arxiv.org/abs/2605.13724), [project page](https://nvlabs.github.io/AnyFlow/), [official code](https://github.com/NVlabs/AnyFlow), [model weights](https://huggingface.co/collections/nvidia/anyflow)) is an any-step video diffusion framework built on flow maps. A single distilled checkpoint can be evaluated at NFE ∈ {1, 2, 4, 8, 16, 32} without retraining, and quality scales **monotonically** with steps — unlike consistency-based distillation, which often degrades as NFE grows.
+
+The student network ``u_θ(x_t, t, r)`` predicts the *average velocity* from time ``t`` back to time ``r``, so one Euler step is
+
+```
+x_r = x_t - ((t - r) / N) · u_θ(x_t, t, r)
+```
+
+for any ``t > r``.
+
+## 📊 Model Overview
+
+NVIDIA publishes four checkpoints under [`nvidia/anyflow`](https://huggingface.co/collections/nvidia/anyflow):
+
+- `nvidia/AnyFlow-Wan2.1-T2V-1.3B-Diffusers` — bidirectional T2V, Wan2.1 1.3B base
+- `nvidia/AnyFlow-Wan2.1-T2V-14B-Diffusers` — bidirectional T2V, Wan2.1 14B base
+- `nvidia/AnyFlow-FAR-Wan2.1-1.3B-Diffusers` — frame-autoregressive variant, 1.3B
+- `nvidia/AnyFlow-FAR-Wan2.1-14B-Diffusers` — frame-autoregressive variant, 14B
+
+FastVideo currently supports the bidirectional T2V variants for training; the FAR variants can be loaded for inference through the diffusers integration.
+
+## ⚙️ Inference
+
+For inference, load the published checkpoint directly through diffusers; FastVideo's training-side ``WanModel`` config maps the HF AnyFlow ``delta_embedder`` weights onto its internal layout via ``param_names_mapping`` so the same checkpoint can be used as the ``init_from`` for the on-policy YAML below.
+
+## 🧠 Algorithm
+
+Training runs in two stages. Both use the dual-timestep Wan backbone — enabled by ``pipeline.dit_config.r_embedder: true`` in the YAML, which allocates a sibling ``condition_embedder.delta_embedder`` and fuses its embedding with the standard timestep embedding via either an additive or a gated mixer.
+
+### Stage 1 — Pretrain (flow-map central-difference)
+
+Method: ``AnyFlowPretrainMethod`` (``fastvideo/train/methods/distribution_matching/anyflow_pretrain.py``)
+
+For each batch, sample ``(t, r) ∈ [0, 1]`` as ``(max, min)`` of two uniform draws, then:
+
+- a ``diffusion_ratio`` fraction (default 0.5) gets ``r = t`` — recovers plain flow matching;
+- a ``consistency_ratio`` fraction (default 0.25) gets ``r = 0`` — forces consistency to clean data;
+- the remainder is free.
+
+The student forward at ``(t, r)`` is trained against the central-difference target
+
+```
+target = (eps - x_0) - (t - r) · dF/dt
+```
+
+where ``dF/dt`` is estimated from the student's own forward at ``(t ± δ, r)`` with the sample also moved along the flow trajectory by ``v_pred · (δ / N)``. Per-timestep weighting uses ``beta08`` (``w(t) = t · sqrt(1 - t)``, renormalized). A stop-gradient scale-balance keeps the non-diffusion branches' loss magnitude aligned with the diffusion branch.
+
+### Stage 2 — On-policy DMD
+
+Method: ``AnyFlowMethod`` (``fastvideo/train/methods/distribution_matching/anyflow.py``)
+
+Inherits ``DMD2Method``. The student is rolled out for ``student_sample_steps`` Euler-flow steps from pure noise; one randomly-chosen step is gradient-enabled (broadcast from rank 0 so every worker agrees), the rest run under ``torch.no_grad``. With ``use_mean_velocity: true`` (default) the rollout uses ``r = t_next`` at each step, matching AnyFlow's ``WanAnyFlowPipeline.training_rollout``.
+
+The inherited ``_dmd_loss`` (VSD with fake-score critic) consumes the rollout output and the teacher's CFG prediction. The optional pinned ``t_list_override`` lets configs reproduce the paper's hand-tuned 4-step schedule ``[999, 937, 833, 624, 0]``.
+
+## 🚀 Training Scripts
+
+### Stage 1 — pretrain
+
+```bash
+bash examples/train/run.sh \
+    examples/train/configs/distribution_matching/wan/anyflow_pretrain_t2v.yaml
+```
+
+**Key configuration** (in ``examples/train/configs/distribution_matching/wan/anyflow_pretrain_t2v.yaml``):
+
+- Global batch size: 32 (8 GPUs × 4 per-GPU)
+- Learning rate: 5e-5
+- Flow shift: 5.0
+- ``diffusion_ratio`` / ``consistency_ratio``: 0.5 / 0.25
+- ``epsilon`` (finite-difference step): 5 (absolute train-timestep units)
+- ``weight_type``: ``beta08``
+- ``fuse_guidance_scale``: 3.0
+- Training steps: 6000
+
+### Stage 2 — on-policy
+
+```bash
+bash examples/train/run.sh \
+    examples/train/configs/distribution_matching/wan/anyflow_onpolicy_t2v.yaml \
+    --models.student.init_from outputs/wan2.1_anyflow_pretrain/checkpoint-final
+```
+
+(Or point ``models.student.init_from`` directly at ``nvidia/AnyFlow-Wan2.1-T2V-1.3B-Diffusers`` to bootstrap from the paper weights and skip Stage 1.)
+
+**Key configuration**:
+
+- Global batch size: 8 (8 GPUs × 1 per-GPU)
+- Learning rate: 2e-6
+- Flow shift: 5.0
+- ``student_sample_steps``: 4
+- ``t_list_override``: ``[999, 937, 833, 624, 0]``
+- ``use_mean_velocity``: ``true`` (i.e. ``r = t_next`` during rollout)
+- ``real_score_guidance_scale``: 3.0
+- ``generator_update_interval``: 5 (DMD2 alternation)
+- Training steps: 4000
+
+## 🔌 Loading published AnyFlow checkpoints
+
+The HF AnyFlow checkpoints expose ``condition_embedder.delta_embedder.*`` weights that FastVideo internally maps onto its ``condition_embedder.delta_embedder.mlp.*`` layout. This rename happens automatically through the regex in ``WanVideoArchConfig.param_names_mapping`` — no separate adapter is needed. The same regex is a no-op on plain Wan checkpoints (which don't contain any ``delta_embedder`` keys).
+
+Set the YAML's ``pipeline.dit_config.r_embedder: true`` to allocate the ``delta_embedder`` module on the FastVideo side; when initializing from a plain Wan checkpoint the delta weights are deep-copied from ``time_embedder`` (matching AnyFlow's ``setup_flowmap_model()`` behavior).
+
+## 🧭 Note on ``fuse_guidance_scale``
+
+Stage 1 optionally fuses classifier-free guidance into the training target so the resulting checkpoint can be sampled at ``guidance_scale=1.0`` (no extra forward pass at inference time). The transformation is
+
+```
+noise_pred ← (noise_pred - (1 - g) · noise_pred_uncond) / g
+```
+
+with ``g = fuse_guidance_scale``. The negative prompt embedding comes from ``WanModel``'s ``ensure_negative_conditioning()`` — i.e. the dataset's configured ``sampling_param.negative_prompt``. Setting ``fuse_guidance_scale: 1.0`` skips the extra unconditional forward entirely.
+
+The on-policy stage's ``real_score_guidance_scale`` (inherited from DMD2) follows the same parameterization conventions documented in [``dmd.md``](dmd.md#-note-on-real_score_guidance_scale).

docs/inference/support_matrix.md

@@ -58,6 +58,8 @@ pipeline initialization and sampling.
 | FastWan2.1 T2V 1.3B | `FastVideo/FastWan2.1-T2V-1.3B-Diffusers` | 480P | ⭕ | ⭕ | ⭕ | ✅ | ⭕ |
 | FastWan2.2 TI2V 5B Full Attn* | `FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers` | 720P | ⭕ | ⭕ | ⭕ | ✅ | ⭕ |
 | Wan2.2 TI2V 5B | `Wan-AI/Wan2.2-TI2V-5B-Diffusers` | 720P | ⭕ | ⭕ | ✅ | ⭕ | ⭕ |
+| DreamX-World 5B Cam | `FastVideo/DreamX-World-5B-Cam-Diffusers` | 480P | ⭕ | ⭕ | ⭕ | ⭕ | ⭕ |
+| DreamX-World 5B AR | `FastVideo/DreamX-World-5B-Diffusers` | 704px1280p | ⭕ | ⭕ | ⭕ | ⭕ | ⭕ |
 | Lucy Edit Dev 5B*** | `decart-ai/Lucy-Edit-Dev` | 480P | ⭕ | ⭕ | ⭕ | ⭕ | ⭕ |
 | Wan2.2 T2V A14B | `Wan-AI/Wan2.2-T2V-A14B-Diffusers` | 480P<br>720P | ❌ | ❌ | ✅ | ⭕ | ⭕ |
 | Wan2.2 I2V A14B | `Wan-AI/Wan2.2-I2V-A14B-Diffusers` | 480P<br>720P | ❌ | ❌ | ✅ | ⭕ | ⭕ |

docs/training/trackers.md

@@ -0,0 +1,91 @@
+# Training Trackers
+
+FastVideo can send training metrics and validation media to Weights & Biases
+or SwanLab. Tracking runs only on global rank 0, and local tracker files are
+stored under `<output_dir>/tracker`.
+
+## Supported Trackers
+
+| Value | Backend | Installation |
+|-------|---------|--------------|
+| `wandb` | Weights & Biases | Included with FastVideo |
+| `swanlab` | SwanLab | Install the optional `swanlab` dependency |
+| `none` | Disable external tracking | No additional package |
+
+You can enable more than one backend, for example `trackers: [wandb, swanlab]`.
+Metrics and validation media are converted to the artifact type required by
+each backend.
+
+## Install SwanLab
+
+For a published FastVideo installation, install the SwanLab extra:
+
+```bash
+uv pip install "fastvideo[swanlab]"
+```
+
+For an editable source checkout, include the same extra during installation:
+
+```bash
+uv pip install -e ".[swanlab]"
+```
+
+If FastVideo is already installed, you can install the compatible SDK directly:
+
+```bash
+uv pip install "swanlab>=0.6.7"
+```
+
+Authenticate once before starting a training run:
+
+```bash
+swanlab login
+```
+
+See the [SwanLab login documentation](https://docs.swanlab.cn/en/api/cli-swanlab-login.html)
+for non-interactive and self-hosted setups.
+
+## Configure Tracking
+
+Select SwanLab in the YAML config used by the modular training framework:
+
+```yaml
+training:
+  checkpoint:
+    output_dir: outputs/my_run
+  tracker:
+    trackers: [swanlab]
+    project_name: my_project
+    run_name: my_run
+```
+
+To log to both supported services:
+
+```yaml
+training:
+  tracker:
+    trackers: [wandb, swanlab]
+    project_name: my_project
+    run_name: my_run
+```
+
+An empty or omitted `trackers` list selects W&B when `project_name` is set.
+Use an explicit `none` entry to disable external tracking:
+
+```yaml
+training:
+  tracker:
+    trackers: [none]
+```
+
+## Validation Videos
+
+SwanLab currently accepts GIF video artifacts. FastVideo converts validation
+MP4 files and in-memory video arrays to GIF automatically before logging them.
+For video files, FastVideo uses the sampling frame rate supplied by the caller,
+or the source file's frame rate when no value is supplied. In-memory arrays use
+the frame rate supplied by the caller. Both forms fall back to 16 FPS when no
+frame rate is available.
+
+For details about configuring validation callbacks, see
+[Training Infrastructure](train_infra.md#callbacks-pluggable-hooks).

docs/training/train_infra.md

@@ -161,6 +161,21 @@ training:
     decay_interval_steps: 0
 ```
 
+`training.data.data_path` can also mix multiple preprocessed datasets by using a mapping from dataset path to repeat count:
+
+```yaml
+training:
+  data:
+    data_path:
+      data/zeldam2-clean: 1
+      data/multi3d_games: 2
+```
+
+The repeat count duplicates that dataset's parquet file list before shuffling/sampling, so the example above trains with roughly twice as much `multi3d_games` exposure as `zeldam2-clean`. Paths are just suggested locations; use any local path that contains a FastVideo preprocessed parquet dataset.
+
+See [Training Trackers](trackers.md) to configure Weights & Biases or SwanLab,
+including SwanLab installation and authentication.
+
 ### `callbacks` — Pluggable hooks
 
 Callbacks run at specific points in the training loop (before/after optimizer
@@ -323,6 +338,40 @@ Self-Forcing inherits all DMD2 parameters, plus:
 | `enable_gradient_in_rollout` | `true` | Enable backprop through rollout |
 | `start_gradient_frame` | `0` | Frame index where gradients begin |
 
+### Streaming Long Tuning
+
+`StreamingLongTuningMethod` extends Self-Forcing for LongLive-style rollouts. It
+keeps a streaming state, generates overlapping chunks, and trains only the new
+frames while preserving context from earlier chunks.
+
+For the MatrixGame2/Zelda world-model example, self-forcing and long tuning are
+separate runs: first train or load the 1k-step self-forcing checkpoint using
+`examples/train/scenario/worldmodel/zelda/self_forcing_causal_i2v.yaml`,
+then run
+`examples/train/scenario/worldmodel/zelda/streaming_long_tuning_causal_i2v.yaml`
+from that checkpoint for the 3k-step streaming long-tuning stage.
+
+```yaml
+method:
+  _target_: fastvideo.train.methods.distribution_matching.streaming_long_tuning.StreamingLongTuningMethod
+  streaming_chunk_size: 9
+  streaming_max_length: 39
+  streaming_fixed_overlap_latents: 3
+  streaming_reencode_overlap_anchor: true
+  streaming_anchor_inject_k: 1
+  streaming_require_full_blocks: true
+  multi_phased_distill_schedule:
+    - stage: streaming_long
+      start_step: 0
+      end_step: 3000
+      num_latent_t: 39
+      streaming_training: true
+```
+
+See
+`examples/train/scenario/worldmodel/zelda/streaming_long_tuning_causal_i2v.yaml`
+for a complete MatrixGame2/Zelda configuration.
+
 ---
 
 ## Callbacks

examples/inference/basic/basic_dreamx_world.py

@@ -0,0 +1,64 @@
+import os
+
+from fastvideo import VideoGenerator
+
+
+OUTPUT_PATH = os.getenv("DREAMX_WORLD_OUTPUT_PATH", "video_samples_dreamx_world")
+
+
+def _env_int(name: str, default: int) -> int:
+    return int(os.getenv(name, str(default)))
+
+
+def _env_float(name: str, default: float) -> float:
+    return float(os.getenv(name, str(default)))
+
+
+def main():
+    model_name = os.getenv("DREAMX_WORLD_MODEL_DIR", "FastVideo/DreamX-World-5B-Cam-Diffusers")
+    generator = VideoGenerator.from_pretrained(
+        model_name,
+        num_gpus=1,
+        use_fsdp_inference=False,
+        dit_cpu_offload=False,
+        vae_cpu_offload=True,
+        text_encoder_cpu_offload=True,
+        pin_cpu_memory=False,
+        override_pipeline_cls_name="DreamXWorldPipeline",
+    )
+
+    prompt = os.getenv(
+        "DREAMX_WORLD_PROMPT",
+        "A cinematic first-person drive through a futuristic coastal city at "
+        "sunrise, reflective glass towers, clean streets, soft volumetric light.",
+    )
+    image_path = os.getenv(
+        "DREAMX_WORLD_IMAGE_PATH",
+        "https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/wan_i2v_input.JPG",
+    )
+
+    kwargs = {
+        "output_path": OUTPUT_PATH,
+        "save_video": os.getenv("DREAMX_WORLD_SAVE_VIDEO", "1") != "0",
+        "height": _env_int("DREAMX_WORLD_HEIGHT", 480),
+        "width": _env_int("DREAMX_WORLD_WIDTH", 832),
+        "num_frames": _env_int("DREAMX_WORLD_NUM_FRAMES", 161),
+        "num_inference_steps": _env_int("DREAMX_WORLD_STEPS", 30),
+        "guidance_scale": _env_float("DREAMX_WORLD_GUIDANCE", 5.0),
+        "action_list": os.getenv("DREAMX_WORLD_ACTIONS", "w,d,w").split(","),
+        "action_speed_list": [
+            float(value)
+            for value in os.getenv("DREAMX_WORLD_ACTION_SPEEDS", "4.0,2.0,4.0").split(",")
+        ],
+    }
+    if image_path:
+        kwargs["image_path"] = image_path
+
+    try:
+        generator.generate_video(prompt, **kwargs)
+    finally:
+        generator.shutdown()
+
+
+if __name__ == "__main__":
+    main()

examples/inference/basic/basic_flux_dev.py

@@ -0,0 +1,140 @@
+# SPDX-License-Identifier: Apache-2.0
+
+from __future__ import annotations
+
+import argparse
+import contextlib
+import os
+import re
+
+DEFAULT_PROMPTS = [
+    "a photo of a cat",
+    (
+        "a cinematic photo of a red panda wearing a tiny backpack, standing on a "
+        "rainy neon-lit street at night, shallow depth of field, sharp focus, "
+        "35mm, bokeh"
+    ),
+]
+
+
+def _safe_filename(text: str, max_len: int = 100) -> str:
+    """Make a stable, filesystem-friendly filename base."""
+    s = text[:max_len].strip()
+    s = s.replace(os.sep, "_")
+    if os.altsep:
+        s = s.replace(os.altsep, "_")
+    s = re.sub(r"\s+", " ", s)
+    s = re.sub(r"[^A-Za-z0-9 .,_-]", "_", s)
+    s = s.strip(" .")
+    return s or "prompt"
+
+
+def _remove_existing_outputs(out_dir: str, filename_base: str) -> None:
+    """Delete prior outputs so reruns do not get _1, _2 suffixes."""
+    if not os.path.isdir(out_dir):
+        return
+
+    pattern = re.compile(rf"^{re.escape(filename_base)}(_\d+)?\.(mp4|png)$")
+    for fn in os.listdir(out_dir):
+        if pattern.match(fn):
+            with contextlib.suppress(FileNotFoundError):
+                os.remove(os.path.join(out_dir, fn))
+
+
+def parse_args() -> argparse.Namespace:
+    p = argparse.ArgumentParser(
+        description="Run FLUX.1-dev text-to-image with FastVideo VideoGenerator.",
+    )
+    p.add_argument(
+        "--model-path",
+        default="official_weights/FLUX.1-dev",
+        help="Local Diffusers checkpoint dir or HF repo id.",
+    )
+    p.add_argument(
+        "--out-dir",
+        "--outdir",
+        default="outputs/flux_dev/samples",
+        help="Directory for saved PNG outputs.",
+    )
+    p.add_argument(
+        "--prompt",
+        action="append",
+        default=None,
+        help="Prompt. Repeat for multiple images.",
+    )
+    p.add_argument(
+        "--backend",
+        default=None,
+        help="Set FASTVIDEO_ATTENTION_BACKEND (e.g. TORCH_SDPA).",
+    )
+    p.add_argument("--seed", type=int, default=42, help="Base seed; each prompt uses seed + index.")
+    p.add_argument("--height", type=int, default=1024, help="Output height.")
+    p.add_argument("--width", type=int, default=1024, help="Output width.")
+    p.add_argument("--steps", type=int, default=28, help="Number of inference steps.")
+    p.add_argument("--guidance", type=float, default=3.5, help="Guidance scale.")
+    p.add_argument("--num-gpus", type=int, default=1, help="GPU count.")
+    return p.parse_args()
+
+
+def main() -> None:
+    args = parse_args()
+    prompts: list[str] = args.prompt if args.prompt else DEFAULT_PROMPTS
+
+    if args.backend:
+        os.environ["FASTVIDEO_ATTENTION_BACKEND"] = args.backend
+
+    from fastvideo import VideoGenerator
+
+    os.makedirs(args.out_dir, exist_ok=True)
+
+    init_kwargs = {
+        "num_gpus": args.num_gpus,
+        "workload_type": "t2i",
+        "sp_size": 1,
+        "tp_size": 1,
+        "dit_cpu_offload": False,
+        "dit_layerwise_offload": False,
+        "text_encoder_cpu_offload": False,
+        "vae_cpu_offload": False,
+        "image_encoder_cpu_offload": False,
+        "pin_cpu_memory": False,
+        "use_fsdp_inference": False,
+    }
+
+    generator = VideoGenerator.from_pretrained(
+        model_path=args.model_path,
+        **init_kwargs,
+    )
+    try:
+        for i, prompt in enumerate(prompts):
+            seed = args.seed + i
+            filename_base = (
+                f"flux_dev_{i:02d}_seed{seed}_{_safe_filename(prompt, max_len=80)}"
+            )
+            _remove_existing_outputs(args.out_dir, filename_base)
+            output_path = os.path.join(args.out_dir, f"{filename_base}.png")
+            print(f"[flux] prompt_idx={i} seed={seed} output_path={output_path}")
+
+            generation_kwargs = {
+                "output_path": output_path,
+                "height": args.height,
+                "width": args.width,
+                "num_frames": 1,
+                "fps": 1,
+                "num_inference_steps": args.steps,
+                "guidance_scale": args.guidance,
+                "use_embedded_guidance": True,
+                "true_cfg_scale": 1.0,
+                "seed": seed,
+                "save_video": True,
+            }
+
+            generator.generate_video(prompt, **generation_kwargs)
+
+        print(f"[flux] done. outputs written to: {args.out_dir}")
+    finally:
+        generator.shutdown()
+
+
+if __name__ == "__main__":
+    main()

examples/inference/basic/basic_glm_image.py

@@ -0,0 +1,107 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Run GLM-Image text-to-image generation through FastVideo.
+
+User story:
+    "I have the HF `zai-org/GLM-Image` checkpoint and want a minimal
+    text-to-image generation command, saved as a PNG."
+"""
+import argparse
+from pathlib import Path
+
+from PIL import Image
+
+from fastvideo import VideoGenerator
+from fastvideo.api import (
+    EngineConfig,
+    GenerationRequest,
+    GeneratorConfig,
+    OutputConfig,
+    ParallelismConfig,
+    PipelineSelection,
+    SamplingConfig,
+)
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="Run GLM-Image text-to-image generation.")
+    parser.add_argument(
+        "--model-path",
+        default="zai-org/GLM-Image",
+        help="HF id or local diffusers-format GLM-Image weights directory.",
+    )
+    parser.add_argument(
+        "--output",
+        default="image_output/landscape.png",
+        help="Output PNG path.",
+    )
+    parser.add_argument(
+        "--prompt",
+        default=("A beautiful landscape photography with rolling hills, "
+                 "a winding river, and a vibrant sunset in the background. "
+                 "Warm golden light, photorealistic style."),
+        help="Text prompt.",
+    )
+    parser.add_argument("--height", type=int, default=1024)
+    parser.add_argument("--width", type=int, default=1024)
+    parser.add_argument("--steps", type=int, default=50)
+    parser.add_argument("--guidance-scale", type=float, default=1.5)
+    parser.add_argument("--seed", type=int, default=1024)
+    parser.add_argument("--num-gpus", type=int, default=1)
+    parser.add_argument("--tp-size", type=int, default=None)
+    parser.add_argument("--sp-size", type=int, default=None)
+    return parser.parse_args()
+
+
+def main() -> None:
+    args = parse_args()
+
+    output = Path(args.output)
+    output.parent.mkdir(parents=True, exist_ok=True)
+    tp_size = args.tp_size if args.tp_size is not None else (args.num_gpus if args.num_gpus > 1 else 1)
+    sp_size = args.sp_size if args.sp_size is not None else (1 if args.num_gpus > 1 else args.num_gpus)
+
+    # GLM-Image needs trust_remote_code for its AR encoder; offload and the
+    # pipeline class come from the model's registered defaults — don't override.
+    generator_config = GeneratorConfig(
+        model_path=args.model_path,
+        trust_remote_code=True,
+        engine=EngineConfig(
+            num_gpus=args.num_gpus,
+            parallelism=ParallelismConfig(tp_size=tp_size, sp_size=sp_size),
+        ),
+        pipeline=PipelineSelection(workload_type="t2i"),
+    )
+
+    generator = VideoGenerator.from_config(generator_config)
+    try:
+        request = GenerationRequest(
+            prompt=args.prompt,
+            sampling=SamplingConfig(
+                height=args.height,
+                width=args.width,
+                num_frames=1,
+                fps=1,
+                num_inference_steps=args.steps,
+                guidance_scale=args.guidance_scale,
+                seed=args.seed,
+            ),
+            output=OutputConfig(
+                output_path=str(output.parent),
+                save_video=False,
+                return_frames=True,
+            ),
+        )
+        result = generator.generate(request)
+        if isinstance(result, list):
+            result = result[0]
+
+        frames = result.frames
+        if frames is not None and len(frames):
+            Image.fromarray(frames[0]).save(output)
+            print(f"Saved image to {output}")
+    finally:
+        generator.shutdown()
+
+
+if __name__ == "__main__":
+    main()

examples/inference/basic/basic_kandinsky5_i2v.py

@@ -0,0 +1,37 @@
+from fastvideo import VideoGenerator
+
+OUTPUT_PATH = "video_samples_kandinsky5_i2v"
+
+IMAGE_PATH = "assets/girl.png"
+
+
+def main():
+    generator = VideoGenerator.from_pretrained(
+        "kandinskylab/Kandinsky-5.0-I2V-Pro-distilled-5s-Diffusers", 
+        # "kandinskylab/Kandinsky-5.0-I2V-Pro-sft-5s-Diffusers"
+        # "kandinskylab/Kandinsky-5.0-I2V-Lite-5s-Diffusers"
+        num_gpus=1,
+        use_fsdp_inference=False,
+        dit_cpu_offload=False,
+        vae_cpu_offload=False,
+        text_encoder_cpu_offload=True,
+        pin_cpu_memory=True,
+        # image_encoder_cpu_offload=False,
+    )
+
+    prompt = (
+        "A woman stands up and walks away"
+    )
+    _ = generator.generate_video(
+        prompt,
+        image_path=IMAGE_PATH,
+        output_path=OUTPUT_PATH,
+        save_video=True,
+        height=1024,
+        width=1024,
+        num_frames=121,
+    )
+
+
+if __name__ == "__main__":
+    main()

examples/inference/basic/basic_kandinsky5_t2v.py

@@ -0,0 +1,37 @@
+from fastvideo import VideoGenerator
+
+OUTPUT_PATH = "video_samples_kandinsky5_t2v"
+
+def main():
+    generator = VideoGenerator.from_pretrained(
+        "kandinskylab/Kandinsky-5.0-T2V-Lite-sft-5s-Diffusers",
+        # "kandinskylab/Kandinsky-5.0-T2V-Pro-sft-5s-Diffusers"
+        # "kandinskylab/Kandinsky-5.0-T2V-Lite-distilled16steps-5s-Diffusers"
+        # "kandinskylab/Kandinsky-5.0-T2V-Pro-distilled-5s-Diffusers"
+        num_gpus=1,
+        use_fsdp_inference=False,
+        dit_cpu_offload=False,
+        vae_cpu_offload=False,
+        text_encoder_cpu_offload=True,
+        pin_cpu_memory=True,
+        # image_encoder_cpu_offload=False,
+    )
+
+    prompt = (
+        "A curious raccoon peers through a vibrant field of yellow sunflowers, its eyes "
+        "wide with interest. The playful yet serene atmosphere is complemented by soft "
+        "natural light filtering through the petals. Mid-shot, warm and cheerful tones."
+    )
+    _ = generator.generate_video(prompt, output_path=OUTPUT_PATH, save_video=True,height=512, width=768, num_frames=121)
+
+    prompt2 = (
+        "A majestic lion strides across the golden savanna, its powerful frame "
+        "glistening under the warm afternoon sun. The tall grass ripples gently in "
+        "the breeze, enhancing the lion's commanding presence. The tone is vibrant, "
+        "embodying the raw energy of the wild. Low angle, steady tracking shot, "
+        "cinematic.")
+    _ = generator.generate_video(prompt2, output_path=OUTPUT_PATH, save_video=True, height=512, width=768, num_frames=121)
+
+
+if __name__ == "__main__":
+    main()

examples/inference/basic/edit_glm_image.py

@@ -0,0 +1,120 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Run GLM-Image image-to-image (edit) generation through FastVideo.
+
+User story:
+    "I have the HF `zai-org/GLM-Image` checkpoint and a condition image, and
+    want a minimal edit command (text + image -> edited image), saved as a PNG."
+
+GLM-Image is a single unified pipeline: passing a condition image switches it
+from text-to-image to the edit path (the condition enters the DiT via a KV-cache
+write pass), so the generator config is identical to `basic_glm_image.py` — the
+`inputs.pil_image` on the request is what selects the edit mode.
+"""
+import argparse
+from pathlib import Path
+
+from PIL import Image
+
+from fastvideo import VideoGenerator
+from fastvideo.api import (
+    EngineConfig,
+    GenerationRequest,
+    GeneratorConfig,
+    InputConfig,
+    OutputConfig,
+    ParallelismConfig,
+    PipelineSelection,
+    SamplingConfig,
+)
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="Run GLM-Image image-to-image (edit) generation.")
+    parser.add_argument(
+        "--model-path",
+        default="zai-org/GLM-Image",
+        help="HF id or local diffusers-format GLM-Image weights directory.",
+    )
+    parser.add_argument(
+        "--image",
+        default="assets/images/couple.jpg",
+        help="Condition image to edit.",
+    )
+    parser.add_argument(
+        "--output",
+        default="image_output/edited.png",
+        help="Output PNG path.",
+    )
+    parser.add_argument(
+        "--prompt",
+        default="Change the background to a snowy mountain landscape at golden hour.",
+        help="Edit instruction.",
+    )
+    parser.add_argument("--height", type=int, default=1024)
+    parser.add_argument("--width", type=int, default=1024)
+    parser.add_argument("--steps", type=int, default=50)
+    parser.add_argument("--guidance-scale", type=float, default=1.5)
+    parser.add_argument("--seed", type=int, default=1024)
+    parser.add_argument("--num-gpus", type=int, default=1)
+    parser.add_argument("--tp-size", type=int, default=None)
+    parser.add_argument("--sp-size", type=int, default=None)
+    return parser.parse_args()
+
+
+def main() -> None:
+    args = parse_args()
+
+    output = Path(args.output)
+    output.parent.mkdir(parents=True, exist_ok=True)
+    condition = Image.open(args.image).convert("RGB")
+    tp_size = args.tp_size if args.tp_size is not None else (args.num_gpus if args.num_gpus > 1 else 1)
+    sp_size = args.sp_size if args.sp_size is not None else (1 if args.num_gpus > 1 else args.num_gpus)
+
+    # GLM-Image needs trust_remote_code for its AR encoder; offload and the
+    # pipeline class come from the model's registered defaults — don't override.
+    # The pipeline is registered as t2i; passing inputs.pil_image below switches
+    # it to the edit path.
+    generator_config = GeneratorConfig(
+        model_path=args.model_path,
+        trust_remote_code=True,
+        engine=EngineConfig(
+            num_gpus=args.num_gpus,
+            parallelism=ParallelismConfig(tp_size=tp_size, sp_size=sp_size),
+        ),
+        pipeline=PipelineSelection(workload_type="t2i"),
+    )
+
+    generator = VideoGenerator.from_config(generator_config)
+    try:
+        request = GenerationRequest(
+            prompt=args.prompt,
+            inputs=InputConfig(pil_image=condition),
+            sampling=SamplingConfig(
+                height=args.height,
+                width=args.width,
+                num_frames=1,
+                fps=1,
+                num_inference_steps=args.steps,
+                guidance_scale=args.guidance_scale,
+                seed=args.seed,
+            ),
+            output=OutputConfig(
+                output_path=str(output.parent),
+                save_video=False,
+                return_frames=True,
+            ),
+        )
+        result = generator.generate(request)
+        if isinstance(result, list):
+            result = result[0]
+
+        frames = result.frames
+        if frames is not None and len(frames):
+            Image.fromarray(frames[0]).save(output)
+            print(f"Saved image to {output}")
+    finally:
+        generator.shutdown()
+
+
+if __name__ == "__main__":
+    main()

examples/inference/ltx2_3/README.md

@@ -0,0 +1,69 @@
+# LTX-2.3 distilled inference configs
+
+Ready-to-run `fastvideo generate` run configs for the LTX-2.3
+distilled model (`FastVideo/LTX-2.3-Distilled-Diffusers`), covering both
+workloads (t2v / i2v), both two-stage step schedules (`5+2`, `8+3` = denoise
++ refine), and four resolutions.
+
+```bash
+fastvideo generate --config examples/inference/ltx2_3/t2v_8s3_1280x832.yaml
+```
+
+Each config is self-contained (no preset registry needed): the two-stage
+refine is wired via `generator.pipeline.preset_overrides.refine`, and the
+base sampling knobs live under `request.sampling`. The refine upsampler
+auto-resolves from the model's `spatial_upscaler`.
+
+## Configs
+
+| workload | schedule | resolution (HxW) | file |
+|---|---|---|---|
+| t2v | 5+2 | 1280x832 | `t2v_5s2_1280x832.yaml` |
+| t2v | 5+2 | 1024x1536 | `t2v_5s2_1024x1536.yaml` |
+| t2v | 5+2 | 768x1280 | `t2v_5s2_768x1280.yaml` |
+| t2v | 5+2 | 512x768 | `t2v_5s2_512x768.yaml` |
+| t2v | 8+3 | 1280x832 | `t2v_8s3_1280x832.yaml` |
+| t2v | 8+3 | 1024x1536 | `t2v_8s3_1024x1536.yaml` |
+| t2v | 8+3 | 768x1280 | `t2v_8s3_768x1280.yaml` |
+| t2v | 8+3 | 512x768 | `t2v_8s3_512x768.yaml` |
+| i2v | 5+2 | 1280x832 | `i2v_5s2_1280x832.yaml` |
+| i2v | 5+2 | 1024x1536 | `i2v_5s2_1024x1536.yaml` |
+| i2v | 5+2 | 768x1280 | `i2v_5s2_768x1280.yaml` |
+| i2v | 5+2 | 512x768 | `i2v_5s2_512x768.yaml` |
+| i2v | 8+3 | 1280x832 | `i2v_8s3_1280x832.yaml` |
+| i2v | 8+3 | 1024x1536 | `i2v_8s3_1024x1536.yaml` |
+| i2v | 8+3 | 768x1280 | `i2v_8s3_768x1280.yaml` |
+| i2v | 8+3 | 512x768 | `i2v_8s3_512x768.yaml` |
+
+## Overriding without editing a file
+
+Dotted overrides (prefixes `generator.` / `request.`) let you tweak any field:
+
+```bash
+# swap prompt
+fastvideo generate --config examples/inference/ltx2_3/t2v_8s3_1280x832.yaml \
+  --request.prompt "a red fox running through fresh snow"
+
+# change output path / gpu count
+fastvideo generate --config examples/inference/ltx2_3/t2v_5s2_512x768.yaml \
+  --request.output.output_path outputs/preview.mp4 \
+  --generator.engine.num_gpus 4
+```
+
+## i2v
+
+The `i2v_*` configs take a first-frame image via
+`request.extensions.ltx2_images` (`[[path, frame_offset, weight]]`). Edit the
+path in the file, or override it:
+
+```bash
+fastvideo generate --config examples/inference/ltx2_3/i2v_8s3_1280x832.yaml \
+  --request.extensions.ltx2_images '[["/data/portrait.jpg", 0, 1.0]]'
+```
+
+## Schedules
+
+`5+2` is the fast preview schedule; `8+3` is the higher-quality distilled
+recipe. Refine (`preset_overrides.refine.num_inference_steps`) only accepts 2
+or 3 steps.
+

examples/inference/ltx2_3/i2v_5s2_1024x1536.yaml

@@ -0,0 +1,38 @@
+# LTX-2.3 distilled i2v — 5+2 two-stage at 1024x1536.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/i2v_5s2_1024x1536.yaml
+#
+# Stage 1 denoises for 5 steps at half resolution; the latents are then
+# spatially upsampled and refined for 2 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: i2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 2
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    The subject slowly turns toward the camera with a soft, natural expression, hair and clothing swaying gently, shallow depth of field, subtle cinematic motion.
+  sampling:
+    height: 1024
+    width: 1536
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 5
+  # i2v conditioning — replace the path with your own first-frame image.
+  extensions:
+    ltx2_images:
+      - ["/path/to/your/first_frame.jpg", 0, 1.0]
+    ltx2_image_crf: 0.0
+  output:
+    output_path: outputs/ltx2_3_i2v_5s2_1024x1536.mp4
+    save_video: true

examples/inference/ltx2_3/i2v_5s2_1280x832.yaml

@@ -0,0 +1,38 @@
+# LTX-2.3 distilled i2v — 5+2 two-stage at 1280x832.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/i2v_5s2_1280x832.yaml
+#
+# Stage 1 denoises for 5 steps at half resolution; the latents are then
+# spatially upsampled and refined for 2 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: i2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 2
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    The subject slowly turns toward the camera with a soft, natural expression, hair and clothing swaying gently, shallow depth of field, subtle cinematic motion.
+  sampling:
+    height: 1280
+    width: 832
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 5
+  # i2v conditioning — replace the path with your own first-frame image.
+  extensions:
+    ltx2_images:
+      - ["/path/to/your/first_frame.jpg", 0, 1.0]
+    ltx2_image_crf: 0.0
+  output:
+    output_path: outputs/ltx2_3_i2v_5s2_1280x832.mp4
+    save_video: true

examples/inference/ltx2_3/i2v_5s2_512x768.yaml

@@ -0,0 +1,38 @@
+# LTX-2.3 distilled i2v — 5+2 two-stage at 512x768.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/i2v_5s2_512x768.yaml
+#
+# Stage 1 denoises for 5 steps at half resolution; the latents are then
+# spatially upsampled and refined for 2 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: i2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 2
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    The subject slowly turns toward the camera with a soft, natural expression, hair and clothing swaying gently, shallow depth of field, subtle cinematic motion.
+  sampling:
+    height: 512
+    width: 768
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 5
+  # i2v conditioning — replace the path with your own first-frame image.
+  extensions:
+    ltx2_images:
+      - ["/path/to/your/first_frame.jpg", 0, 1.0]
+    ltx2_image_crf: 0.0
+  output:
+    output_path: outputs/ltx2_3_i2v_5s2_512x768.mp4
+    save_video: true

examples/inference/ltx2_3/i2v_5s2_768x1280.yaml

@@ -0,0 +1,38 @@
+# LTX-2.3 distilled i2v — 5+2 two-stage at 768x1280.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/i2v_5s2_768x1280.yaml
+#
+# Stage 1 denoises for 5 steps at half resolution; the latents are then
+# spatially upsampled and refined for 2 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: i2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 2
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    The subject slowly turns toward the camera with a soft, natural expression, hair and clothing swaying gently, shallow depth of field, subtle cinematic motion.
+  sampling:
+    height: 768
+    width: 1280
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 5
+  # i2v conditioning — replace the path with your own first-frame image.
+  extensions:
+    ltx2_images:
+      - ["/path/to/your/first_frame.jpg", 0, 1.0]
+    ltx2_image_crf: 0.0
+  output:
+    output_path: outputs/ltx2_3_i2v_5s2_768x1280.mp4
+    save_video: true

examples/inference/ltx2_3/i2v_8s3_1024x1536.yaml

@@ -0,0 +1,38 @@
+# LTX-2.3 distilled i2v — 8+3 two-stage at 1024x1536.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/i2v_8s3_1024x1536.yaml
+#
+# Stage 1 denoises for 8 steps at half resolution; the latents are then
+# spatially upsampled and refined for 3 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: i2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 3
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    The subject slowly turns toward the camera with a soft, natural expression, hair and clothing swaying gently, shallow depth of field, subtle cinematic motion.
+  sampling:
+    height: 1024
+    width: 1536
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 8
+  # i2v conditioning — replace the path with your own first-frame image.
+  extensions:
+    ltx2_images:
+      - ["/path/to/your/first_frame.jpg", 0, 1.0]
+    ltx2_image_crf: 0.0
+  output:
+    output_path: outputs/ltx2_3_i2v_8s3_1024x1536.mp4
+    save_video: true

examples/inference/ltx2_3/i2v_8s3_1280x832.yaml

@@ -0,0 +1,38 @@
+# LTX-2.3 distilled i2v — 8+3 two-stage at 1280x832.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/i2v_8s3_1280x832.yaml
+#
+# Stage 1 denoises for 8 steps at half resolution; the latents are then
+# spatially upsampled and refined for 3 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: i2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 3
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    The subject slowly turns toward the camera with a soft, natural expression, hair and clothing swaying gently, shallow depth of field, subtle cinematic motion.
+  sampling:
+    height: 1280
+    width: 832
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 8
+  # i2v conditioning — replace the path with your own first-frame image.
+  extensions:
+    ltx2_images:
+      - ["/path/to/your/first_frame.jpg", 0, 1.0]
+    ltx2_image_crf: 0.0
+  output:
+    output_path: outputs/ltx2_3_i2v_8s3_1280x832.mp4
+    save_video: true

examples/inference/ltx2_3/i2v_8s3_512x768.yaml

@@ -0,0 +1,38 @@
+# LTX-2.3 distilled i2v — 8+3 two-stage at 512x768.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/i2v_8s3_512x768.yaml
+#
+# Stage 1 denoises for 8 steps at half resolution; the latents are then
+# spatially upsampled and refined for 3 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: i2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 3
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    The subject slowly turns toward the camera with a soft, natural expression, hair and clothing swaying gently, shallow depth of field, subtle cinematic motion.
+  sampling:
+    height: 512
+    width: 768
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 8
+  # i2v conditioning — replace the path with your own first-frame image.
+  extensions:
+    ltx2_images:
+      - ["/path/to/your/first_frame.jpg", 0, 1.0]
+    ltx2_image_crf: 0.0
+  output:
+    output_path: outputs/ltx2_3_i2v_8s3_512x768.mp4
+    save_video: true

examples/inference/ltx2_3/i2v_8s3_768x1280.yaml

@@ -0,0 +1,38 @@
+# LTX-2.3 distilled i2v — 8+3 two-stage at 768x1280.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/i2v_8s3_768x1280.yaml
+#
+# Stage 1 denoises for 8 steps at half resolution; the latents are then
+# spatially upsampled and refined for 3 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: i2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 3
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    The subject slowly turns toward the camera with a soft, natural expression, hair and clothing swaying gently, shallow depth of field, subtle cinematic motion.
+  sampling:
+    height: 768
+    width: 1280
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 8
+  # i2v conditioning — replace the path with your own first-frame image.
+  extensions:
+    ltx2_images:
+      - ["/path/to/your/first_frame.jpg", 0, 1.0]
+    ltx2_image_crf: 0.0
+  output:
+    output_path: outputs/ltx2_3_i2v_8s3_768x1280.mp4
+    save_video: true

examples/inference/ltx2_3/t2v_5s2_1024x1536.yaml

@@ -0,0 +1,33 @@
+# LTX-2.3 distilled t2v — 5+2 two-stage at 1024x1536.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/t2v_5s2_1024x1536.yaml
+#
+# Stage 1 denoises for 5 steps at half resolution; the latents are then
+# spatially upsampled and refined for 2 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: t2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 2
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    A cinematic drone shot flying over dramatic coastal cliffs at sunrise, golden light spilling across the water, gentle waves breaking on the rocks below, ultra-detailed, smooth camera motion.
+  sampling:
+    height: 1024
+    width: 1536
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 5
+  output:
+    output_path: outputs/ltx2_3_t2v_5s2_1024x1536.mp4
+    save_video: true

examples/inference/ltx2_3/t2v_5s2_1280x832.yaml

@@ -0,0 +1,33 @@
+# LTX-2.3 distilled t2v — 5+2 two-stage at 1280x832.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/t2v_5s2_1280x832.yaml
+#
+# Stage 1 denoises for 5 steps at half resolution; the latents are then
+# spatially upsampled and refined for 2 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: t2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 2
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    A cinematic drone shot flying over dramatic coastal cliffs at sunrise, golden light spilling across the water, gentle waves breaking on the rocks below, ultra-detailed, smooth camera motion.
+  sampling:
+    height: 1280
+    width: 832
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 5
+  output:
+    output_path: outputs/ltx2_3_t2v_5s2_1280x832.mp4
+    save_video: true

examples/inference/ltx2_3/t2v_5s2_512x768.yaml

@@ -0,0 +1,33 @@
+# LTX-2.3 distilled t2v — 5+2 two-stage at 512x768.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/t2v_5s2_512x768.yaml
+#
+# Stage 1 denoises for 5 steps at half resolution; the latents are then
+# spatially upsampled and refined for 2 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: t2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 2
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    A cinematic drone shot flying over dramatic coastal cliffs at sunrise, golden light spilling across the water, gentle waves breaking on the rocks below, ultra-detailed, smooth camera motion.
+  sampling:
+    height: 512
+    width: 768
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 5
+  output:
+    output_path: outputs/ltx2_3_t2v_5s2_512x768.mp4
+    save_video: true

examples/inference/ltx2_3/t2v_5s2_768x1280.yaml

@@ -0,0 +1,33 @@
+# LTX-2.3 distilled t2v — 5+2 two-stage at 768x1280.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/t2v_5s2_768x1280.yaml
+#
+# Stage 1 denoises for 5 steps at half resolution; the latents are then
+# spatially upsampled and refined for 2 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: t2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 2
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    A cinematic drone shot flying over dramatic coastal cliffs at sunrise, golden light spilling across the water, gentle waves breaking on the rocks below, ultra-detailed, smooth camera motion.
+  sampling:
+    height: 768
+    width: 1280
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 5
+  output:
+    output_path: outputs/ltx2_3_t2v_5s2_768x1280.mp4
+    save_video: true

examples/inference/ltx2_3/t2v_8s3_1024x1536.yaml

@@ -0,0 +1,33 @@
+# LTX-2.3 distilled t2v — 8+3 two-stage at 1024x1536.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/t2v_8s3_1024x1536.yaml
+#
+# Stage 1 denoises for 8 steps at half resolution; the latents are then
+# spatially upsampled and refined for 3 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: t2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 3
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    A cinematic drone shot flying over dramatic coastal cliffs at sunrise, golden light spilling across the water, gentle waves breaking on the rocks below, ultra-detailed, smooth camera motion.
+  sampling:
+    height: 1024
+    width: 1536
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 8
+  output:
+    output_path: outputs/ltx2_3_t2v_8s3_1024x1536.mp4
+    save_video: true

examples/inference/ltx2_3/t2v_8s3_1280x832.yaml

@@ -0,0 +1,33 @@
+# LTX-2.3 distilled t2v — 8+3 two-stage at 1280x832.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/t2v_8s3_1280x832.yaml
+#
+# Stage 1 denoises for 8 steps at half resolution; the latents are then
+# spatially upsampled and refined for 3 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: t2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 3
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    A cinematic drone shot flying over dramatic coastal cliffs at sunrise, golden light spilling across the water, gentle waves breaking on the rocks below, ultra-detailed, smooth camera motion.
+  sampling:
+    height: 1280
+    width: 832
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 8
+  output:
+    output_path: outputs/ltx2_3_t2v_8s3_1280x832.mp4
+    save_video: true

examples/inference/ltx2_3/t2v_8s3_512x768.yaml

@@ -0,0 +1,33 @@
+# LTX-2.3 distilled t2v — 8+3 two-stage at 512x768.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/t2v_8s3_512x768.yaml
+#
+# Stage 1 denoises for 8 steps at half resolution; the latents are then
+# spatially upsampled and refined for 3 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: t2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 3
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    A cinematic drone shot flying over dramatic coastal cliffs at sunrise, golden light spilling across the water, gentle waves breaking on the rocks below, ultra-detailed, smooth camera motion.
+  sampling:
+    height: 512
+    width: 768
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 8
+  output:
+    output_path: outputs/ltx2_3_t2v_8s3_512x768.mp4
+    save_video: true

examples/inference/ltx2_3/t2v_8s3_768x1280.yaml

@@ -0,0 +1,33 @@
+# LTX-2.3 distilled t2v — 8+3 two-stage at 768x1280.
+#
+# Run:
+#   fastvideo generate --config examples/inference/ltx2_3/t2v_8s3_768x1280.yaml
+#
+# Stage 1 denoises for 8 steps at half resolution; the latents are then
+# spatially upsampled and refined for 3 steps (refine only supports 2 or 3).
+# The refine upsampler auto-resolves from the model's `spatial_upscaler`.
+generator:
+  model_path: FastVideo/LTX-2.3-Distilled-Diffusers
+  engine:
+    num_gpus: 1
+  pipeline:
+    workload_type: t2v
+    preset_overrides:
+      refine:
+        enabled: true
+        num_inference_steps: 3
+        guidance_scale: 1.0
+        add_noise: true
+request:
+  prompt: >-
+    A cinematic drone shot flying over dramatic coastal cliffs at sunrise, golden light spilling across the water, gentle waves breaking on the rocks below, ultra-detailed, smooth camera motion.
+  sampling:
+    height: 768
+    width: 1280
+    num_frames: 121
+    fps: 24
+    guidance_scale: 1.0
+    num_inference_steps: 8
+  output:
+    output_path: outputs/ltx2_3_t2v_8s3_768x1280.mp4
+    save_video: true

examples/inference/optimizations/FastWan_QAD_TAEHV.py

@@ -1,316 +0,0 @@
-"""Fast NVFP4 linear inference for Wan2.1-T2V-1.3B with TAEHV decoding.
-
-This is the FP4-linear fast path from ``fp4_linear_wan2_1_1_3b.py`` with the
-heavy Wan VAE swapped out for TAEHV -- a tiny autoencoder that decodes Wan2.1
-latents directly (no denormalization) and is dramatically faster / lighter.
-
-How it works: the generator runs with ``output_type="latent"`` so the pipeline
-returns raw denoised latents instead of pixels (the Wan VAE is offloaded and
-never used). We then decode those latents with TAEHV in this script and save
-the frames ourselves. This mirrors the FastVideo-Quantization
-``quantization_example_taehv.py`` proof-of-concept, but kept clean: TAEHV is a
-pip package (no ``sys.path`` hacks), the latent->uint8 conversion is vectorized,
-and there is no dead profiler / sanitization code.
-
-Requirements:
-    - Blackwell GPU (B200/B300, sm100a/sm103a) for the FP4 linear path
-    - flashinfer (``pip install flashinfer-python``)
-    - TAEHV weights ``taew2_1.pth`` (https://github.com/madebyollin/taehv)
-
-Usage:
-    python fp4_linear_taehv_wan2_1_1_3b.py                    # FP4 + TAEHV + compile
-    python fp4_linear_taehv_wan2_1_1_3b.py --no-taehv         # FP4 + full Wan VAE
-    python fp4_linear_taehv_wan2_1_1_3b.py --no-compile       # eager
-    python fp4_linear_taehv_wan2_1_1_3b.py --baseline         # dense bf16 reference
-    python fp4_linear_taehv_wan2_1_1_3b.py --distilled_model ''  # base Wan2.1 weights
-    python fp4_linear_taehv_wan2_1_1_3b.py --warmups 5 --benchmark-runs 20  # timing stats (default)
-"""
-
-import argparse
-import contextlib
-import logging
-import os
-import statistics
-import time
-
-import imageio
-import torch
-
-from fastvideo import VideoGenerator
-from fastvideo.configs.pipelines.base import PipelineConfig
-from fastvideo.layers.quantization.nvfp4_qat_config import NVFP4QATConfig
-
-OUTPUT_PATH = "video_samples"
-
-# Distilled, quantization-aware (QAD) transformer for Wan2.1-1.3B (3 steps,
-# guidance 1.0). Loaded on top of the base Wan2.1 pipeline; pass
-# ``--distilled_model ''`` to run the base weights instead.
-DEFAULT_DISTILLED_MODEL = "FastVideo/FastWan-QAD-1.3B"
-DISTILLED_WEIGHTS_FILE = (
-    "generator_inference_transformer/diffusion_pytorch_model.safetensors"
-)
-
-# TAEHV checkpoint for Wan2.1. Clone https://github.com/madebyollin/taehv to get
-# ``taew2_1.pth`` (Wan 2.1 / Wan 2.2-14B / Qwen-Image all use this VAE).
-DEFAULT_TAEHV_CHECKPOINT = "/root/taehv/taew2_1.pth"
-
-PROMPT = (
-    "A curious raccoon peers through a vibrant field of yellow sunflowers, its eyes "
-    "wide with interest. The playful yet serene atmosphere is complemented by soft "
-    "natural light filtering through the petals. Mid-shot, warm and cheerful tones."
-)
-
-
-class TaehvDecoder:
-    """Thin wrapper around the TAEHV tiny autoencoder for Wan2.1 latents.
-
-    TAEHV consumes the *normalized* latents the diffusion model produces (the
-    same representation FastVideo carries internally), so no denormalization is
-    needed -- unlike the full Wan VAE path.
-    """
-
-    def __init__(self, checkpoint_path: str, device: str = "cuda",
-                 dtype: torch.dtype = torch.float16) -> None:
-        from taehv import TAEHV  # pip-installed; no sys.path manipulation
-        self.device = device
-        self.dtype = dtype
-        print(f"Loading TAEHV from {checkpoint_path} ...")
-        self.model = TAEHV(checkpoint_path=checkpoint_path).to(device, dtype).eval()
-
-    @torch.no_grad()
-    def decode(self, latents: torch.Tensor):
-        """Decode FastVideo latents into uint8 RGB frames.
-
-        Args:
-            latents: ``[B, C, T, H, W]`` (NCTHW) normalized latent tensor.
-
-        Returns:
-            A ``(T, H, W, 3)`` uint8 numpy array ready for ``imageio.mimsave``.
-        """
-        # NCTHW -> NTCHW (TAEHV's expected layout), on the TAEHV device/dtype.
-        latents = latents.permute(0, 2, 1, 3, 4).to(self.device, self.dtype)
-        decoded = self.model.decode_video(
-            latents, parallel=True, show_progress_bar=False)
-        # decoded: [B, T, 3, H, W] in [0, 1]. Take batch 0, vectorize to uint8.
-        frames = (decoded[0].clamp(0, 1) * 255).to(torch.uint8)
-        return frames.permute(0, 2, 3, 1).cpu().numpy()
-
-
-def resolve_distilled_weights(hf_id: str) -> str:
-    """Return a local path to the distilled transformer safetensors."""
-    if os.path.exists(hf_id):
-        return hf_id
-    from huggingface_hub import hf_hub_download
-    return hf_hub_download(repo_id=hf_id, filename=DISTILLED_WEIGHTS_FILE)
-
-
-@contextlib.contextmanager
-def silence_request_log():
-    """Quiet ``VideoGenerator.generate``'s per-request config printout.
-
-    Each ``generate(...)`` call logs a multi-line debug block (height/width/
-    prompt/steps/...) at INFO via ``logger.info`` in
-    ``fastvideo.entrypoints.video_generator``. There is no built-in switch,
-    so this context manager raises that logger's level to WARNING while the
-    warmup calls run, then restores it for the timed run.
-    """
-    vg_logger = logging.getLogger("fastvideo.entrypoints.video_generator")
-    prev_level = vg_logger.level
-    vg_logger.setLevel(logging.WARNING)
-    try:
-        yield
-    finally:
-        vg_logger.setLevel(prev_level)
-
-
-def resolve_taehv_checkpoint(path: str) -> str:
-    """Validate the TAEHV checkpoint path, with a helpful error if missing."""
-    if os.path.exists(path):
-        return path
-    raise FileNotFoundError(
-        f"TAEHV checkpoint not found at {path!r}. Clone the weights with:\n"
-        "    git clone https://github.com/madebyollin/taehv\n"
-        "and pass --taehv_checkpoint <repo>/taew2_1.pth")
-
-
-def build_generator(args: argparse.Namespace) -> VideoGenerator:
-    model_id = args.model
-
-    # Half precision everywhere; DiT linears are additionally NVFP4-quantized
-    # via dit_config.quant_config below.
-    pipeline_config = PipelineConfig.from_pretrained(model_id)
-    pipeline_config.dit_precision = "bf16"
-    pipeline_config.vae_precision = "bf16"
-    pipeline_config.text_encoder_precisions = ("bf16",)
-
-    if not args.baseline:
-        pipeline_config.dit_config.quant_config = NVFP4QATConfig()
-
-    compile_enabled = not args.no_compile
-
-    extra_kwargs = {}
-    if args.distilled_model:
-        weights_path = resolve_distilled_weights(args.distilled_model)
-        print(f"Using distilled weights: {args.distilled_model} -> {weights_path}")
-        extra_kwargs["init_weights_from_safetensors"] = weights_path
-
-    if args.taehv:
-        # Skip the in-pipeline VAE decode entirely: the pipeline returns raw
-        # latents, the Wan VAE is offloaded to CPU (and not compiled) since we
-        # decode with TAEHV in this script instead.
-        extra_kwargs["output_type"] = "latent"
-
-    generator = VideoGenerator.from_pretrained(
-        model_id,
-        pipeline_config=pipeline_config,
-        num_gpus=args.num_gpus,
-        # Keep everything resident on the GPU -- no offloading, except the
-        # unused Wan VAE when TAEHV handles decoding.
-        use_fsdp_inference=False,
-        dit_cpu_offload=False,
-        dit_layerwise_offload=False,
-        vae_cpu_offload=args.taehv,
-        text_encoder_cpu_offload=False,
-        pin_cpu_memory=False,
-        enable_torch_compile=compile_enabled,
-        enable_torch_compile_text_encoder=compile_enabled,
-        enable_torch_compile_vae=compile_enabled and not args.taehv,
-        **extra_kwargs,
-    )
-    return generator
-
-
-def main() -> None:
-    parser = argparse.ArgumentParser(
-        description="FP4 linear Wan2.1-1.3B with TAEHV decoding benchmark")
-    parser.add_argument("--model", default="Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
-                        help="Model path or HuggingFace ID")
-    parser.add_argument("--baseline", action="store_true",
-                        help="Run dense bf16 instead of FP4 linear")
-    parser.add_argument("--no-compile", action="store_true",
-                        help="Disable torch.compile (eager)")
-    parser.add_argument("--taehv", action=argparse.BooleanOptionalAction,
-                        default=True,
-                        help="Decode with TAEHV instead of the full Wan VAE "
-                             "(use --no-taehv for the Wan VAE path)")
-    parser.add_argument("--taehv_checkpoint", default=DEFAULT_TAEHV_CHECKPOINT,
-                        help="Path to the TAEHV taew2_1.pth checkpoint")
-    parser.add_argument("--distilled_model", default=DEFAULT_DISTILLED_MODEL,
-                        help="HuggingFace ID (or local path) of a distilled "
-                             "transformer checkpoint to load on top of --model. "
-                             "Pass '' to use the base --model weights instead.")
-    parser.add_argument("--num_gpus", type=int, default=1)
-    parser.add_argument("--infer_steps", type=int, default=3)
-    parser.add_argument("--guidance_scale", type=float, default=1.0)
-    parser.add_argument("--warmups", type=int, default=5,
-                        help="Warmup runs before timing (default: 5).")
-    parser.add_argument("--benchmark-runs", type=int, default=20,
-                        help="Timed runs to collect min/max/mean/std over (default: 20).")
-    args = parser.parse_args()
-
-    if not torch.cuda.is_available():
-        raise SystemExit("CUDA is required for FP4 inference.")
-
-    cap = torch.cuda.get_device_capability()
-    print(f"GPU: {torch.cuda.get_device_name()} (capability {cap[0]}.{cap[1]})")
-    if not args.baseline and cap[0] < 10:
-        print("Warning: NVFP4 requires Blackwell (capability 10.0+); "
-              "FP4 kernels may be unavailable on this GPU.")
-
-    mode = "bf16" if args.baseline else "fp4_linear"
-    mode += "_taehv" if args.taehv else "_wanvae"
-    if not args.no_compile:
-        mode += "_compile"
-    print(f"Mode: {mode.upper()}")
-
-    # Load TAEHV before the (slow) generator build so a bad checkpoint path
-    # fails fast.
-    taehv = TaehvDecoder(resolve_taehv_checkpoint(args.taehv_checkpoint)) \
-        if args.taehv else None
-
-    generator = build_generator(args)
-
-    os.makedirs(OUTPUT_PATH, exist_ok=True)
-
-    # Warmup: pay the DiT torch.compile cost and warm TAEHV's cuDNN algo
-    # selection + allocator growth (~0.2s on the first decode) so the timed
-    # runs below measure steady-state latency only.
-    with silence_request_log():
-        for _ in range(args.warmups):
-            warm = generator.generate(request={
-                "prompt": PROMPT,
-                "sampling": {"num_inference_steps": 2, "guidance_scale": args.guidance_scale},
-                "output": {"save_video": False, "return_frames": args.taehv},
-            })
-            if args.taehv:
-                taehv.decode(warm.samples)
-
-    output_path = os.path.join(OUTPUT_PATH, f"raccoon_{mode}.mp4")
-
-    # Benchmark: time each run end-to-end. ``denoise`` is the generator's own
-    # generation_time; for TAEHV we add the in-script decode. The mp4 is not
-    # written inside the loop (only the final run's frames are saved below) so
-    # disk I/O never pollutes the timings.
-    denoise_times: list[float] = []
-    decode_times: list[float] = []
-    totals: list[float] = []
-    frames = None
-    with silence_request_log():
-        for i in range(args.benchmark_runs):
-            result = generator.generate(request={
-                "prompt": PROMPT,
-                "sampling": {
-                    "num_inference_steps": args.infer_steps,
-                    "guidance_scale": args.guidance_scale,
-                },
-                "output": {
-                    "save_video": False,
-                    "return_frames": args.taehv,
-                    "output_path": output_path,
-                },
-            })
-            denoise_elapsed = result.generation_time
-            denoise_times.append(denoise_elapsed)
-
-            if args.taehv:
-                torch.cuda.synchronize()
-                decode_start = time.perf_counter()
-                frames = taehv.decode(result.samples)
-                torch.cuda.synchronize()
-                decode_elapsed = time.perf_counter() - decode_start
-                decode_times.append(decode_elapsed)
-                total = denoise_elapsed + decode_elapsed
-            else:
-                total = denoise_elapsed
-            totals.append(total)
-
-            line = f"  run {i + 1:02d}/{args.benchmark_runs}: {total:.3f}s"
-            if args.taehv:
-                line += (f" (denoise {denoise_elapsed:.3f}s + "
-                         f"decode {decode_elapsed:.3f}s)")
-            print(line)
-
-    if args.taehv and frames is not None:
-        imageio.mimsave(output_path, frames, fps=16, format="mp4")
-        print(f"Saved video to {output_path}")
-
-    # Report min / max / mean / std over the timed runs.
-    def _stat_row(name: str, xs: list[float]) -> str:
-        std = statistics.stdev(xs) if len(xs) > 1 else 0.0
-        return (f"  {name:<11}{min(xs):>8.3f}{max(xs):>9.3f}"
-                f"{statistics.mean(xs):>9.3f}{std:>9.3f}")
-
-    if totals:
-        print(f"\n[{mode.upper()}] {args.benchmark_runs} runs, {args.warmups} warmup, "
-              f"{args.infer_steps} steps:")
-        print(f"  {'metric':<11}{'min':>8}{'max':>9}{'mean':>9}{'std':>9}   (s)")
-        print(_stat_row("total", totals))
-        if args.taehv:
-            print(_stat_row("denoise", denoise_times))
-            print(_stat_row("decode", decode_times))
-
-    generator.shutdown()
-
-
-if __name__ == "__main__":
-    main()

examples/inference/optimizations/_qad_common.py

@@ -0,0 +1,30 @@
+"""Shared GPU-arch helpers for the QAD 5090 example scripts.
+
+The default (non ``--bf16``) path of the NVFP4 scripts runs DiT linears through
+flashinfer's cutlass FP4 gemm, which ships sm_120a cubins only (RTX 5090-class
+consumer Blackwell). These helpers pick the right FLASHINFER_CUDA_ARCH_LIST and
+fail fast with a readable error instead of an opaque flashinfer ValueError.
+"""
+
+import torch
+
+# Capabilities the flashinfer FP4 gemm path has cubins for.
+FP4_CAPABILITIES = ((12, 0), (12, 1))
+
+
+def flashinfer_arch_list() -> str:
+    """FLASHINFER_CUDA_ARCH_LIST derived from the local GPU, defaulting to 12.0a."""
+    if torch.cuda.is_available():
+        major, minor = torch.cuda.get_device_capability()
+        if (major, minor) in FP4_CAPABILITIES:
+            return f"{major}.{minor}a"
+    return "12.0a"
+
+
+def require_fp4_capable_gpu() -> None:
+    """Exit with a one-line error if the GPU cannot run the NVFP4 gemm path."""
+    cap = torch.cuda.get_device_capability() if torch.cuda.is_available() else None
+    if cap not in FP4_CAPABILITIES:
+        raise SystemExit(
+            f"The default NVFP4 path requires an sm_120a GPU (RTX 5090-class Blackwell, "
+            f"compute capability 12.0/12.1); found {cap}. Re-run with --bf16 on other GPUs.")

examples/inference/optimizations/benchmark_qad_wan2_1_1_3b.py

@@ -0,0 +1,222 @@
+"""Standalone benchmark harness for the FastWan-QAD Wan2.1-1.3B recipes.
+
+Measures steady-state end-to-end generation latency (denoise + optional TAEHV
+decode) over N timed runs after a warmup phase, and prints a min/max/mean/std
+stats table. This is the multi-run timing machinery that used to live in
+``FastWan_QAD_TAEHV.py``, kept as a separate harness so the example scripts
+stay lean single-run examples.
+
+Pipeline setup mirrors the corresponding example script in this directory:
+    fp8        -> fp8_wan2_1_1_3b.py        (FastVideo/FastWan-QAD-FP8-1.3B, SAGE_ATTN)
+    nvfp4_qat  -> nvfp4_qat_wan2_1_1_3b.py  (FastVideo/FastWan-QAD-1.3B, ATTN_QAT_INFER)
+    nvfp4_sa2  -> nvfp4_sa2_wan2_1_1_3b.py  (FastVideo/FastWan-QAD-1.3B-SA2, SAGE_ATTN)
+
+Hardware requirements per mode (default quantized paths; --bf16 relaxes them):
+    fp8        sm89+ (H100, L40S, RTX 4090, Ada Lovelace, or newer)
+    nvfp4_qat  RTX 5090-class Blackwell (sm_120a): flashinfer FP4 gemm + attn_qat_infer
+    nvfp4_sa2  RTX 5090-class Blackwell (sm_120a): flashinfer FP4 gemm + SageAttention2
+
+Usage:
+    python benchmark_qad_wan2_1_1_3b.py --mode nvfp4_qat --taehv-checkpoint /path/to/taehv/taew2_1.pth
+    python benchmark_qad_wan2_1_1_3b.py --mode fp8 --warmups 5 --runs 20
+    python benchmark_qad_wan2_1_1_3b.py --mode nvfp4_sa2 --bf16    # BF16 baseline
+"""
+
+import argparse
+import contextlib
+import logging
+import os
+import statistics
+import time
+
+import torch
+
+from _qad_common import flashinfer_arch_list, require_fp4_capable_gpu
+
+OUTPUT_PATH = "video_samples"
+
+# mode -> (default model, attention backend, needs the flashinfer FP4 gemm path)
+MODES = {
+    "fp8": ("FastVideo/FastWan-QAD-FP8-1.3B", "SAGE_ATTN", False),
+    "nvfp4_qat": ("FastVideo/FastWan-QAD-1.3B", "ATTN_QAT_INFER", True),
+    "nvfp4_sa2": ("FastVideo/FastWan-QAD-1.3B-SA2", "SAGE_ATTN", True),
+}
+
+PROMPT = (
+    "A curious raccoon peers through a vibrant field of yellow sunflowers, its eyes "
+    "wide with interest. The playful yet serene atmosphere is complemented by soft "
+    "natural light filtering through the petals. Mid-shot, warm and cheerful tones."
+)
+
+
+class TaehvDecoder:
+    def __init__(self, checkpoint_path: str, device: str = "cuda",
+                 dtype: torch.dtype = torch.float16) -> None:
+        from taehv import TAEHV
+        self.device = device
+        self.dtype = dtype
+        print(f"Loading TAEHV from {checkpoint_path} ...")
+        self.model = TAEHV(checkpoint_path=checkpoint_path).to(device, dtype).eval()
+
+    @torch.no_grad()
+    def decode(self, latents: torch.Tensor):
+        latents = latents.permute(0, 2, 1, 3, 4).to(self.device, self.dtype)
+        decoded = self.model.decode_video(latents, parallel=True, show_progress_bar=False)
+        frames = (decoded[0].clamp(0, 1) * 255).to(torch.uint8)
+        return frames.permute(0, 2, 3, 1).cpu().numpy()
+
+
+@contextlib.contextmanager
+def silence_request_log():
+    """Quiet ``VideoGenerator.generate``'s per-request multi-line INFO printout
+    so the per-run timing lines stay readable."""
+    vg_logger = logging.getLogger("fastvideo.entrypoints.video_generator")
+    prev_level = vg_logger.level
+    vg_logger.setLevel(logging.WARNING)
+    try:
+        yield
+    finally:
+        vg_logger.setLevel(prev_level)
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Multi-run QAD Wan2.1-1.3B benchmark")
+    parser.add_argument("--mode", required=True, choices=sorted(MODES),
+                        help="Which QAD recipe to benchmark (mirrors the example script of the same name)")
+    parser.add_argument("--warmups", type=int, default=5,
+                        help="Warmup runs before timing (default: 5)")
+    parser.add_argument("--runs", type=int, default=20,
+                        help="Timed runs to collect min/max/mean/std over (default: 20)")
+    parser.add_argument("--bf16", action="store_true", help="BF16 baseline (no quantization)")
+    parser.add_argument("--granularity", choices=["tensor", "channel"], default="tensor",
+                        help="FP8 weight scale granularity (fp8 mode only)")
+    parser.add_argument("--taehv-checkpoint", default=None, metavar="PATH",
+                        help="Path to taew2_1.pth; enables TAEHV tiny autoencoder decoding")
+    parser.add_argument("--model", default=None,
+                        help="Model path or HuggingFace ID (default: per --mode)")
+    parser.add_argument("--no-compile", action="store_true", help="Disable torch.compile for the DiT")
+    parser.add_argument("--num_gpus", type=int, default=1)
+    parser.add_argument("--infer_steps", type=int, default=3)
+    args = parser.parse_args()
+
+    model, backend, needs_fp4 = MODES[args.mode]
+    if args.model:
+        model = args.model
+
+    if needs_fp4 and not args.bf16:
+        require_fp4_capable_gpu()
+
+    os.environ.setdefault("FASTVIDEO_ATTENTION_BACKEND", backend)
+    if needs_fp4:
+        os.environ["FASTVIDEO_DISABLE_ATTENTION_COMPILE"] = "0"
+        os.environ.setdefault("FLASHINFER_CUDA_ARCH_LIST", flashinfer_arch_list())
+
+    from fastvideo import VideoGenerator
+    from fastvideo.configs.pipelines.base import PipelineConfig
+
+    mode = "bf16" if args.bf16 else args.mode
+    if not args.no_compile:
+        mode += "_compile"
+    use_taehv = args.taehv_checkpoint is not None
+    print(f"Mode: {mode.upper()}  model={model}  " + ("decoder=TAEHV" if use_taehv else "decoder=VAE"))
+
+    taehv = TaehvDecoder(args.taehv_checkpoint) if use_taehv else None
+
+    pipeline_config = PipelineConfig.from_pretrained(model)
+    pipeline_config.text_encoder_precisions = ("bf16",)
+    if not args.bf16:
+        if args.mode == "fp8":
+            from fastvideo.layers.quantization import get_quantization_config
+            pipeline_config.dit_config.quant_config = get_quantization_config("FP8")(
+                granularity=args.granularity)
+        else:
+            from fastvideo.layers.quantization.nvfp4_qat_config import NVFP4QATConfig
+            pipeline_config.dit_config.quant_config = NVFP4QATConfig()
+
+    generator = VideoGenerator.from_pretrained(
+        model,
+        pipeline_config=pipeline_config,
+        num_gpus=args.num_gpus,
+        use_fsdp_inference=False,
+        dit_cpu_offload=False,
+        dit_layerwise_offload=False,
+        vae_cpu_offload=use_taehv,
+        text_encoder_cpu_offload=False,
+        pin_cpu_memory=False,
+        enable_torch_compile=not args.no_compile,
+        enable_torch_compile_text_encoder=not args.no_compile,
+        enable_torch_compile_vae=not args.no_compile and not use_taehv,
+        output_type="latent" if use_taehv else "pil",
+    )
+
+    request = {
+        "prompt": PROMPT,
+        "sampling": {"num_inference_steps": args.infer_steps, "guidance_scale": 1.0},
+        "output": {"save_video": False},
+    }
+
+    # Warmup: pay the torch.compile cost and warm TAEHV's cuDNN algo selection
+    # so the timed runs below measure steady-state latency only.
+    with silence_request_log():
+        for _ in range(args.warmups):
+            warm = generator.generate(request=request)
+            if use_taehv:
+                taehv.decode(warm.samples)
+
+    # Benchmark: time each run end-to-end. ``denoise`` is the generator's own
+    # generation_time; for TAEHV we add the in-script decode. Nothing is written
+    # to disk inside the loop so I/O never pollutes the timings.
+    denoise_times: list[float] = []
+    decode_times: list[float] = []
+    totals: list[float] = []
+    frames = None
+    with silence_request_log():
+        for i in range(args.runs):
+            result = generator.generate(request=request)
+            denoise_elapsed = result.generation_time
+            denoise_times.append(denoise_elapsed)
+
+            if use_taehv:
+                torch.cuda.synchronize()
+                decode_start = time.perf_counter()
+                frames = taehv.decode(result.samples)
+                torch.cuda.synchronize()
+                decode_elapsed = time.perf_counter() - decode_start
+                decode_times.append(decode_elapsed)
+                total = denoise_elapsed + decode_elapsed
+            else:
+                total = denoise_elapsed
+            totals.append(total)
+
+            line = f"  run {i + 1:02d}/{args.runs}: {total:.3f}s"
+            if use_taehv:
+                line += f" (denoise {denoise_elapsed:.3f}s + decode {decode_elapsed:.3f}s)"
+            print(line)
+
+    if frames is not None:
+        os.makedirs(OUTPUT_PATH, exist_ok=True)
+        import imageio
+        output_path = os.path.join(OUTPUT_PATH, f"raccoon_{mode}.mp4")
+        imageio.mimsave(output_path, frames, fps=16, format="mp4")
+        print(f"Saved video to {output_path}")
+
+    # Report min / max / mean / std over the timed runs.
+    def _stat_row(name: str, xs: list[float]) -> str:
+        std = statistics.stdev(xs) if len(xs) > 1 else 0.0
+        return (f"  {name:<11}{min(xs):>8.3f}{max(xs):>9.3f}"
+                f"{statistics.mean(xs):>9.3f}{std:>9.3f}")
+
+    if totals:
+        print(f"\n[{mode.upper()}] {args.runs} runs, {args.warmups} warmup, "
+              f"{args.infer_steps} steps:")
+        print(f"  {'metric':<11}{'min':>8}{'max':>9}{'mean':>9}{'std':>9}   (s)")
+        print(_stat_row("total", totals))
+        if use_taehv:
+            print(_stat_row("denoise", denoise_times))
+            print(_stat_row("decode", decode_times))
+
+    generator.shutdown()
+
+
+if __name__ == "__main__":
+    main()

examples/inference/optimizations/fp8_wan2_1_1_3b.py

@@ -1,25 +1,23 @@
 """FP8 weight quantization inference example.
 
-Runs Wan2.1-T2V-1.3B with FP8 e4m3 quantized DiT linear layers (attention
-projections and FFN). Weights are quantized in-place after loading; activations
-are quantized dynamically at runtime. Reduces GPU memory relative to BF16 and
-can improve throughput on sm89+ GPUs.
+Runs FastWan-QAD-FP8-1.3B (a distilled Wan2.1-T2V-1.3B-Diffusers checkpoint) with
+FP8 e4m3 quantized DiT linear layers. Uses the SAGE_ATTN attention backend.
 
 Requirements:
     - GPU: sm89+ (H100, L40S, RTX 4090, Ada Lovelace, or newer)
       Falls back to a bf16 dequant path on older GPUs.
+    - sageattention: pip install sageattention (the script defaults to the
+      SAGE_ATTN attention backend)
     - TAEHV (optional): Follow install instructions at https://github.com/madebyollin/taehv
 
 Usage:
-    python fp8_wan2_1_1_3b.py              # FP8 per-tensor (default)
-    python fp8_wan2_1_1_3b.py --bf16       # BF16 baseline
-    python fp8_wan2_1_1_3b.py --granularity channel  # per-channel (higher accuracy but slower)
-    python fp8_wan2_1_1_3b.py --taehv-checkpoint /path/to/taew2_1.pth
+    python fp8_wan2_1_1_3b.py --taehv-checkpoint /path/to/taehv/taew2_1.pth                           # FP8 per-tensor
+    python fp8_wan2_1_1_3b.py --taehv-checkpoint /path/to/taehv/taew2_1.pth --bf16                    # BF16 baseline
+    python fp8_wan2_1_1_3b.py --taehv-checkpoint /path/to/taehv/taew2_1.pth --granularity channel     # per-channel
 """
 
 import argparse
 import os
-import sys
 import time
 
 import torch
@@ -27,28 +25,21 @@ import torch
 OUTPUT_PATH = "video_samples"
 
 
-def load_taehv(checkpoint_path, device="cuda", dtype=torch.float16):
-    repo_dir = os.path.dirname(checkpoint_path)
-    if repo_dir not in sys.path:
-        sys.path.insert(0, repo_dir)
-    from taehv import TAEHV
-    print(f"Loading TAEHV from {checkpoint_path}...")
-    model = TAEHV(checkpoint_path=checkpoint_path).to(device, dtype)
-    print("TAEHV loaded.")
-    return model
+class TaehvDecoder:
+    def __init__(self, checkpoint_path: str, device: str = "cuda",
+                 dtype: torch.dtype = torch.float16) -> None:
+        from taehv import TAEHV
+        self.device = device
+        self.dtype = dtype
+        print(f"Loading TAEHV from {checkpoint_path} ...")
+        self.model = TAEHV(checkpoint_path=checkpoint_path).to(device, dtype).eval()
 
-
-@torch.no_grad()  # type: ignore[misc]
-def decode_with_taehv(taehv_model, latents):
-    latents = latents.permute(0, 2, 1, 3, 4)
-    latents = latents.to(device=next(taehv_model.parameters()).device,
-                         dtype=next(taehv_model.parameters()).dtype)
-    decoded = taehv_model.decode_video(latents, parallel=False, show_progress_bar=False)
-    frames = []
-    for frame in decoded[0]:
-        frame_np = (frame.clamp(0, 1) * 255).byte().cpu().permute(1, 2, 0).numpy()
-        frames.append(frame_np)
-    return frames
+    @torch.no_grad()
+    def decode(self, latents: torch.Tensor):
+        latents = latents.permute(0, 2, 1, 3, 4).to(self.device, self.dtype)
+        decoded = self.model.decode_video(latents, parallel=True, show_progress_bar=False)
+        frames = (decoded[0].clamp(0, 1) * 255).to(torch.uint8)
+        return frames.permute(0, 2, 3, 1).cpu().numpy()
 
 
 def main():
@@ -69,7 +60,7 @@ def main():
     os.environ.setdefault("FASTVIDEO_ATTENTION_BACKEND", "SAGE_ATTN")
 
     from fastvideo import VideoGenerator
-    from fastvideo.layers.quantization import get_quantization_config
+    from fastvideo.configs.pipelines.base import PipelineConfig
 
     mode = "bf16" if args.bf16 else f"fp8_{args.granularity}"
     if not args.no_compile:
@@ -77,15 +68,17 @@ def main():
     use_taehv = args.taehv_checkpoint is not None
     print(f"Mode: {mode.upper()}" + ("  decoder=TAEHV" if use_taehv else "  decoder=VAE"))
 
-    taehv_model = load_taehv(args.taehv_checkpoint) if use_taehv else None
+    taehv = TaehvDecoder(args.taehv_checkpoint) if use_taehv else None
+
+    pipeline_config = PipelineConfig.from_pretrained(args.model)
+    pipeline_config.text_encoder_precisions = ("bf16",)
+    if not args.bf16:
+        from fastvideo.layers.quantization import get_quantization_config
+        pipeline_config.dit_config.quant_config = get_quantization_config("FP8")(granularity=args.granularity)
 
-    # transformer_quant needs a QuantizationConfig *instance* — the bare string
-    # is not resolved on the from_pretrained kwarg path.
-    extra = {} if args.bf16 else {
-        "transformer_quant": get_quantization_config("FP8")(granularity=args.granularity)
-    }
     generator = VideoGenerator.from_pretrained(
         args.model,
+        pipeline_config=pipeline_config,
         num_gpus=args.num_gpus,
         use_fsdp_inference=False,
         dit_cpu_offload=False,
@@ -94,9 +87,9 @@ def main():
         text_encoder_cpu_offload=False,
         pin_cpu_memory=False,
         enable_torch_compile=not args.no_compile,
+        enable_torch_compile_text_encoder=not args.no_compile,
         enable_torch_compile_vae=not args.no_compile and not use_taehv,
         output_type="latent" if use_taehv else "pil",
-        **extra,
     )
 
     prompt = (
@@ -105,33 +98,42 @@ def main():
         "natural light filtering through the petals. Mid-shot, warm and cheerful tones."
     )
 
-    n_warmup = 1 if not args.no_compile else 0
+    n_warmup = 2 if not args.no_compile else 0
     for _ in range(n_warmup):
-        generator.generate(request={"prompt": prompt, "sampling": {"num_inference_steps": 3, "guidance_scale": 1.0},
-                                    "output": {"save_video": False}})
+        warmup_result = generator.generate(request={"prompt": prompt, "sampling": {"num_inference_steps": 3, "guidance_scale": 1.0},
+                                                    "output": {"save_video": False}})
+        if use_taehv:
+            taehv.decode(warmup_result.samples)
 
     os.makedirs(OUTPUT_PATH, exist_ok=True)
-    start = time.time()
+    video_path = os.path.join(OUTPUT_PATH, f"raccoon_{mode}.mp4")
     if use_taehv:
+        import imageio
         result = generator.generate(request={
             "prompt": prompt,
             "sampling": {"num_inference_steps": args.infer_steps, "guidance_scale": 1.0},
             "output": {"save_video": False},
         })
-        import imageio
-        frames = decode_with_taehv(taehv_model, result.samples)
-        video_path = os.path.join(OUTPUT_PATH, f"raccoon_{mode}.mp4")
+        denoise_elapsed = result.generation_time
+        torch.cuda.synchronize()
+        t_decode = time.perf_counter()
+        frames = taehv.decode(result.samples)
+        torch.cuda.synchronize()
+        decode_elapsed = time.perf_counter() - t_decode
+        total = denoise_elapsed + decode_elapsed
         imageio.mimsave(video_path, frames, fps=16, format="mp4")
         print(f"Saved TAEHV-decoded video to: {video_path}")
+        print(f"[{mode.upper()}] {args.infer_steps} steps in {total:.3f}s "
+              f"(denoise {denoise_elapsed:.3f}s + decode {decode_elapsed:.3f}s)")
     else:
-        generator.generate(request={
+        result = generator.generate(request={
             "prompt": prompt,
             "sampling": {"num_inference_steps": args.infer_steps, "guidance_scale": 1.0},
-            "output": {"save_video": True, "output_path": os.path.join(OUTPUT_PATH, f"raccoon_{mode}.mp4")},
+            "output": {"save_video": True, "output_path": video_path},
         })
-    elapsed = time.time() - start
-    print(f"[{mode.upper()}] {args.infer_steps} steps in {elapsed:.2f}s "
-          f"({args.infer_steps / elapsed:.2f} it/s)")
+        elapsed = result.generation_time
+        print(f"[{mode.upper()}] {args.infer_steps} steps in {elapsed:.3f}s "
+              f"({args.infer_steps / elapsed:.2f} it/s)")
 
     generator.shutdown()
 

examples/inference/optimizations/nvfp4_qat_wan2_1_1_3b.py

@@ -1,69 +1,100 @@
-"""NVFP4 + Attn-QAT (modified SageAttention3) inference on Blackwell.
+"""NVFP4 QAD inference example.
 
-Runs Wan2.1-T2V-1.3B fully in 4-bit: NVFP4 linear layers (activations
-quantized on the fly) together with the modified SageAttention3 FP4 attention
-backend (``ATTN_QAT_INFER``). This is the inference half of the
-Quantization-Aware Distillation (QAD) recipe.
+Runs FastWan-QAD-1.3B (a distilled Wan2.1-T2V-1.3B-Diffusers checkpoint) with
+NVFP4QATConfig quantization. Uses ATTN_QAT_INFER attention backend.
 
 Requirements:
-    - RTX 5090 / consumer Blackwell (sm_120a). The attn_qat_infer kernel hard
-      gates on sm_120; on other GPUs it falls back to Flash Attention.
-    - The attn_qat_infer kernel built into fastvideo-kernel (see #1455) and
-      flashinfer for the NVFP4 linear matmuls.
+    - GPU (default NVFP4 mode): RTX 5090-class Blackwell (sm_120a). The FP4
+      linear path uses flashinfer's sm_120a cubins, which do not run on
+      sm_100 (B200/B300), and the attn_qat_infer kernel hard-gates on sm_120.
+    - GPU (--bf16 baseline): any supported GPU (the ATTN_QAT_INFER backend
+      falls back to Flash Attention off sm_120).
+    - TAEHV (optional): Follow install instructions at https://github.com/madebyollin/taehv
 
 Usage:
-    python nvfp4_qat_wan2_1_1_3b.py                 # NVFP4 linear + Attn-QAT attn
-    python nvfp4_qat_wan2_1_1_3b.py --bf16          # BF16 baseline
+    python nvfp4_qat_wan2_1_1_3b.py --taehv-checkpoint /path/to/taehv/taew2_1.pth           # NVFP4 QAD
+    python nvfp4_qat_wan2_1_1_3b.py --taehv-checkpoint /path/to/taehv/taew2_1.pth --bf16    # BF16 baseline
 """
 
 import argparse
 import os
 import time
 
+import torch
+
 OUTPUT_PATH = "video_samples"
 
 
+class TaehvDecoder:
+    def __init__(self, checkpoint_path: str, device: str = "cuda",
+                 dtype: torch.dtype = torch.float16) -> None:
+        from taehv import TAEHV
+        self.device = device
+        self.dtype = dtype
+        print(f"Loading TAEHV from {checkpoint_path} ...")
+        self.model = TAEHV(checkpoint_path=checkpoint_path).to(device, dtype).eval()
+
+    @torch.no_grad()
+    def decode(self, latents: torch.Tensor):
+        latents = latents.permute(0, 2, 1, 3, 4).to(self.device, self.dtype)
+        decoded = self.model.decode_video(latents, parallel=True, show_progress_bar=False)
+        frames = (decoded[0].clamp(0, 1) * 255).to(torch.uint8)
+        return frames.permute(0, 2, 3, 1).cpu().numpy()
+
+
 def main():
-    parser = argparse.ArgumentParser(description="NVFP4 + Attn-QAT video generation")
+    parser = argparse.ArgumentParser(description="NVFP4 QAD video generation benchmark")
     parser.add_argument("--bf16", action="store_true",
-                        help="BF16 baseline (no NVFP4 linear, default attention)")
-    parser.add_argument("--model", default="Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+                        help="BF16 baseline (no NVFP4 quantization)")
+    parser.add_argument("--taehv-checkpoint", default=None, metavar="PATH",
+                        help="Path to taew2_1.pth; enables TAEHV tiny autoencoder decoding")
+    parser.add_argument("--model", default="FastVideo/FastWan-QAD-1.3B",
                         help="Model path or HuggingFace ID")
-    parser.add_argument("--quant-method", default="nvfp4_qat", choices=["nvfp4_qat", "NVFP4"],
-                        help="Linear quantization config. Wan-2.1 uses nvfp4_qat (matches its "
-                             "to_q/k/v/out + ffn layers); NVFP4 is LTX2-specific and will NOT "
-                             "quantize Wan.")
-    parser.add_argument("--compile", action="store_true", help="Enable torch.compile for the DiT")
+    parser.add_argument("--no-compile", action="store_true", help="Disable torch.compile for the DiT")
     parser.add_argument("--num_gpus", type=int, default=1)
-    parser.add_argument("--infer_steps", type=int, default=50)
+    parser.add_argument("--infer_steps", type=int, default=3)
     args = parser.parse_args()
 
-    # The attention backend is selected via env var before the engine starts.
-    # ATTN_QAT_INFER -> AttnQatInferBackend (modified SageAttention3 FP4).
+    from _qad_common import flashinfer_arch_list, require_fp4_capable_gpu
+
     if not args.bf16:
-        os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "ATTN_QAT_INFER"
+        require_fp4_capable_gpu()
+
+    os.environ.setdefault("FASTVIDEO_ATTENTION_BACKEND", "ATTN_QAT_INFER")
+    os.environ["FASTVIDEO_DISABLE_ATTENTION_COMPILE"] = "0"
+    os.environ.setdefault("FLASHINFER_CUDA_ARCH_LIST", flashinfer_arch_list())
 
-    # Import after the env var so the platform picks up the selection.
     from fastvideo import VideoGenerator
-    from fastvideo.layers.quantization import get_quantization_config
+    from fastvideo.configs.pipelines.base import PipelineConfig
 
-    mode = "bf16" if args.bf16 else args.quant_method
-    if args.compile:
+    mode = "bf16" if args.bf16 else "nvfp4_qad"
+    if not args.no_compile:
         mode += "_compile"
-    print(f"Mode: {mode.upper()}")
+    use_taehv = args.taehv_checkpoint is not None
+    print(f"Mode: {mode.upper()}" + ("  decoder=TAEHV" if use_taehv else "  decoder=VAE"))
+
+    taehv = TaehvDecoder(args.taehv_checkpoint) if use_taehv else None
+
+    pipeline_config = PipelineConfig.from_pretrained(args.model)
+    pipeline_config.text_encoder_precisions = ("bf16",)
+    if not args.bf16:
+        from fastvideo.layers.quantization.nvfp4_qat_config import NVFP4QATConfig
+        pipeline_config.dit_config.quant_config = NVFP4QATConfig()
 
-    # transformer_quant needs a QuantizationConfig *instance* — the bare string
-    # is not resolved on the from_pretrained kwarg path.
-    extra = {} if args.bf16 else {"transformer_quant": get_quantization_config(args.quant_method)()}
     generator = VideoGenerator.from_pretrained(
         args.model,
+        pipeline_config=pipeline_config,
         num_gpus=args.num_gpus,
-        use_fsdp_inference=args.bf16,
+        use_fsdp_inference=False,
         dit_cpu_offload=False,
-        vae_cpu_offload=True,
-        text_encoder_cpu_offload=True,
-        enable_torch_compile=args.compile,
-        **extra,
+        dit_layerwise_offload=False,
+        vae_cpu_offload=use_taehv,
+        text_encoder_cpu_offload=False,
+        pin_cpu_memory=False,
+        enable_torch_compile=not args.no_compile,
+        enable_torch_compile_text_encoder=not args.no_compile,
+        enable_torch_compile_vae=not args.no_compile and not use_taehv,
+        output_type="latent" if use_taehv else "pil",
     )
 
     prompt = (
@@ -72,21 +103,42 @@ def main():
         "natural light filtering through the petals. Mid-shot, warm and cheerful tones."
     )
 
-    n_warmup = 2 if args.compile else 1
+    n_warmup = 2 if not args.no_compile else 0
     for _ in range(n_warmup):
-        generator.generate(request={"prompt": prompt, "sampling": {"num_inference_steps": 2},
-                                    "output": {"save_video": False}})
+        warmup_result = generator.generate(request={"prompt": prompt, "sampling": {"num_inference_steps": 3, "guidance_scale": 1.0},
+                                                    "output": {"save_video": False}})
+        if use_taehv:
+            taehv.decode(warmup_result.samples)
 
     os.makedirs(OUTPUT_PATH, exist_ok=True)
-    start = time.time()
-    generator.generate(request={
-        "prompt": prompt,
-        "sampling": {"num_inference_steps": args.infer_steps},
-        "output": {"save_video": True, "output_path": os.path.join(OUTPUT_PATH, f"raccoon_{mode}.mp4")},
-    })
-    elapsed = time.time() - start
-    print(f"[{mode.upper()}] {args.infer_steps} steps in {elapsed:.2f}s "
-          f"({args.infer_steps / elapsed:.2f} it/s)")
+    video_path = os.path.join(OUTPUT_PATH, f"raccoon_{mode}.mp4")
+    if use_taehv:
+        import imageio
+        result = generator.generate(request={
+            "prompt": prompt,
+            "sampling": {"num_inference_steps": args.infer_steps, "guidance_scale": 1.0},
+            "output": {"save_video": False},
+        })
+        denoise_elapsed = result.generation_time
+        torch.cuda.synchronize()
+        t_decode = time.perf_counter()
+        frames = taehv.decode(result.samples)
+        torch.cuda.synchronize()
+        decode_elapsed = time.perf_counter() - t_decode
+        total = denoise_elapsed + decode_elapsed
+        imageio.mimsave(video_path, frames, fps=16, format="mp4")
+        print(f"Saved TAEHV-decoded video to: {video_path}")
+        print(f"[{mode.upper()}] {args.infer_steps} steps in {total:.3f}s "
+              f"(denoise {denoise_elapsed:.3f}s + decode {decode_elapsed:.3f}s)")
+    else:
+        result = generator.generate(request={
+            "prompt": prompt,
+            "sampling": {"num_inference_steps": args.infer_steps, "guidance_scale": 1.0},
+            "output": {"save_video": True, "output_path": video_path},
+        })
+        elapsed = result.generation_time
+        print(f"[{mode.upper()}] {args.infer_steps} steps in {elapsed:.3f}s "
+              f"({args.infer_steps / elapsed:.2f} it/s)")
 
     generator.shutdown()
 

examples/inference/optimizations/nvfp4_sa2_wan2_1_1_3b.py

@@ -0,0 +1,146 @@
+"""NVFP4 QAD inference example with SageAttention 2 backend.
+
+Runs FastWan-QAD-1.3B-SA2 (a distilled Wan2.1-T2V-1.3B-Diffusers checkpoint) with
+NVFP4QATConfig quantization. Uses the SAGE_ATTN attention backend.
+
+Requirements:
+    - GPU (default NVFP4 mode): RTX 5090-class Blackwell (sm_120a). The NVFP4
+      linear path uses flashinfer's cutlass FP4 gemm, which has no fallback.
+    - GPU (--bf16 baseline): sm89+ (H100, L40S, RTX 4090, Ada Lovelace, or newer)
+    - sageattention: pip install sageattention
+    - TAEHV (optional): Follow install instructions at https://github.com/madebyollin/taehv
+
+Usage:
+    python nvfp4_sa2_wan2_1_1_3b.py --taehv-checkpoint /path/to/taehv/taew2_1.pth           # NVFP4 + SageAttn2
+    python nvfp4_sa2_wan2_1_1_3b.py --taehv-checkpoint /path/to/taehv/taew2_1.pth --bf16    # BF16 baseline
+"""
+
+import argparse
+import os
+import time
+
+import torch
+
+OUTPUT_PATH = "video_samples"
+
+
+class TaehvDecoder:
+    def __init__(self, checkpoint_path: str, device: str = "cuda",
+                 dtype: torch.dtype = torch.float16) -> None:
+        from taehv import TAEHV
+        self.device = device
+        self.dtype = dtype
+        print(f"Loading TAEHV from {checkpoint_path} ...")
+        self.model = TAEHV(checkpoint_path=checkpoint_path).to(device, dtype).eval()
+
+    @torch.no_grad()
+    def decode(self, latents: torch.Tensor):
+        latents = latents.permute(0, 2, 1, 3, 4).to(self.device, self.dtype)
+        decoded = self.model.decode_video(latents, parallel=True, show_progress_bar=False)
+        frames = (decoded[0].clamp(0, 1) * 255).to(torch.uint8)
+        return frames.permute(0, 2, 3, 1).cpu().numpy()
+
+
+def main():
+    parser = argparse.ArgumentParser(description="NVFP4 QAD + SageAttention2 video generation benchmark")
+    parser.add_argument("--bf16", action="store_true",
+                        help="BF16 baseline (no NVFP4 quantization)")
+    parser.add_argument("--taehv-checkpoint", default=None, metavar="PATH",
+                        help="Path to taew2_1.pth; enables TAEHV tiny autoencoder decoding")
+    parser.add_argument("--model", default="FastVideo/FastWan-QAD-1.3B-SA2",
+                        help="Model path or HuggingFace ID")
+    parser.add_argument("--no-compile", action="store_true", help="Disable torch.compile for the DiT")
+    parser.add_argument("--num_gpus", type=int, default=1)
+    parser.add_argument("--infer_steps", type=int, default=3)
+    args = parser.parse_args()
+
+    from _qad_common import flashinfer_arch_list, require_fp4_capable_gpu
+
+    if not args.bf16:
+        require_fp4_capable_gpu()
+
+    os.environ.setdefault("FASTVIDEO_ATTENTION_BACKEND", "SAGE_ATTN")
+    os.environ["FASTVIDEO_DISABLE_ATTENTION_COMPILE"] = "0"
+    os.environ.setdefault("FLASHINFER_CUDA_ARCH_LIST", flashinfer_arch_list())
+
+    from fastvideo import VideoGenerator
+    from fastvideo.configs.pipelines.base import PipelineConfig
+
+    mode = "bf16" if args.bf16 else "nvfp4_sa2"
+    if not args.no_compile:
+        mode += "_compile"
+    use_taehv = args.taehv_checkpoint is not None
+    print(f"Mode: {mode.upper()}" + ("  decoder=TAEHV" if use_taehv else "  decoder=VAE"))
+
+    taehv = TaehvDecoder(args.taehv_checkpoint) if use_taehv else None
+
+    pipeline_config = PipelineConfig.from_pretrained(args.model)
+    pipeline_config.text_encoder_precisions = ("bf16",)
+    if not args.bf16:
+        from fastvideo.layers.quantization.nvfp4_qat_config import NVFP4QATConfig
+        pipeline_config.dit_config.quant_config = NVFP4QATConfig()
+
+    generator = VideoGenerator.from_pretrained(
+        args.model,
+        pipeline_config=pipeline_config,
+        num_gpus=args.num_gpus,
+        use_fsdp_inference=False,
+        dit_cpu_offload=False,
+        dit_layerwise_offload=False,
+        vae_cpu_offload=use_taehv,
+        text_encoder_cpu_offload=False,
+        pin_cpu_memory=False,
+        enable_torch_compile=not args.no_compile,
+        enable_torch_compile_text_encoder=not args.no_compile,
+        enable_torch_compile_vae=not args.no_compile and not use_taehv,
+        output_type="latent" if use_taehv else "pil",
+    )
+
+    prompt = (
+        "A curious raccoon peers through a vibrant field of yellow sunflowers, its eyes "
+        "wide with interest. The playful yet serene atmosphere is complemented by soft "
+        "natural light filtering through the petals. Mid-shot, warm and cheerful tones."
+    )
+
+    n_warmup = 2 if not args.no_compile else 0
+    for _ in range(n_warmup):
+        warmup_result = generator.generate(request={"prompt": prompt, "sampling": {"num_inference_steps": 3, "guidance_scale": 1.0},
+                                                    "output": {"save_video": False}})
+        if use_taehv:
+            taehv.decode(warmup_result.samples)
+
+    os.makedirs(OUTPUT_PATH, exist_ok=True)
+    video_path = os.path.join(OUTPUT_PATH, f"raccoon_{mode}.mp4")
+    if use_taehv:
+        import imageio
+        result = generator.generate(request={
+            "prompt": prompt,
+            "sampling": {"num_inference_steps": args.infer_steps, "guidance_scale": 1.0},
+            "output": {"save_video": False},
+        })
+        denoise_elapsed = result.generation_time
+        torch.cuda.synchronize()
+        t_decode = time.perf_counter()
+        frames = taehv.decode(result.samples)
+        torch.cuda.synchronize()
+        decode_elapsed = time.perf_counter() - t_decode
+        total = denoise_elapsed + decode_elapsed
+        imageio.mimsave(video_path, frames, fps=16, format="mp4")
+        print(f"Saved TAEHV-decoded video to: {video_path}")
+        print(f"[{mode.upper()}] {args.infer_steps} steps in {total:.3f}s "
+              f"(denoise {denoise_elapsed:.3f}s + decode {decode_elapsed:.3f}s)")
+    else:
+        result = generator.generate(request={
+            "prompt": prompt,
+            "sampling": {"num_inference_steps": args.infer_steps, "guidance_scale": 1.0},
+            "output": {"save_video": True, "output_path": video_path},
+        })
+        elapsed = result.generation_time
+        print(f"[{mode.upper()}] {args.infer_steps} steps in {elapsed:.3f}s "
+              f"({args.infer_steps / elapsed:.2f} it/s)")
+
+    generator.shutdown()
+
+
+if __name__ == "__main__":
+    main()

examples/train/configs/consistency_model/wan/causal_cd_t2v.yaml

@@ -0,0 +1,78 @@
+# Causal Consistency Distillation: Wan 2.1 T2V 1.3B Causal
+#
+# ODE-data-free distillation. A frozen teacher takes a single CFG Euler step;
+# the student matches an EMA copy of itself at the next timestep, all under
+# clean-history teacher forcing.
+#
+# All three roles initialize from the SAME checkpoint (the teacher-forcing
+# AR-diffusion model). Point init_from at that checkpoint for a real run.
+
+models:
+  student:
+    _target_: fastvideo.train.models.wan.WanCausalModel
+    init_from: Wan-AI/Wan2.1-T2V-1.3B-Diffusers
+    trainable: true
+  teacher:
+    _target_: fastvideo.train.models.wan.WanCausalModel
+    init_from: Wan-AI/Wan2.1-T2V-1.3B-Diffusers
+    trainable: false
+  ema:
+    _target_: fastvideo.train.models.wan.WanCausalModel
+    init_from: Wan-AI/Wan2.1-T2V-1.3B-Diffusers
+    trainable: false
+
+method:
+  _target_: fastvideo.train.methods.consistency_model.causal_cd.CausalConsistencyDistillationMethod
+  discrete_cd_N: 48
+  guidance_scale: 3.0
+  ema_decay: 0.99
+  ema_start_step: 200
+
+training:
+  distributed:
+    num_gpus: 8
+    sp_size: 1
+    tp_size: 1
+    hsdp_replicate_dim: 1
+    hsdp_shard_dim: 8
+
+  data:
+    data_path: data/Wan-Syn_77x448x832_600k
+    dataloader_num_workers: 4
+    train_batch_size: 1
+    training_cfg_rate: 0.0
+    seed: 1000
+    num_latent_t: 18
+    num_height: 448
+    num_width: 832
+    num_frames: 69
+
+  optimizer:
+    learning_rate: 2.0e-6
+    betas: [0.9, 0.999]
+    weight_decay: 0.01
+    lr_scheduler: constant
+    lr_warmup_steps: 0
+
+  loop:
+    max_train_steps: 3000
+    gradient_accumulation_steps: 1
+
+  checkpoint:
+    output_dir: outputs/wan2.1_causal_cd
+    training_state_checkpointing_steps: 1000
+    checkpoints_total_limit: 3
+
+  tracker:
+    project_name: distillation_wan_r
+    run_name: wan2.1_causal_cd_shift5
+
+  model:
+    enable_gradient_checkpointing_type: full
+
+callbacks:
+  grad_clip:
+    max_grad_norm: 1.0
+
+pipeline:
+  flow_shift: 5

examples/train/configs/distribution_matching/wan/anyflow_onpolicy_t2v.yaml

@@ -0,0 +1,105 @@
+# AnyFlow on-policy DMD — Wan 2.1 T2V 1.3B.
+#
+# Stage 2 of the AnyFlow two-stage recipe. Continues from the pretrain
+# checkpoint; refines the student via DMD2 with a multi-step Euler-flow
+# rollout from pure noise. Teacher provides the real score, critic
+# learns the fake score; both inherited from DMD2Method.
+#
+# Replace <PATH_TO_PRETRAIN_CKPT> with the output of the pretrain stage,
+# or with the NVIDIA-released checkpoint
+# nvidia/AnyFlow-Wan2.1-T2V-1.3B-Diffusers to bootstrap directly from
+# the paper weights (the delta_embedder rename is handled by the
+# param_names_mapping in WanVideoArchConfig).
+
+models:
+  student:
+    _target_: fastvideo.train.models.wan.WanModel
+    init_from: <PATH_TO_PRETRAIN_CKPT>
+    trainable: true
+  teacher:
+    _target_: fastvideo.train.models.wan.WanModel
+    init_from: Wan-AI/Wan2.1-T2V-14B-Diffusers
+    trainable: false
+    disable_custom_init_weights: true
+  critic:
+    _target_: fastvideo.train.models.wan.WanModel
+    init_from: Wan-AI/Wan2.1-T2V-1.3B-Diffusers
+    trainable: true
+    disable_custom_init_weights: true
+
+method:
+  _target_: fastvideo.train.methods.distribution_matching.anyflow.AnyFlowMethod
+  rollout_mode: simulate
+  generator_update_interval: 5
+  real_score_guidance_scale: 3.0
+  dmd_denoising_steps: [999, 937, 833, 624]
+  warp_denoising_step: false
+
+  # AnyFlow rollout knobs.
+  student_sample_steps: 4
+  use_mean_velocity: true
+  t_list_override: [999.0, 937.0, 833.0, 624.0, 0.0]
+  dmd_score_r_value: 0.0  # DMD scoring conditioning is at r=0 (consistency target).
+
+  # Critic optimizer (DMD2 inherited).
+  fake_score_learning_rate: 8.0e-6
+  fake_score_betas: [0.0, 0.999]
+  fake_score_lr_scheduler: constant
+
+  attn_kind: vsa
+
+training:
+  distributed:
+    num_gpus: 8
+    sp_size: 1
+    tp_size: 1
+    hsdp_replicate_dim: 1
+    hsdp_shard_dim: 8
+
+  data:
+    data_path: data/preprocessed
+    dataloader_num_workers: 4
+    train_batch_size: 1
+    training_cfg_rate: 0.0
+    seed: 1000
+    num_latent_t: 21
+    num_height: 480
+    num_width: 832
+    num_frames: 81
+
+  optimizer:
+    learning_rate: 2.0e-6
+    betas: [0.0, 0.999]
+    weight_decay: 0.01
+    lr_scheduler: constant
+    lr_warmup_steps: 0
+
+  loop:
+    max_train_steps: 4000
+    gradient_accumulation_steps: 1
+
+  checkpoint:
+    output_dir: outputs/wan2.1_anyflow_onpolicy
+    training_state_checkpointing_steps: 500
+    checkpoints_total_limit: 3
+    resume_from_checkpoint: latest
+
+  tracker:
+    project_name: anyflow-wan
+    run_name: wan2.1_t2v_anyflow_onpolicy
+
+  model:
+    enable_gradient_checkpointing_type: full
+
+callbacks:
+  grad_clip:
+    _target_: fastvideo.train.callbacks.grad_clip.GradNormClipCallback
+    max_grad_norm: 1.0
+
+pipeline:
+  flow_shift: 5.0
+  dit_config:
+    r_embedder: true
+    r_embedder_fusion: gated
+    r_embedder_gate_value: 0.25
+    r_embedder_deltatime_type: r

examples/train/configs/distribution_matching/wan/anyflow_pretrain_t2v.yaml

@@ -0,0 +1,83 @@
+# AnyFlow pretrain (flow-map central-difference) — Wan 2.1 T2V 1.3B.
+#
+# Stage 1 of the AnyFlow two-stage recipe. Trains the dual-timestep
+# u_θ(x_t, t, r) on the central-difference target so the same checkpoint
+# can be sampled at arbitrary NFE in the on-policy stage.
+#
+# Initialize from base Wan 2.1 T2V 1.3B. No teacher or critic at this
+# stage; AnyFlowPretrainMethod owns a single student + one optimizer.
+
+models:
+  student:
+    _target_: fastvideo.train.models.wan.WanModel
+    init_from: Wan-AI/Wan2.1-T2V-1.3B-Diffusers
+    trainable: true
+
+method:
+  _target_: fastvideo.train.methods.distribution_matching.anyflow_pretrain.AnyFlowPretrainMethod
+  diffusion_ratio: 0.5
+  consistency_ratio: 0.25
+  epsilon: 5            # finite-difference step in absolute train-timestep units
+  weight_type: beta08   # per-timestep loss weight = t * sqrt(1 - t), renormalized
+  fuse_guidance_scale: 3.0
+  # shift is taken from pipeline.flow_shift below.
+
+training:
+  distributed:
+    num_gpus: 8
+    sp_size: 1
+    tp_size: 1
+    hsdp_replicate_dim: 1
+    hsdp_shard_dim: 8
+
+  data:
+    data_path: data/preprocessed
+    dataloader_num_workers: 4
+    train_batch_size: 4
+    training_cfg_rate: 0.0
+    seed: 1000
+    num_latent_t: 21
+    num_height: 480
+    num_width: 832
+    num_frames: 81
+
+  optimizer:
+    learning_rate: 5.0e-5
+    betas: [0.9, 0.999]
+    weight_decay: 0.0
+    lr_scheduler: constant
+    lr_warmup_steps: 0
+
+  loop:
+    max_train_steps: 6000
+    gradient_accumulation_steps: 1
+
+  checkpoint:
+    output_dir: outputs/wan2.1_anyflow_pretrain
+    training_state_checkpointing_steps: 500
+    checkpoints_total_limit: 3
+    resume_from_checkpoint: latest
+
+  tracker:
+    project_name: anyflow-wan
+    run_name: wan2.1_t2v_anyflow_pretrain
+
+  model:
+    enable_gradient_checkpointing_type: full
+
+callbacks:
+  grad_clip:
+    _target_: fastvideo.train.callbacks.grad_clip.GradNormClipCallback
+    max_grad_norm: 1.0
+
+pipeline:
+  flow_shift: 5.0
+  dit_config:
+    # Enable AnyFlow dual-timestep conditioning. The student loads from
+    # base Wan 2.1 — its checkpoint has no delta_embedder weights, so they
+    # get initialized identically to time_embedder via deep-copy in
+    # WanTimeTextImageEmbedding.__init__.
+    r_embedder: true
+    r_embedder_fusion: gated
+    r_embedder_gate_value: 0.25
+    r_embedder_deltatime_type: r

examples/train/configs/distribution_matching/wan/self_forcing_causal_t2v.yaml

@@ -100,9 +100,6 @@ callbacks:
     sampling_steps: [4]
     sampling_timesteps: [1000, 750, 500, 250]
     num_frames: 81
-    # Validation/inference uses standard CFG in both clean and Self-Forcing,
-    # so this directly matches Self-Forcing guidance_scale=3.0.
-    guidance_scale: 3.0
 
 pipeline:
   flow_shift: 5

examples/train/configs/fine_tuning/wan/dfsft_causal_t2v_framewise.yaml

@@ -0,0 +1,73 @@
+# DFSFT (Diffusion-Forcing SFT), frame-wise: Wan 2.1 T2V 1.3B Causal
+#
+# - Student: trainable causal Wan model with a block size of 1 frame
+# - Training: each frame gets its own independent noise level (frame-wise
+#   diffusion forcing), versus the chunk-wise variant that shares one noise
+#   level across num_frames_per_block frames.
+
+models:
+  student:
+    _target_: fastvideo.train.models.wan.WanCausalModel
+    init_from: Wan-AI/Wan2.1-T2V-1.3B-Diffusers
+    trainable: true
+    num_frames_per_block: 1
+
+method:
+  _target_: fastvideo.train.methods.fine_tuning.dfsft.DiffusionForcingSFTMethod
+  chunk_size: 1
+
+training:
+  distributed:
+    num_gpus: 8
+    sp_size: 1
+    tp_size: 1
+    hsdp_replicate_dim: 1
+    hsdp_shard_dim: 8
+
+  data:
+    data_path: data/Wan-Syn_77x448x832_600k
+    dataloader_num_workers: 4
+    train_batch_size: 1
+    training_cfg_rate: 0.0
+    seed: 1000
+    num_latent_t: 18
+    num_height: 448
+    num_width: 832
+    num_frames: 69
+
+  optimizer:
+    learning_rate: 2.0e-6
+    betas: [0.9, 0.999]
+    weight_decay: 0.01
+    lr_scheduler: constant
+    lr_warmup_steps: 0
+
+  loop:
+    max_train_steps: 4000
+    gradient_accumulation_steps: 1
+
+  checkpoint:
+    output_dir: outputs/wan2.1_causal_dfsft_framewise
+    training_state_checkpointing_steps: 1000
+    checkpoints_total_limit: 3
+
+  tracker:
+    project_name: distillation_wan_r
+    run_name: wan2.1_causal_dfsft_framewise_shift5_gauss_weight
+
+  model:
+    enable_gradient_checkpointing_type: full
+
+callbacks:
+  grad_clip:
+    max_grad_norm: 1.0
+  validation:
+    pipeline_target: fastvideo.pipelines.basic.wan.wan_causal_pipeline.WanCausalPipeline
+    dataset_file: examples/training/finetune/Wan2.1-VSA/Wan-Syn-Data/validation_4.json
+    every_steps: 50
+    sampling_steps: [40]
+    guidance_scale: 6.0
+    num_frames: 69
+
+pipeline:
+  flow_shift: 5

examples/train/configs/fine_tuning/wan/tfsft_causal_t2v.yaml

@@ -0,0 +1,72 @@
+# TFSFT (Teacher-Forcing SFT): Wan 2.1 T2V 1.3B Causal
+#
+# - Student: trainable causal Wan model
+# - Training: inhomogeneous timesteps per chunk, but the causal transformer
+#   denoises the current block while attending to *clean* history (clean_x),
+#   not its own noisy rollout.
+
+models:
+  student:
+    _target_: fastvideo.train.models.wan.WanCausalModel
+    init_from: Wan-AI/Wan2.1-T2V-1.3B-Diffusers
+    trainable: true
+
+method:
+  _target_: fastvideo.train.methods.fine_tuning.tfsft.TeacherForcingSFTMethod
+  chunk_size: 3
+
+training:
+  distributed:
+    num_gpus: 8
+    sp_size: 1
+    tp_size: 1
+    hsdp_replicate_dim: 1
+    hsdp_shard_dim: 8
+
+  data:
+    data_path: data/Wan-Syn_77x448x832_600k
+    dataloader_num_workers: 4
+    train_batch_size: 1
+    training_cfg_rate: 0.0
+    seed: 1000
+    num_latent_t: 18
+    num_height: 448
+    num_width: 832
+    num_frames: 69
+
+  optimizer:
+    learning_rate: 2.0e-6
+    betas: [0.9, 0.999]
+    weight_decay: 0.01
+    lr_scheduler: constant
+    lr_warmup_steps: 0
+
+  loop:
+    max_train_steps: 4000
+    gradient_accumulation_steps: 1
+
+  checkpoint:
+    output_dir: outputs/wan2.1_causal_tfsft
+    training_state_checkpointing_steps: 1000
+    checkpoints_total_limit: 3
+
+  tracker:
+    project_name: distillation_wan_r
+    run_name: wan2.1_causal_tfsft_shift5_gauss_weight
+
+  model:
+    enable_gradient_checkpointing_type: full
+
+callbacks:
+  grad_clip:
+    max_grad_norm: 1.0
+  validation:
+    pipeline_target: fastvideo.pipelines.basic.wan.wan_causal_pipeline.WanCausalPipeline
+    dataset_file: examples/training/finetune/Wan2.1-VSA/Wan-Syn-Data/validation_4.json
+    every_steps: 50
+    sampling_steps: [40]
+    guidance_scale: 6.0
+    num_frames: 69
+
+pipeline:
+  flow_shift: 5

examples/train/scenario/worldmodel/README.md

@@ -1,32 +1,119 @@
 # World-Model: Matrix-Game 2.0 I2V
 
-Three training scenarios for the Matrix-Game 2.0 I2V world model on the
-new YAML-driven trainer (`fastvideo/train/entrypoint/train.py`).
+Training scenarios for the Matrix-Game 2.0 I2V world model on Solaris (Minecraft)
+data and Zelda data, using the new YAML-driven trainer
+(`fastvideo/train/entrypoint/train.py`).
+
+## Solaris Configs
 
 | Config | Method | Student | Notes |
 |---|---|---|---|
-| `finetune_i2v.yaml` | `FineTuneMethod` | `MatrixGame2Model` (bidirectional) | Multi-step SFT from `mg_bidirectional_Solaris`. |
-| `dfsft_causal_i2v.yaml` | `DiffusionForcingSFTMethod` | `MatrixGame2CausalModel` | Diffusion-Forcing SFT with chunkwise timesteps. |
-| `self_forcing_causal_i2v.yaml` | `SelfForcingMethod` | `MatrixGame2CausalModel` | DMD/Self-Forcing distillation; teacher = bidirectional, critic = bidirectional. |
+| `solaris/finetune_i2v.yaml` | `FineTuneMethod` | `MatrixGame2Model` (bidirectional) | Multi-step SFT from `mg_bidirectional_Solaris`. |
+| `solaris/dfsft_causal_i2v.yaml` | `DiffusionForcingSFTMethod` | `MatrixGame2CausalModel` | Diffusion-Forcing SFT with chunkwise timesteps. |
+| `solaris/self_forcing_causal_i2v.yaml` | `SelfForcingMethod` | `MatrixGame2CausalModel` | Matrix-Game 2.0 DMD/Self-Forcing distillation; teacher = bidirectional, critic = bidirectional. |
+
+## Zelda Configs
+
+| Config | Method | Student | Notes |
+|---|---|---|---|
+| `zelda/finetune_i2v.yaml` | `FineTuneMethod` | `MatrixGame2Model` (bidirectional) | Zelda bidirectional I2V finetuning from `FastVideo/Matrix-Game-2.0-Base-Diffusers`. Uses 33-frame clips and Zelda validation with action overlays. |
+| `zelda/dfsft_causal_i2v.yaml` | `DiffusionForcingSFTMethod` | `MatrixGame2CausalModel` | Zelda causal Diffusion-Forcing SFT from `mignonjia/mg_bidirectional_zelda`. Uses the same Zelda data, resolution, optimizer, and validation defaults as the Zelda finetune config. |
+| `zelda/self_forcing_causal_i2v.yaml` | `SelfForcingMethod` | `MatrixGame2CausalModel` | Zelda DMD/Self-Forcing distillation; student init = `mignonjia/mg_causal_zelda`, teacher = bidirectional (`mignonjia/mg_bidirectional_zelda`), critic = bidirectional. |
+| `zelda/streaming_long_tuning_causal_i2v.yaml` | `StreamingLongTuningMethod` | `MatrixGame2CausalModel` | LongLive-style streaming long tuning from the 1k-step Zelda self-forcing checkpoint. |
+
+Zelda world-model distillation is a two-run workflow: first run
+`zelda/self_forcing_causal_i2v.yaml` to train or load the 1k-step
+self-forcing checkpoint (`mignonjia/mg_sf_distilled_zelda_1k_steps`), then run
+`zelda/streaming_long_tuning_causal_i2v.yaml` for the 3k-step streaming
+long-tuning stage. The long-tuning YAML starts from that 1k-step checkpoint; it
+does not run the short self-forcing stage inside the same config.
+
+## Zelda Training Data
+
+The Zelda training configs use `data/zeldam2-clean` as a suggested local path.
+Download the dataset from Hugging Face before running those configs:
+
+```bash
+python scripts/huggingface/download_hf.py \
+    --repo_id mignonjia/zeldam2-clean \
+    --local_dir data/zeldam2-clean \
+    --repo_type dataset
+```
+
+You can store the dataset elsewhere; update `training.data.data_path` in the
+YAML to point at that location.
+
+## Multi3D Training Data
+
+`zelda/finetune_i2v.yaml` includes an optional, commented-out Multi3D entry.
+Enable it only when you want to mix Zelda with multi-game data from
+`data/multi3d_games`. You can store this dataset anywhere; before enabling it,
+update the matching commented `training.data.data_path` key in the YAML to the
+correct location.
+
+To mix datasets in a training YAML, set `training.data.data_path` to a
+path-to-repeat-count mapping. For example, `zelda/finetune_i2v.yaml` can use
+`data/zeldam2-clean: 1` and `# data/multi3d_games: 10`; uncommenting the
+Multi3D entry repeats the multi-game parquet list ten times before training
+samples are shuffled.
+
+## World Model Validation Data
+
+The Zelda validation configs expect a small public validation bundle under
+`data/zelda_validation_data`.
+
+Download it from Hugging Face before running the Zelda scenarios:
+
+```bash
+python scripts/huggingface/download_hf.py \
+    --repo_id mignonjia/zelda_validation_data \
+    --local_dir data/zelda_validation_data \
+    --repo_type dataset
+```
+
+The bundle contains `validation_zelda.json`, `images/`, and `actions/`.
+The Zelda configs point
+`callbacks.validation.dataset_file` at
+`data/zelda_validation_data/validation_zelda.json`.
 
 ## Usage
 
+### Solaris
+
 ```bash
 bash examples/train/run.sh \
-    examples/train/scenario/worldmodel/finetune_i2v.yaml
+    examples/train/scenario/worldmodel/solaris/finetune_i2v.yaml
 
 bash examples/train/run.sh \
-    examples/train/scenario/worldmodel/dfsft_causal_i2v.yaml
+    examples/train/scenario/worldmodel/solaris/dfsft_causal_i2v.yaml
 
 bash examples/train/run.sh \
-    examples/train/scenario/worldmodel/self_forcing_causal_i2v.yaml
+    examples/train/scenario/worldmodel/solaris/self_forcing_causal_i2v.yaml
+```
+
+### Zelda
+
+```bash
+# Finetuning / DFSFT
+bash examples/train/run.sh \
+    examples/train/scenario/worldmodel/zelda/finetune_i2v.yaml
+
+bash examples/train/run.sh \
+    examples/train/scenario/worldmodel/zelda/dfsft_causal_i2v.yaml
+
+# Distillation / long tuning
+bash examples/train/run.sh \
+    examples/train/scenario/worldmodel/zelda/self_forcing_causal_i2v.yaml
+
+bash examples/train/run.sh \
+    examples/train/scenario/worldmodel/zelda/streaming_long_tuning_causal_i2v.yaml
 ```
 
 Override any field on the command line:
 
 ```bash
 bash examples/train/run.sh \
-    examples/train/scenario/worldmodel/dfsft_causal_i2v.yaml \
+    examples/train/scenario/worldmodel/solaris/dfsft_causal_i2v.yaml \
     --training.distributed.num_gpus 8 \
     --training.optimizer.learning_rate 1e-5
 ```

examples/train/scenario/worldmodel/solaris/self_forcing_causal_i2v.yaml

@@ -97,4 +97,4 @@ callbacks:
     guidance_scale: 6.0
 
 pipeline:
-  flow_shift: 5
+  flow_shift: 5

examples/train/scenario/worldmodel/zelda/dfsft_causal_i2v.yaml

@@ -0,0 +1,94 @@
+# Diffusion-Forcing SFT: Zelda world model I2V Causal
+
+models:
+  student:
+    _target_: fastvideo.train.models.matrixgame2.matrixgame2_causal.MatrixGame2CausalModel
+    init_from: mignonjia/mg_bidirectional_zelda
+    trainable: true
+
+method:
+  _target_: fastvideo.train.methods.fine_tuning.dfsft.DiffusionForcingSFTMethod
+  chunk_size: 3
+
+training:
+  distributed:
+    num_gpus: 8
+    sp_size: 1
+    tp_size: 1
+    hsdp_replicate_dim: 1
+    hsdp_shard_dim: 8
+
+  data:
+    data_path:
+      data/zeldam2-clean: 1
+    dataloader_num_workers: 1
+    train_batch_size: 1
+    training_cfg_rate: 0.0
+    seed: 42
+    num_latent_t: 9
+    num_height: 480
+    num_width: 832
+    num_frames: 33
+
+  optimizer:
+    learning_rate: 2.0e-5
+    betas: [0.9, 0.95]
+    weight_decay: 0.0
+    lr_scheduler: constant
+    lr_warmup_steps: 0
+
+  loop:
+    max_train_steps: 60000
+    gradient_accumulation_steps: 1
+
+  checkpoint:
+    output_dir: outputs/matrixgame_finetune/checkpoints/zelda_causal_dfsft
+    training_state_checkpointing_steps: 5000
+    checkpoints_total_limit: 3
+
+  tracker:
+    entity: hapo-exp
+    project_name: mg_1.3b_zelda
+    run_name: zelda_causal_dfsft
+
+  model:
+    enable_gradient_checkpointing_type: full
+
+  dit_precision: fp32
+
+callbacks:
+  grad_clip:
+    max_grad_norm: 1.0
+  validation:
+    pipeline_target: fastvideo.pipelines.basic.matrixgame2.matrixgame2_causal_dmd_pipeline.MatrixGame2CausalDMDPipeline
+    dataset_file: data/zelda_validation_data/validation_zelda.json
+    every_steps: 200
+    sampling_steps: [40]
+    sampling_timesteps: [1000, 975, 950, 925, 900, 875, 850, 825, 800, 775,
+                         750, 725, 700, 675, 650, 625, 600, 575, 550, 525,
+                         500, 475, 450, 425, 400, 375, 350, 325, 300, 275,
+                         250, 225, 200, 175, 150, 125, 100, 75, 50, 25]
+    num_frames: 33
+    overlay_actions: true
+    guidance_scale: 6.0
+    metrics:
+      enabled: true
+      names:
+        - vbench.imaging_quality
+        - vbench.aesthetic_quality
+        - vbench.temporal_flickering
+        - vbench.motion_smoothness
+        - vbench.subject_consistency
+        - vbench.background_consistency
+        - vbench.dynamic_degree
+        - optical_flow.synthetic_optical_flow
+      calibration_path: assets/eval/worldmodel_synthetic_flow_calibration.json
+      skip_missing_deps: true
+      strict: false
+      unload_after_validation: true
+
+pipeline:
+  flow_shift: 5
+  dit_config:
+    local_attn_size: 6
+    sink_size: 1

examples/train/scenario/worldmodel/zelda/finetune_i2v.yaml

@@ -0,0 +1,88 @@
+# Matrix-Game 2.0 Zelda + multi-game I2V finetune.
+
+models:
+  student:
+    _target_: fastvideo.train.models.matrixgame2.matrixgame2.MatrixGame2Model
+    init_from: FastVideo/Matrix-Game-2.0-Base-Diffusers
+    trainable: true
+
+method:
+  _target_: fastvideo.train.methods.fine_tuning.finetune.FineTuneMethod
+
+training:
+  distributed:
+    num_gpus: 8
+    sp_size: 1
+    tp_size: 1
+    hsdp_replicate_dim: 1
+    hsdp_shard_dim: 8
+
+  data:
+    data_path:
+      data/zeldam2-clean: 1
+      # data/multi3d_games: 10
+    dataloader_num_workers: 1
+    train_batch_size: 1
+    training_cfg_rate: 0.0  # unused for MatrixGame2 I2V; no text_embedding CFG dropout
+    seed: 42
+    num_latent_t: 9
+    num_height: 480
+    num_width: 832
+    num_frames: 33
+
+  optimizer:
+    learning_rate: 2.0e-5
+    betas: [0.9, 0.95]
+    weight_decay: 0.0
+    lr_scheduler: constant
+    lr_warmup_steps: 0
+
+  loop:
+    max_train_steps: 60000
+    gradient_accumulation_steps: 1
+
+  checkpoint:
+    output_dir: outputs/matrixgame_finetune/checkpoints/zelda_with_mg_init
+    training_state_checkpointing_steps: 5000
+    checkpoints_total_limit: 3
+
+  tracker:
+    project_name: mg_1.3b_zelda
+    run_name: zelda_with_mg_init
+
+  model:
+    enable_gradient_checkpointing_type: full
+
+  dit_precision: fp32
+
+callbacks:
+  grad_clip:
+    _target_: fastvideo.train.callbacks.grad_clip.GradNormClipCallback
+    max_grad_norm: 1.0
+  validation:
+    _target_: fastvideo.train.callbacks.validation.ValidationCallback
+    pipeline_target: fastvideo.pipelines.basic.matrixgame2.matrixgame2_i2v_pipeline.MatrixGame2I2VPipeline
+    dataset_file: data/zelda_validation_data/validation_zelda.json
+    every_steps: 200
+    sampling_steps: [40]
+    num_frames: 33
+    overlay_actions: true
+    guidance_scale: 6.0
+    metrics:
+      enabled: true
+      names:
+        - vbench.imaging_quality
+        - vbench.aesthetic_quality
+        - vbench.temporal_flickering
+        - vbench.motion_smoothness
+        - vbench.subject_consistency
+        - vbench.background_consistency
+        - vbench.dynamic_degree
+        - optical_flow.synthetic_optical_flow
+      calibration_path: assets/eval/worldmodel_synthetic_flow_calibration.json
+      skip_missing_deps: true
+      strict: false
+      unload_after_validation: true
+
+pipeline:
+  flow_shift: 5

examples/train/scenario/worldmodel/zelda/self_forcing_causal_i2v.yaml

@@ -0,0 +1,120 @@
+# Self-forcing distillation: Zelda world model I2V Causal
+
+models:
+  student:
+    _target_: fastvideo.train.models.matrixgame2.matrixgame2_causal.MatrixGame2CausalModel
+    init_from: mignonjia/mg_causal_zelda
+    trainable: true
+  teacher:
+    _target_: fastvideo.train.models.matrixgame2.matrixgame2.MatrixGame2Model
+    init_from: mignonjia/mg_bidirectional_zelda
+    trainable: false
+    disable_custom_init_weights: true
+  critic:
+    _target_: fastvideo.train.models.matrixgame2.matrixgame2.MatrixGame2Model
+    init_from: mignonjia/mg_bidirectional_zelda
+    trainable: true
+    disable_custom_init_weights: true
+
+method:
+  _target_: fastvideo.train.methods.distribution_matching.self_forcing.SelfForcingMethod
+  rollout_mode: simulate
+  generator_update_interval: 5
+  dmd_denoising_steps: [1000, 750, 500, 250]
+  warp_denoising_step: true
+  min_timestep_ratio: 0.02
+  max_timestep_ratio: 0.98
+
+  chunk_size: 3
+  student_sample_type: sde
+  same_step_across_blocks: true
+  last_step_only: false
+  context_noise: 0.0
+  enable_gradient_in_rollout: true
+  start_gradient_frame: 0
+
+  # Critic optimizer
+  fake_score_learning_rate: 3.0e-7
+  fake_score_betas: [0.9, 0.95]
+  fake_score_lr_scheduler: constant
+
+training:
+  distributed:
+    num_gpus: 8
+    sp_size: 1
+    tp_size: 1
+    hsdp_replicate_dim: 1
+    hsdp_shard_dim: 8
+
+  data:
+    data_path: data/zeldam2-clean
+    dataloader_num_workers: 4
+    train_batch_size: 1
+    training_cfg_rate: 0.0
+    seed: 1001
+    num_latent_t: 9
+    num_height: 480
+    num_width: 832
+    num_frames: 33
+
+  optimizer:
+    learning_rate: 3.0e-6
+    betas: [0.9, 0.95]
+    weight_decay: 0.0
+    lr_scheduler: constant
+    lr_warmup_steps: 0
+
+  loop:
+    max_train_steps: 1000
+    gradient_accumulation_steps: 1
+
+  checkpoint:
+    output_dir: outputs/zelda_causal_self_forcing
+    training_state_checkpointing_steps: 1000
+    checkpoints_total_limit: 1
+
+  tracker:
+    project_name: wangame_sf
+    run_name: mg2_self_forcing_9_latents
+
+  model:
+    enable_gradient_checkpointing_type: full
+
+  dit_precision: fp32
+
+callbacks:
+  ema:
+    decay: 0.99
+    start_iter: 200
+  grad_clip:
+    max_grad_norm: 1.0
+  validation:
+    pipeline_target: fastvideo.pipelines.basic.matrixgame2.matrixgame2_causal_dmd_pipeline.MatrixGame2CausalDMDPipeline
+    dataset_file: data/zelda_validation_data/validation_zelda.json
+    every_steps: 100
+    sampling_steps: [4]
+    sampling_timesteps: [1000, 750, 500, 250]
+    num_frames: 153
+    overlay_actions: true
+    keyboard_value_scale: 1.0
+    metrics:
+      enabled: true
+      names:
+        - vbench.imaging_quality
+        - vbench.aesthetic_quality
+        - vbench.temporal_flickering
+        - vbench.motion_smoothness
+        - vbench.subject_consistency
+        - vbench.background_consistency
+        - vbench.dynamic_degree
+        - optical_flow.synthetic_optical_flow
+      calibration_path: assets/eval/worldmodel_synthetic_flow_calibration.json
+      skip_missing_deps: true
+      strict: false
+      unload_after_validation: true
+
+pipeline:
+  flow_shift: 5
+  dit_config:
+    local_attn_size: 6
+    sink_size: 1

examples/train/scenario/worldmodel/zelda/streaming_long_tuning_causal_i2v.yaml

@@ -0,0 +1,140 @@
+# MatrixGame2 I2V LongLive-style streaming distillation: Zelda world model I2V Causal
+# Student init from self forcing after 1k steps
+
+models:
+  student:
+    _target_: fastvideo.train.models.matrixgame2.matrixgame2_causal.MatrixGame2CausalModel
+    init_from: mignonjia/mg_sf_distilled_zelda_1k_steps 
+    trainable: true
+    # transformer_override_safetensor: outputs/matrixgame_dmd/checkpoints/mg_zelda_sf_m2/checkpoint-1000_weight_only/ema/generator_ema.safetensors
+  teacher:
+    _target_: fastvideo.train.models.matrixgame2.matrixgame2.MatrixGame2Model
+    init_from: mignonjia/mg_bidirectional_zelda
+    trainable: false
+    disable_custom_init_weights: true
+  critic:
+    _target_: fastvideo.train.models.matrixgame2.matrixgame2.MatrixGame2Model
+    init_from: mignonjia/mg_bidirectional_zelda
+    trainable: true
+    disable_custom_init_weights: true
+
+method:
+  _target_: fastvideo.train.methods.distribution_matching.streaming_long_tuning.StreamingLongTuningMethod
+  rollout_mode: simulate
+  generator_update_interval: 5
+  dmd_denoising_steps: [1000, 750, 500, 250]
+  warp_denoising_step: true
+  min_timestep_ratio: 0.02
+  max_timestep_ratio: 0.98
+
+  chunk_size: 3
+  student_sample_type: sde
+  same_step_across_blocks: true
+  last_step_only: false
+  context_noise: 0.0
+  enable_gradient_in_rollout: true
+  start_gradient_frame: 0
+
+  streaming_training: true
+  streaming_chunk_size: 9
+  streaming_max_length: 39
+  streaming_fixed_overlap_latents: 3
+  streaming_reencode_overlap_anchor: true
+  streaming_anchor_inject_k: 1
+  streaming_require_full_blocks: true
+
+  multi_phased_distill_schedule:
+    - stage: streaming_long
+      start_step: 0
+      end_step: 3000
+      num_latent_t: 39
+      streaming_training: true
+      streaming_chunk_size: 9
+      streaming_max_length: 39
+      streaming_fixed_overlap_latents: 3
+
+  # Critic optimizer
+  fake_score_learning_rate: 3.0e-7
+  fake_score_betas: [0.9, 0.95]
+  fake_score_lr_scheduler: constant
+
+training:
+  distributed:
+    num_gpus: 8
+    sp_size: 1
+    tp_size: 1
+    hsdp_replicate_dim: 1
+    hsdp_shard_dim: 8
+
+  data:
+    data_path: data/zeldam2-clean
+    dataloader_num_workers: 4
+    train_batch_size: 1
+    training_cfg_rate: 0.0
+    seed: 1001
+    num_latent_t: 39
+    num_height: 480
+    num_width: 832
+    num_frames: 153
+
+  optimizer:
+    learning_rate: 3.0e-6
+    betas: [0.9, 0.95]
+    weight_decay: 0.0
+    lr_scheduler: constant
+    lr_warmup_steps: 0
+
+  loop:
+    max_train_steps: 3000
+    gradient_accumulation_steps: 1
+
+  checkpoint:
+    output_dir: outputs/zelda_causal_long_tuning
+    training_state_checkpointing_steps: 1000
+    checkpoints_total_limit: 2
+
+  tracker:
+    project_name: wangame_sf
+    run_name: mg2_39only_streaming_long
+
+  model:
+    enable_gradient_checkpointing_type: full
+
+  dit_precision: fp32
+
+callbacks:
+  ema:
+    decay: 0.99
+    start_iter: 200
+  grad_clip:
+    max_grad_norm: 1.0
+  validation:
+    pipeline_target: fastvideo.pipelines.basic.matrixgame2.matrixgame2_causal_dmd_pipeline.MatrixGame2CausalDMDPipeline
+    dataset_file: data/zelda_validation_data/validation_zelda.json
+    every_steps: 100
+    sampling_steps: [4]
+    sampling_timesteps: [1000, 750, 500, 250]
+    num_frames: 153
+    overlay_actions: true
+    keyboard_value_scale: 1.0
+    metrics:
+      enabled: true
+      names:
+        - vbench.imaging_quality
+        - vbench.aesthetic_quality
+        - vbench.temporal_flickering
+        - vbench.motion_smoothness
+        - vbench.subject_consistency
+        - vbench.background_consistency
+        - vbench.dynamic_degree
+        - optical_flow.synthetic_optical_flow
+      calibration_path: assets/eval/worldmodel_synthetic_flow_calibration.json
+      skip_missing_deps: true
+      strict: false
+      unload_after_validation: true
+
+pipeline:
+  flow_shift: 5
+  dit_config:
+    local_attn_size: 6
+    sink_size: 1

fastvideo/api/compat.py

@@ -159,16 +159,6 @@ def legacy_from_pretrained_to_config(
             preset_refine["guidance_scale"] = value
         elif key in {"enable_stage_verification", "use_fsdp_inference", "disable_autocast"}:
             engine[key] = value
-        elif key == "batching_mode":
-            engine.setdefault("batching", {})["mode"] = value
-        elif key == "batching_max_size":
-            engine.setdefault("batching", {})["max_size"] = value
-        elif key == "batching_delay_ms":
-            engine.setdefault("batching", {})["delay_ms"] = value
-        elif key == "batching_config":
-            engine.setdefault("batching", {})["config_path"] = value
-        elif key == "enable_batching_metrics":
-            engine.setdefault("batching", {})["enable_metrics"] = value
         elif key == "override_text_encoder_quant":
             quantization["text_encoder_quant"] = value
         elif key == "workload_type":
@@ -254,11 +244,6 @@ def generator_config_to_fastvideo_args(config: GeneratorConfig | Mapping[str, An
         "enable_stage_verification": engine.enable_stage_verification,
         "use_fsdp_inference": engine.use_fsdp_inference,
         "disable_autocast": engine.disable_autocast,
-        "batching_mode": engine.batching.mode,
-        "batching_max_size": engine.batching.max_size,
-        "batching_delay_ms": engine.batching.delay_ms,
-        "batching_config": engine.batching.config_path,
-        "enable_batching_metrics": engine.batching.enable_metrics,
     }
     if normalized.pipeline.workload_type is not None:
         kwargs["workload_type"] = normalized.pipeline.workload_type

fastvideo/api/flux.py

@@ -0,0 +1,27 @@
+# SPDX-License-Identifier: Apache-2.0
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+from fastvideo.api.sampling_param import SamplingParam
+
+
+@dataclass
+class FluxSamplingParam(SamplingParam):
+
+    prompt: str | None = "a photo of a cat"
+    negative_prompt: str = ""
+
+    num_videos_per_prompt: int = 1
+    seed: int = 0
+
+    num_frames: int = 1
+    height: int = 1024
+    width: int = 1024
+    fps: int = 1
+
+    num_inference_steps: int = 28
+    guidance_scale: float = 3.5
+    use_embedded_guidance: bool = True
+    true_cfg_scale: float = 1.0

fastvideo/api/sampling_param.py

@@ -90,6 +90,10 @@ class SamplingParam:
     num_inference_steps_sr: int = 50
     guidance_scale: float = 1.0
     guidance_scale_2: float | None = None
+    # Embedded guidance (FLUX): do not treat ``guidance_scale > 1`` as classic CFG.
+    use_embedded_guidance: bool = False
+    # Diffusers-style true CFG for FLUX when > 1 (requires negative prompt encoding).
+    true_cfg_scale: float = 1.0
     guidance_rescale: float = 0.0
     boundary_ratio: float | None = None
     sigmas: list[float] | None = None
@@ -325,6 +329,18 @@ class SamplingParam:
             default=SamplingParam.guidance_rescale,
             help="Guidance rescale factor",
         )
+        parser.add_argument(
+            "--use-embedded-guidance",
+            action="store_true",
+            default=SamplingParam.use_embedded_guidance,
+            help="Use embedded guidance scale (FLUX-style) instead of classic CFG",
+        )
+        parser.add_argument(
+            "--true-cfg-scale",
+            type=float,
+            default=SamplingParam.true_cfg_scale,
+            help="True CFG scale for FLUX when > 1 (requires negative prompt encoding)",
+        )
         parser.add_argument(
             "--boundary-ratio",
             type=float,

fastvideo/api/schema.py

@@ -71,15 +71,6 @@ class QuantizationConfig:
     transformer_quant: str | None = None
 
 
-@dataclass
-class BatchingConfig:
-    mode: Literal["disabled", "dynamic"] = "disabled"
-    max_size: int = 1
-    delay_ms: float = 0.0
-    config_path: str | None = None
-    enable_metrics: bool = False
-
-
 @dataclass
 class EngineConfig:
     num_gpus: int = 1
@@ -91,7 +82,6 @@ class EngineConfig:
     use_fsdp_inference: bool = False
     disable_autocast: bool = False
     quantization: QuantizationConfig | None = None
-    batching: BatchingConfig = field(default_factory=BatchingConfig)
 
 
 @dataclass
@@ -160,6 +150,7 @@ class SamplingConfig:
     guidance_scale_2: float | None = None
     guidance_rescale: float = 0.0
     true_cfg_scale: float | None = None
+    use_embedded_guidance: bool | None = None
     boundary_ratio: float | None = None
     sigmas: list[float] | None = None
 
@@ -290,7 +281,6 @@ class ServeConfig:
 
 
 __all__ = [
-    "BatchingConfig",
     "CompileConfig",
     "ComponentConfig",
     "ContinuationState",

fastvideo/attention/backends/flash_attn.py

@@ -1,13 +1,15 @@
 # SPDX-License-Identifier: Apache-2.0
 
-import importlib.util
 import os
-
 import torch
 import torch.nn.functional as F
 from dataclasses import dataclass
 
-from fastvideo import envs
+from fastvideo.attention.utils.flash_attn_default import (
+    fa_version,
+    flash_attn_func_compilable,
+)
+
 from fastvideo.attention.backends.abstract import (
     AttentionBackend,
     AttentionImpl,
@@ -17,119 +19,6 @@ from fastvideo.attention.backends.abstract import (
 from fastvideo.logger import init_logger
 
 logger = init_logger(__name__)
-
-# FA4 (flash_attn.cute) is explicit opt-in via FASTVIDEO_FA4=1, mirroring the
-# kernel package's FASTVIDEO_VSA_CUTEDSL: its CuTeDSL kernels JIT-compile per
-# shape family and can fail at runtime on some arch/shape combinations, so it
-# is never auto-selected just because it is installed. Below sm90 a capability
-# gate in flash_attn_cute routes to FA2 the calls FA4 cannot serve there:
-# grad-enabled (its backward asserts sm90+) and GQA (pack_gqa fails CuTeDSL
-# JIT, observed on sm_89).
-if envs.FASTVIDEO_FA4:
-    try:
-        from fastvideo.attention.utils.flash_attn_cute import flash_attn_func
-    except ImportError as e:
-        raise RuntimeError(f"FASTVIDEO_FA4=1 but flash_attn.cute (FA4) is not usable ({e}); "
-                           "fix the FA4 install (see the flash-attn-4 pin in pyproject.toml) "
-                           "or unset FASTVIDEO_FA4.") from e
-    fa_version = "4"
-else:
-    try:
-        from flash_attn_interface import flash_attn_func as flash_attn_3_func
-
-        # flash_attn 3 no longer have a different API, see following commit:
-        # https://github.com/Dao-AILab/flash-attention/commit/ed209409acedbb2379f870bbd03abce31a7a51b7
-        flash_attn_func = flash_attn_3_func
-        fa_version = "3"
-    except ImportError:
-        from flash_attn import flash_attn_func as flash_attn_2_func
-        flash_attn_func = flash_attn_2_func
-        fa_version = "2"
-    try:
-        if importlib.util.find_spec("flash_attn.cute") is not None:
-            logger.info("flash_attn.cute (FA4) is installed but not enabled; "
-                        "set FASTVIDEO_FA4=1 to use it for inference.")
-    except ImportError:
-        pass
-
-# torch.compile traceability: the FA4/cute path (fa_version=="4") is
-# already a registered torch.library custom op, so dynamo treats it as a
-# graph node. The external FA2/FA3 `flash_attn_func` is NOT — dynamo
-# breaks the graph at the call site (observed: wanvideo.py self-attn,
-# once per layer every step), which fragments the compiled region and
-# blocks CUDA-graph capture. Wrap the FA2/FA3 default call in a custom
-# op (mirrors the FP4 `flash_attn_cute` template) so it becomes an
-# opaque-but-traceable node. The kernel still runs eager inside the op
-# (correct — flash-attn must run eager); only dynamo's treatment of the
-# boundary changes, so numerics are unchanged (SSIM-gate to confirm).
-if fa_version in ("2", "3"):
-    _fa_default = flash_attn_func
-
-    # Scope: this op covers exactly the q/k/v + softmax_scale + causal
-    # call shape used by FlashAttentionImpl.forward's default branch
-    # (see `flash_attn_func_compilable(...)` call site below). The
-    # masked/no-pad and varlen / cross-attn paths use different
-    # entry points (`flash_attn_no_pad`, `flash_attn_varlen_*`) which
-    # are intentionally out of scope for this PR — wrapping them is a
-    # natural follow-up. The wrapper's signature is the contract: any
-    # extra kwarg (dropout_p, window_size, alibi_slopes, deterministic,
-    # return_attn_probs, ...) raises TypeError at the call site, so
-    # silent loss of kwargs is not a failure mode.
-    @torch.library.custom_op(
-        "fastvideo::_flash_attn_default_forward",
-        mutates_args=(),
-        device_types="cuda",
-    )
-    def _flash_attn_default_forward(
-        q: torch.Tensor,
-        k: torch.Tensor,
-        v: torch.Tensor,
-        softmax_scale: float | None,
-        causal: bool,
-    ) -> torch.Tensor:
-        return _fa_default(q, k, v, softmax_scale=softmax_scale, causal=causal)
-
-    @torch.library.register_fake("fastvideo::_flash_attn_default_forward")
-    def _flash_attn_default_forward_fake(
-        q: torch.Tensor,
-        k: torch.Tensor,
-        v: torch.Tensor,
-        softmax_scale: float | None,
-        causal: bool,
-    ) -> torch.Tensor:
-        del softmax_scale, causal
-        # FA2/FA3 default path: [batch, seqlen_q, nheads, head_dim_v],
-        # same dtype/device as q (head dim taken from v).
-        return q.new_empty(q.shape[0], q.shape[1], q.shape[2], v.shape[-1])
-
-    def flash_attn_func_compilable(q, k, v, softmax_scale=None, causal=False):
-        # Autograd carve-out. The custom op above registers a forward + fake
-        # kernel but NO backward (register_autograd), so it is opaque to
-        # autograd. Inference runs under no_grad / inference_mode and routes
-        # through the traceable custom op — that is the torch.compile win, and
-        # the only path this PR claims. Training backprops through attention,
-        # so route grad-enabled calls to the original FA2/FA3 `flash_attn_func`
-        # (itself an autograd.Function, so backward is correct) at the cost of a
-        # dynamo graph break on the training path — i.e. pre-PR behavior, no
-        # regression. Full autograd parity for the custom op (mirroring the FP4
-        # cute template) is a tracked follow-up.
-        if torch.is_grad_enabled() and (q.requires_grad or k.requires_grad or v.requires_grad):
-            return _fa_default(q, k, v, softmax_scale=softmax_scale, causal=causal)
-        return torch.ops.fastvideo._flash_attn_default_forward(q, k, v, softmax_scale, causal)
-elif fa_version == "4":
-    # FA4 path: `flash_attn_func` (from `flash_attn_cute`) goes through a
-    # registered torch.library custom op (with an FA4 backward on sm90+;
-    # grad-enabled and GQA calls below sm90 route to FA2), so a passthrough
-    # is enough — no extra registration needed.
-    def flash_attn_func_compilable(q, k, v, softmax_scale=None, causal=False):
-        return flash_attn_func(q, k, v, softmax_scale=softmax_scale, causal=causal)
-else:
-    # Defensive: the probe above only ever sets fa_version to "2", "3",
-    # or "4"; an unexpected value means an import/probe regression and
-    # we want a loud error at import, not a silent NameError later.
-    raise RuntimeError(f"Unsupported FlashAttention version: {fa_version!r} — expected "
-                       f"'2', '3', or '4' from the import probe above.")
-
 logger.info("Using FlashAttention-%s backend", fa_version)
 
 # FP4 FA4 support: quantize Q/K to NVFP4 E2M1 for block-scaled MMA on Blackwell.
@@ -309,9 +198,17 @@ class FlashAttentionImpl(AttentionImpl):
         attn_metadata: FlashAttnMetadata,
     ):
         if (attn_metadata is not None and hasattr(attn_metadata, "attn_mask") and attn_metadata.attn_mask is not None):
+            # Route through the *_compilable wrappers so dynamo sees one
+            # traceable node for each masked entry point (the unpad/pad
+            # bookkeeping runs eager inside the custom op). On FA2 these
+            # wrappers go through ops with full register_autograd, so
+            # training also backprops through the op (no graph break on
+            # the training path); on FA3/FA4 they carve out to the
+            # autograd.Function for grad-enabled calls — see
+            # fastvideo/attention/utils/flash_attn_no_pad.py.
             from fastvideo.attention.utils.flash_attn_no_pad import (
-                flash_attn_no_pad,
-                flash_attn_varlen_qk_no_pad,
+                flash_attn_no_pad_compilable as flash_attn_no_pad,
+                flash_attn_varlen_qk_no_pad_compilable as flash_attn_varlen_qk_no_pad,
             )
 
             attn_mask = attn_metadata.attn_mask
@@ -338,7 +235,11 @@ class FlashAttentionImpl(AttentionImpl):
                 )
 
             qkv = torch.stack([query, key, value], dim=2)
-            attn_mask_padded = F.pad(attn_mask, (qkv.shape[1] - attn_mask.shape[1], 0), value=True)
+            key_padding_mask = _key_padding_mask_from_attn_mask(attn_mask, attn_mask.shape[-1]).to(device=query.device)
+            if key_padding_mask.shape[-1] > qkv.shape[1]:
+                raise ValueError("Invalid key padding mask length for FLASH_ATTN: "
+                                 f"expected at most {qkv.shape[1]}, got {key_padding_mask.shape[-1]}")
+            attn_mask_padded = F.pad(key_padding_mask, (qkv.shape[1] - key_padding_mask.shape[-1], 0), value=True)
             output = flash_attn_no_pad(qkv, attn_mask_padded, causal=False, dropout_p=0, softmax_scale=None)
         elif self.nvfp4_fa4:
             output = self._forward_nvfp4(query, key, value)

fastvideo/attention/backends/nabla.py

@@ -0,0 +1,147 @@
+# SPDX-License-Identifier: Apache-2.0
+"""NABLA block-sparse flex-attention backend (Kandinsky5 "nabla" checkpoints).
+
+The block mask is data-dependent: nablaT_v2 mean-pools 64-token blocks of the
+fractal-ordered sequence, thresholds the softmaxed block map, and ORs it with a
+precomputed spatio-temporal-window (STA) mask carried on the attention
+metadata. The mask spans the full sequence, so this backend does not support
+sequence parallelism — use it via LocalAttention only.
+"""
+
+import math
+from dataclasses import dataclass
+from typing import Any
+
+import torch
+
+try:
+    from torch.nn.attention.flex_attention import BlockMask, flex_attention
+    flex_attention = torch.compile(flex_attention, dynamic=False, mode="max-autotune-no-cudagraphs")
+    CAN_USE_FLEX_ATTN = True
+except ImportError:
+    CAN_USE_FLEX_ATTN = False
+
+from fastvideo.attention.backends.abstract import (
+    AttentionBackend,
+    AttentionImpl,
+    AttentionMetadata,
+    AttentionMetadataBuilder,
+)
+
+
+def nablaT_v2(
+    q: torch.Tensor,
+    k: torch.Tensor,
+    sta: torch.Tensor,
+    thr: float = 0.9,
+) -> "BlockMask":
+    q = q.transpose(1, 2).contiguous()
+    k = k.transpose(1, 2).contiguous()
+
+    # Map estimation
+    B, h, S, D = q.shape
+    s1 = S // 64
+    qa = q.reshape(B, h, s1, 64, D).mean(-2)
+    ka = k.reshape(B, h, s1, 64, D).mean(-2).transpose(-2, -1)
+    map = qa @ ka
+
+    map = torch.softmax(map / math.sqrt(D), dim=-1)
+    # Map binarization
+    vals, inds = map.sort(-1)
+    cvals = vals.cumsum_(-1)
+    mask = (cvals >= 1 - thr).int()
+    mask = mask.gather(-1, inds.argsort(-1))
+
+    mask = torch.logical_or(mask, sta)
+
+    # BlockMask creation
+    kv_nb = mask.sum(-1).to(torch.int32)
+    kv_inds = mask.argsort(dim=-1, descending=True).to(torch.int32)
+    return BlockMask.from_kv_blocks(torch.zeros_like(kv_nb), kv_inds, kv_nb, kv_inds, BLOCK_SIZE=64, mask_mod=None)
+
+
+class NablaAttentionBackend(AttentionBackend):
+
+    @staticmethod
+    def get_name() -> str:
+        return "NABLA_ATTN"
+
+    @staticmethod
+    def get_impl_cls() -> type["NablaAttentionImpl"]:
+        return NablaAttentionImpl
+
+    @staticmethod
+    def get_metadata_cls() -> type["NablaAttentionMetadata"]:
+        return NablaAttentionMetadata
+
+    @staticmethod
+    def get_builder_cls() -> type["NablaAttentionMetadataBuilder"]:
+        return NablaAttentionMetadataBuilder
+
+
+@dataclass
+class NablaAttentionMetadata(AttentionMetadata):
+    # Block-level STA window mask [1, 1, S/64, S/64], precomputed once per run.
+    sta_mask: torch.Tensor = None  # type: ignore[assignment]
+    # Cumulative-probability threshold for block-map binarization.
+    P: float = 0.9
+    visual_shape: tuple[int, int, int] = (0, 0, 0)
+
+
+class NablaAttentionMetadataBuilder(AttentionMetadataBuilder):
+
+    def __init__(self) -> None:
+        pass
+
+    def prepare(self) -> None:
+        pass
+
+    def build(
+        self,
+        current_timestep: int,
+        sta_mask: torch.Tensor,
+        P: float,
+        visual_shape: tuple[int, int, int],
+        **kwargs: Any,
+    ) -> NablaAttentionMetadata:
+        return NablaAttentionMetadata(
+            current_timestep=current_timestep,
+            sta_mask=sta_mask,
+            P=P,
+            visual_shape=visual_shape,
+        )
+
+
+class NablaAttentionImpl(AttentionImpl):
+
+    def __init__(
+        self,
+        num_heads: int,
+        head_size: int,
+        softmax_scale: float,
+        causal: bool = False,
+        num_kv_heads: int | None = None,
+        prefix: str = "",
+        **extra_impl_args,
+    ) -> None:
+        if not CAN_USE_FLEX_ATTN:
+            raise RuntimeError("NABLA attention requires torch.nn.attention.flex_attention, "
+                               "which is unavailable in this PyTorch build.")
+        if causal:
+            raise ValueError("NABLA attention does not support causal masking.")
+
+    def forward(
+        self,
+        query: torch.Tensor,
+        key: torch.Tensor,
+        value: torch.Tensor,
+        attn_metadata: NablaAttentionMetadata,
+    ) -> torch.Tensor:
+        # q/k/v: [B, S, heads, head_dim], fractal-ordered by the model; S % 64 == 0.
+        block_mask = nablaT_v2(query, key, attn_metadata.sta_mask, thr=attn_metadata.P)
+        return flex_attention(
+            query=query.transpose(1, 2),
+            key=key.transpose(1, 2),
+            value=value.transpose(1, 2),
+            block_mask=block_mask,
+        ).transpose(1, 2)

fastvideo/attention/backends/sdpa.py

@@ -2,6 +2,7 @@
 
 import torch
 from dataclasses import dataclass
+from torch.nn import functional as F
 from fastvideo.attention.backends.abstract import (  # FlashAttentionMetadata,
     AttentionBackend, AttentionImpl, AttentionMetadata, AttentionMetadataBuilder)
 from fastvideo.logger import init_logger
@@ -14,12 +15,18 @@ class SDPABackend(AttentionBackend):
     accept_output_buffer: bool = True
 
     @staticmethod
-    def get_supported_head_sizes() -> list[int]:
-        return [32, 64, 96, 128, 160, 192, 224, 256]
+    def get_supported_head_sizes() -> list[int] | None:
+        # torch.nn.functional.scaled_dot_product_attention is head-size
+        # agnostic: the math backend handles any size and the fused kernels
+        # fall back internally when they cannot. The previous list
+        # ([32, 64, ..., 256]) was copied from FlashAttentionBackend in the
+        # v1 refactor (#270) and wrongly excluded sizes such as 80 (e.g.
+        # CLIP vision encoders). None means "no restriction".
+        return None
 
     @staticmethod
     def get_name() -> str:
-        return "SDPA"
+        return "TORCH_SDPA"
 
     @staticmethod
     def get_impl_cls() -> type["SDPAImpl"]:
@@ -49,9 +56,51 @@ class SDPAMetadataBuilder(AttentionMetadataBuilder):
             current_timestep: int,
             attn_mask: torch.Tensor,
     ) -> SDPAMetadata:
+        # Store the mask exactly as passed. The metadata is cross-backend:
+        # call sites (HYWorld, HunyuanVideo15) build SDPAMetadata while the
+        # layer's selector may pick FLASH_ATTN, and the shared convention for
+        # padding masks is the tokenizer-style 2D [batch, key_len]. Any
+        # reshaping for torch.sdpa happens inside the SDPA impl
+        # (_normalize_attn_mask_for_sdpa).
         return SDPAMetadata(current_timestep=current_timestep, attn_mask=attn_mask)
 
 
+def _normalize_attn_mask_for_sdpa(
+    attn_mask: torch.Tensor | None,
+    query: torch.Tensor,
+    key: torch.Tensor,
+) -> torch.Tensor | None:
+    if attn_mask is None:
+        return None
+
+    attn_mask = attn_mask.to(device=query.device)
+    # F.scaled_dot_product_attention only accepts bool or float masks;
+    # tokenizers commonly produce int64 0/1 padding masks.
+    if attn_mask.dtype != torch.bool and not attn_mask.dtype.is_floating_point:
+        attn_mask = attn_mask != 0
+
+    key_len = key.shape[-2]
+    if attn_mask.shape[-1] > key_len:
+        raise ValueError("Invalid attention mask length for SDPA: "
+                         f"expected at most {key_len}, got {attn_mask.shape[-1]}")
+    if attn_mask.shape[-1] < key_len:
+        # Front-pad as "attend": double-stream layouts (HYWorld) prepend
+        # non-text tokens the tokenizer mask does not cover.
+        valid_value = True if attn_mask.dtype == torch.bool else 0.0
+        attn_mask = F.pad(attn_mask, (key_len - attn_mask.shape[-1], 0), value=valid_value)
+
+    if attn_mask.dim() == 2:
+        # In-tree producers pass 2D [batch, key_len] padding masks; lift to a
+        # broadcastable [batch, 1, 1, key_len] here so torch.sdpa does not
+        # reinterpret 2D as its documented [query_len, key_len] broadcast.
+        return attn_mask[:, None, None, :]
+    if attn_mask.dim() == 3:
+        return attn_mask[:, None, :, :]
+    if attn_mask.dim() == 4:
+        return attn_mask
+    raise ValueError(f"Unsupported attention mask shape for SDPA: {attn_mask.shape}")
+
+
 class SDPAImpl(AttentionImpl):
 
     def __init__(
@@ -82,6 +131,7 @@ class SDPAImpl(AttentionImpl):
 
         attn_mask = attn_metadata.attn_mask if (attn_metadata is not None
                                                 and hasattr(attn_metadata, "attn_mask")) else None
+        attn_mask = _normalize_attn_mask_for_sdpa(attn_mask, query, key)
         attn_kwargs = {
             "attn_mask": attn_mask,
             "dropout_p": self.dropout,

fastvideo/attention/layer.py

@@ -252,6 +252,7 @@ class LocalAttention(nn.Module):
                  causal: bool = False,
                  supported_attention_backends: tuple[AttentionBackendEnum, ...]
                  | None = None,
+                 default_backend: AttentionBackendEnum | None = None,
                  **extra_impl_args) -> None:
         super().__init__()
         if softmax_scale is None:
@@ -262,7 +263,10 @@ class LocalAttention(nn.Module):
             num_kv_heads = num_heads
 
         dtype = get_compute_dtype()
-        attn_backend = get_attn_backend(head_size, dtype, supported_attention_backends=supported_attention_backends)
+        attn_backend = get_attn_backend(head_size,
+                                        dtype,
+                                        supported_attention_backends=supported_attention_backends,
+                                        default_backend=default_backend)
         impl_cls = attn_backend.get_impl_cls()
         self.attn_impl = impl_cls(num_heads=num_heads,
                                   head_size=head_size,

fastvideo/attention/selector.py

@@ -84,8 +84,9 @@ def get_attn_backend(
     dtype: torch.dtype,
     supported_attention_backends: tuple[AttentionBackendEnum, ...]
     | None = None,
+    default_backend: AttentionBackendEnum | None = None,
 ) -> type[AttentionBackend]:
-    return _cached_get_attn_backend(head_size, dtype, supported_attention_backends)
+    return _cached_get_attn_backend(head_size, dtype, supported_attention_backends, default_backend)
 
 
 @cache
@@ -94,6 +95,7 @@ def _cached_get_attn_backend(
     dtype: torch.dtype,
     supported_attention_backends: tuple[AttentionBackendEnum, ...]
     | None = None,
+    default_backend: AttentionBackendEnum | None = None,
 ) -> type[AttentionBackend]:
     # Check whether a particular choice of backend was
     # previously forced.
@@ -112,6 +114,12 @@ def _cached_get_attn_backend(
         if backend_by_env_var is not None:
             selected_backend = backend_name_to_enum(backend_by_env_var)
 
+    # Layer-level default (e.g. a checkpoint that requires a specific sparse
+    # backend). Lower precedence than the global force and the env var, so
+    # users can still override it.
+    if selected_backend is None and default_backend is not None:
+        selected_backend = default_backend
+
     # get device-specific attn_backend
     from fastvideo.platforms import current_platform
 

fastvideo/attention/utils/flash_attn_default.py

@@ -0,0 +1,251 @@
+# SPDX-License-Identifier: Apache-2.0
+"""torch.compile-traceable wrapper for the FA2/FA3/FA4 default attention path.
+
+The FA4/cute path (`fa_version == "4"`) is already a registered
+`torch.library.custom_op` in `fastvideo.attention.utils.flash_attn_cute`, so
+dynamo treats it as a graph node. The external FA2/FA3 ``flash_attn_func`` is
+NOT — dynamo breaks the graph at the call site (observed: wanvideo.py
+self-attn, once per layer every step), which fragments the compiled region
+and blocks CUDA-graph capture. Wrap the FA2/FA3 default call in a custom op
+(mirrors the FP4 `flash_attn_cute` template) so it becomes an
+opaque-but-traceable node. The kernel still runs eager inside the op
+(correct — flash-attn must run eager); only dynamo's treatment of the
+boundary changes, so numerics are unchanged (SSIM-gated).
+
+Autograd: FA2 has full ``register_autograd`` parity — the custom op's
+backward calls flash_attn's ``_flash_attn_backward`` directly, so training
+backprops *through* the op (no graph break on the training path either).
+FA3 currently keeps the no-backward + carve-out pattern from PR #1373
+because FA3's private backward signature wants validation on a real Hopper
+box (gated on Kuan-Hao's Modal FA3 setup PR). Once that lands the FA3 path
+can mirror FA2.
+
+Lives in `attention/utils/` (sibling of `flash_attn_cute.py` and
+`flash_attn_no_pad.py`) so it can be imported by any backend that wants the
+traceable FA default call without pulling in backend dispatch logic. The
+backend (`attention/backends/flash_attn.py`) just imports
+`flash_attn_func_compilable` and `fa_version` from here.
+"""
+
+import importlib.util
+
+import torch
+
+from fastvideo import envs
+from fastvideo.logger import init_logger
+
+logger = init_logger(__name__)
+
+# Pick the same backend the rest of FastVideo picked for `flash_attn_func`
+# (FA4/cute → FA3 → FA2). Mirror the precedence used in
+# `attention/utils/flash_attn_no_pad.py` so the two probes always agree.
+#
+# FA4 (flash_attn.cute) is explicit opt-in via FASTVIDEO_FA4=1: its CuTeDSL
+# kernels JIT-compile per shape family and can fail at runtime on some
+# arch/shape combinations, so it is never auto-selected just because it is
+# installed.
+if envs.FASTVIDEO_FA4:
+    try:
+        from fastvideo.attention.utils.flash_attn_cute import flash_attn_func
+    except ImportError as e:
+        raise RuntimeError(f"FASTVIDEO_FA4=1 but flash_attn.cute (FA4) is not usable ({e}); "
+                           "fix the FA4 install (see the flash-attn-4 pin in pyproject.toml) "
+                           "or unset FASTVIDEO_FA4.") from e
+    fa_version = "4"
+else:
+    try:
+        from flash_attn_interface import flash_attn_func as flash_attn_3_func
+
+        # flash_attn 3 no longer has a different API, see following commit:
+        # https://github.com/Dao-AILab/flash-attention/commit/ed209409acedbb2379f870bbd03abce31a7a51b7
+        flash_attn_func = flash_attn_3_func
+        fa_version = "3"
+    except ImportError:
+        from flash_attn import flash_attn_func as flash_attn_2_func
+        flash_attn_func = flash_attn_2_func
+        fa_version = "2"
+    try:
+        if importlib.util.find_spec("flash_attn.cute") is not None:
+            logger.info("flash_attn.cute (FA4) is installed but not enabled; "
+                        "set FASTVIDEO_FA4=1 to use it for inference.")
+    except ImportError:
+        pass
+
+if fa_version == "2":
+    # Scope: this op covers exactly the q/k/v + softmax_scale + causal call
+    # shape used by FlashAttentionImpl.forward's default branch (see
+    # `flash_attn_func_compilable(...)` call site in
+    # `attention/backends/flash_attn.py`). The masked/no-pad and varlen /
+    # cross-attn paths use different entry points
+    # (`flash_attn_no_pad`, `flash_attn_varlen_*`) which live in
+    # `attention/utils/flash_attn_no_pad.py`. The wrapper's signature is the
+    # contract: any extra kwarg (dropout_p, window_size, alibi_slopes,
+    # deterministic, return_attn_probs, ...) raises TypeError at the call
+    # site, so silent loss of kwargs is not a failure mode.
+    from flash_attn.flash_attn_interface import _flash_attn_backward as _fa2_backward
+    _fa_default = flash_attn_func
+
+    @torch.library.custom_op(
+        "fastvideo::_flash_attn_default_forward",
+        mutates_args=(),
+        device_types="cuda",
+    )
+    def _flash_attn_default_forward(
+        q: torch.Tensor,
+        k: torch.Tensor,
+        v: torch.Tensor,
+        softmax_scale: float | None,
+        causal: bool,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        # `return_attn_probs=True` asks FA2 to also return softmax_lse +
+        # S_dmask. We need softmax_lse to feed the backward; S_dmask is the
+        # dropout mask (always None here since dropout_p is fixed at 0).
+        out, softmax_lse, _ = _fa_default(q, k, v, softmax_scale=softmax_scale, causal=causal, return_attn_probs=True)
+        return out, softmax_lse
+
+    @torch.library.register_fake("fastvideo::_flash_attn_default_forward")
+    def _flash_attn_default_forward_fake(
+        q: torch.Tensor,
+        k: torch.Tensor,
+        v: torch.Tensor,
+        softmax_scale: float | None,
+        causal: bool,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        del softmax_scale, causal
+        # FA2 default path: out = [batch, seqlen_q, nheads, head_dim_v],
+        # softmax_lse = [batch, nheads, seqlen_q], fp32 regardless of q dtype.
+        b, sq, hq = q.shape[0], q.shape[1], q.shape[2]
+        out = q.new_empty(b, sq, hq, v.shape[-1])
+        lse = q.new_empty(b, hq, sq, dtype=torch.float32)
+        return out, lse
+
+    def _flash_attn_default_setup_context(ctx, inputs, output):
+        q, k, v, softmax_scale, causal = inputs
+        out, lse = output
+        ctx.save_for_backward(q, k, v, out, lse)
+        # `lse` is an auxiliary output we save to feed FA2's backward; nobody
+        # should differentiate through it. Mark it non-differentiable so
+        # autograd errors loudly if a caller wires it into a loss, rather
+        # than silently producing zero/None grads through the `del grad_lse`
+        # in our backward.
+        ctx.mark_non_differentiable(lse)
+        # FA2's *forward* substitutes `1 / sqrt(head_dim)` for `softmax_scale=None`
+        # internally; FA2's *backward* (`_flash_attn_backward`) demands a concrete
+        # float in its C++ schema and rejects None at the binding boundary. Resolve
+        # the default here so the value saved on ctx (and passed to backward) is
+        # always a real float — matches what FA2's own autograd.Function does.
+        if softmax_scale is None:
+            softmax_scale = q.shape[-1]**-0.5
+        ctx.softmax_scale = softmax_scale
+        ctx.causal = causal
+
+    def _flash_attn_default_backward(ctx, grad_out, grad_lse):
+        # We only differentiate `out`; softmax_lse is saved-for-backward, not
+        # a real differentiable output. (Mirrors the FP4 cute template.)
+        del grad_lse
+        q, k, v, out, lse = ctx.saved_tensors
+        dq = torch.empty_like(q)
+        dk = torch.empty_like(k)
+        dv = torch.empty_like(v)
+        # FA2's `_flash_attn_backward` writes into dq/dk/dv in place. The
+        # extra kwargs (window_size_*, softcap, alibi_slopes, deterministic,
+        # rng_state) are pinned to the same defaults the forward wrapper
+        # uses — flash-attn==2.8.1 (the version FastVideo pins) requires
+        # all of them explicitly. `rng_state=None` is correct for our
+        # `dropout_p=0` configuration.
+        _fa2_backward(
+            grad_out,
+            q,
+            k,
+            v,
+            out,
+            lse,
+            dq,
+            dk,
+            dv,
+            dropout_p=0.0,
+            softmax_scale=ctx.softmax_scale,
+            causal=ctx.causal,
+            window_size_left=-1,
+            window_size_right=-1,
+            softcap=0.0,
+            alibi_slopes=None,
+            deterministic=False,
+            rng_state=None,
+        )
+        return dq, dk, dv, None, None
+
+    torch.library.register_autograd(
+        "fastvideo::_flash_attn_default_forward",
+        _flash_attn_default_backward,
+        setup_context=_flash_attn_default_setup_context,
+    )
+
+    def flash_attn_func_compilable(q, k, v, softmax_scale=None, causal=False):
+        # Backward is registered: autograd flows through the op (training
+        # path is also traceable; no carve-out needed). Public API matches
+        # `flash_attn_func` — returns just `out`; we drop the saved-for-
+        # backward `lse` here so callers see the original single-tensor
+        # contract.
+        out, _ = torch.ops.fastvideo._flash_attn_default_forward(q, k, v, softmax_scale, causal)
+        return out
+elif fa_version == "3":
+    # FA3 path: same forward+fake custom op as the original PR #1373, with
+    # the autograd carve-out kept. The full backward (mirroring the FA2 leg
+    # above) wants a Hopper box for grad-check validation, which we don't
+    # have until Kuan-Hao's Modal FA3 setup PR lands. Until then this keeps
+    # inference traceable + training correct (via the original
+    # autograd.Function path + a pre-PR-style graph break on training).
+    _fa_default = flash_attn_func
+
+    @torch.library.custom_op(
+        "fastvideo::_flash_attn_default_forward",
+        mutates_args=(),
+        device_types="cuda",
+    )
+    def _flash_attn_default_forward(
+        q: torch.Tensor,
+        k: torch.Tensor,
+        v: torch.Tensor,
+        softmax_scale: float | None,
+        causal: bool,
+    ) -> torch.Tensor:
+        return _fa_default(q, k, v, softmax_scale=softmax_scale, causal=causal)
+
+    @torch.library.register_fake("fastvideo::_flash_attn_default_forward")
+    def _flash_attn_default_forward_fake(
+        q: torch.Tensor,
+        k: torch.Tensor,
+        v: torch.Tensor,
+        softmax_scale: float | None,
+        causal: bool,
+    ) -> torch.Tensor:
+        del softmax_scale, causal
+        return q.new_empty(q.shape[0], q.shape[1], q.shape[2], v.shape[-1])
+
+    def flash_attn_func_compilable(q, k, v, softmax_scale=None, causal=False):
+        # Autograd carve-out. The custom op above registers a forward + fake
+        # kernel but NO backward (register_autograd), so it is opaque to
+        # autograd. Inference runs under no_grad / inference_mode and routes
+        # through the traceable custom op — that is the torch.compile win, and
+        # the only path this PR claims. Training backprops through attention,
+        # so route grad-enabled calls to the original FA2/FA3 `flash_attn_func`
+        # (itself an autograd.Function, so backward is correct) at the cost of a
+        # dynamo graph break on the training path — i.e. pre-PR behavior, no
+        # regression. Full autograd parity for the custom op (mirroring the FP4
+        # cute template) is a tracked follow-up.
+        if torch.is_grad_enabled() and (q.requires_grad or k.requires_grad or v.requires_grad):
+            return _fa_default(q, k, v, softmax_scale=softmax_scale, causal=causal)
+        return torch.ops.fastvideo._flash_attn_default_forward(q, k, v, softmax_scale, causal)
+elif fa_version == "4":
+    # FA4 path: `flash_attn_func` is already a torch.library custom op
+    # (registered in `fastvideo.attention.utils.flash_attn_cute`), so a
+    # passthrough is enough — no extra registration needed.
+    def flash_attn_func_compilable(q, k, v, softmax_scale=None, causal=False):
+        return flash_attn_func(q, k, v, softmax_scale=softmax_scale, causal=causal)
+else:
+    # Defensive: the probe above only ever sets fa_version to "2", "3",
+    # or "4"; an unexpected value means an import/probe regression and
+    # we want a loud error at import, not a silent NameError later.
+    raise RuntimeError(f"Unsupported FlashAttention version: {fa_version!r} — expected "
+                       f"'2', '3', or '4' from the import probe above.")

fastvideo/attention/utils/flash_attn_no_pad.py

@@ -24,7 +24,7 @@ from flash_attn.bert_padding import pad_input, unpad_input
 from fastvideo import envs
 
 
-def _resolve_flash_attn_varlen_func() -> Any:
+def _resolve_flash_attn_varlen_func() -> tuple[Any, str]:
     if envs.FASTVIDEO_FA4:
         # FA4 cute is explicit opt-in (see fastvideo/attention/backends/
         # flash_attn.py); with FASTVIDEO_FA4=1 an unimportable FA4 build must
@@ -39,20 +39,28 @@ def _resolve_flash_attn_varlen_func() -> Any:
                                "fix the FA4 install (see the flash-attn-4 pin in pyproject.toml) "
                                "or unset FASTVIDEO_FA4.") from e
 
-        return flash_attn_varlen_func_cute
+        return flash_attn_varlen_func_cute, "4"
     try:
         from flash_attn_interface import (
             flash_attn_varlen_func as flash_attn_varlen_func_interface, )
 
-        return flash_attn_varlen_func_interface
+        return flash_attn_varlen_func_interface, "3"
     except ImportError:
         from flash_attn import (
             flash_attn_varlen_func as flash_attn_varlen_func_flash, )
 
-        return flash_attn_varlen_func_flash
+        return flash_attn_varlen_func_flash, "2"
 
 
-flash_attn_varlen_func_impl = _resolve_flash_attn_varlen_func()
+flash_attn_varlen_func_impl, _FA_VARLEN_VERSION = _resolve_flash_attn_varlen_func()
+
+# FA2-only: the private varlen backward we register against the custom ops
+# below. FA3 / FA4 have different private signatures and validation paths
+# (Hopper / Blackwell boxes) — those legs keep the autograd carve-out
+# pattern from PR #1373 until their setup PRs land.
+if _FA_VARLEN_VERSION == "2":
+    from flash_attn.flash_attn_interface import (
+        _flash_attn_varlen_backward as _fa2_varlen_backward, )
 
 
 def flash_attn_no_pad(
@@ -191,3 +199,473 @@ def flash_attn_varlen_qk_no_pad(
         h=nheads,
     )
     return output
+
+
+# ---------------------------------------------------------------------------
+# torch.compile traceability + register_autograd parity for the masked /
+# varlen attention paths.
+#
+# Wraps the two entry points `FlashAttentionImpl.forward` calls
+# (`flash_attn_no_pad`, `flash_attn_varlen_qk_no_pad`) as
+# `torch.library.custom_op`s so dynamo sees one traceable node — the
+# internal unpad / pad bookkeeping (data-dependent `nnz` shapes) runs
+# eager inside the op, and the op's outputs are the statically-shaped
+# padded tensors. This mirrors the FA2 default-path wrapper in
+# `fastvideo/attention/backends/flash_attn.py`.
+#
+# Autograd: on FA2 we register a real backward (`register_autograd`)
+# that calls FA2's `_flash_attn_varlen_backward` on the unpadded form
+# — re-unpadding the saved padded tensors using the saved mask. The
+# `softmax_lse` from the varlen forward is naturally unpadded
+# (`[nheads, total_q]`); we pad it to `[batch, nheads, seqlen]` on
+# the way out (statically shaped) and re-unpad in backward. So
+# training backprops *through* the op (no graph break on the training
+# path either).
+#
+# FA3 / FA4 keep the autograd carve-out pattern from PR #1373: the
+# custom op has forward + fake only, and `*_compilable` falls back to
+# the original autograd.Function for grad-enabled calls. Those legs
+# are gated on Hopper-class / Blackwell-class boxes for backward
+# validation and ship as separate follow-ups.
+
+if _FA_VARLEN_VERSION == "2":
+    # ---------- masked self-attention: flash_attn_no_pad (FA2) ----------
+
+    @torch.library.custom_op(
+        "fastvideo::_flash_attn_no_pad_forward",
+        mutates_args=(),
+        device_types="cuda",
+    )
+    def _flash_attn_no_pad_forward(
+        qkv: torch.Tensor,
+        key_padding_mask: torch.Tensor,
+        causal: bool,
+        dropout_p: float,
+        softmax_scale: float | None,
+        deterministic: bool,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        b, s, _three, h, d = qkv.shape
+        x = rearrange(qkv, "b s three h d -> b s (three h d)")
+        x_unpad, indices, cu_seqlens, max_s, _ = unpad_input(x, key_padding_mask)
+        x_unpad = rearrange(x_unpad, "nnz (three h d) -> nnz three h d", three=3, h=h)
+        out_unpad, lse_unpad, _ = flash_attn_varlen_qkvpacked_func(x_unpad,
+                                                                   cu_seqlens,
+                                                                   max_s,
+                                                                   dropout_p,
+                                                                   softmax_scale=softmax_scale,
+                                                                   causal=causal,
+                                                                   deterministic=deterministic,
+                                                                   return_attn_probs=True)
+        # Pad out: [nnz, h, d] -> [b, s, h, d]
+        out_padded = rearrange(pad_input(rearrange(out_unpad, "nnz h d -> nnz (h d)"), indices, b, s),
+                               "b s (h d) -> b s h d",
+                               h=h)
+        # Pad lse: FA2 varlen returns [nheads, total_q]. Transpose to [total_q,
+        # nheads], pad to [b, s, nheads], permute to [b, nheads, s] — statically
+        # shaped so register_fake matches.
+        lse_padded = pad_input(lse_unpad.t().contiguous(), indices, b, s).permute(0, 2, 1).contiguous()
+        return out_padded, lse_padded
+
+    @torch.library.register_fake("fastvideo::_flash_attn_no_pad_forward")
+    def _flash_attn_no_pad_forward_fake(
+        qkv: torch.Tensor,
+        key_padding_mask: torch.Tensor,
+        causal: bool,
+        dropout_p: float,
+        softmax_scale: float | None,
+        deterministic: bool,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        del key_padding_mask, causal, dropout_p, softmax_scale, deterministic
+        b, s, _three, h, d = qkv.shape
+        out = qkv.new_empty(b, s, h, d)
+        lse = qkv.new_empty(b, h, s, dtype=torch.float32)
+        return out, lse
+
+    def _flash_attn_no_pad_setup_context(ctx, inputs, output):
+        qkv, key_padding_mask, causal, dropout_p, softmax_scale, deterministic = inputs
+        out, lse = output
+        ctx.save_for_backward(qkv, out, lse, key_padding_mask)
+        # Auxiliary output, not differentiable — see default-path note.
+        ctx.mark_non_differentiable(lse)
+        # FA2's varlen backward requires a concrete float for softmax_scale.
+        if softmax_scale is None:
+            softmax_scale = qkv.shape[-1]**-0.5  # head_dim from qkv's last dim
+        ctx.softmax_scale = softmax_scale
+        ctx.causal = causal
+        ctx.dropout_p = dropout_p
+        ctx.deterministic = deterministic
+
+    def _flash_attn_no_pad_backward(ctx, grad_out, grad_lse):
+        # lse is saved-for-backward, not differentiated.
+        del grad_lse
+        qkv, out_padded, lse_padded, key_padding_mask = ctx.saved_tensors
+        b, s, _three, h, d = qkv.shape
+
+        # One `unpad_input` call (on qkv) gives us indices + cu_seqlens + max_s;
+        # reuse those for out / dout / lse below via direct indexing instead
+        # of redundant `unpad_input` calls (each of which would re-run
+        # `nonzero` + `cumsum` + a `.max().item()` GPU→CPU sync).
+        x = rearrange(qkv, "b s three h d -> b s (three h d)")
+        x_unpad, indices, cu_seqlens, max_s, _ = unpad_input(x, key_padding_mask)
+        x_unpad = rearrange(x_unpad, "nnz (three h d) -> nnz three h d", three=3, h=h)
+        q_unpad, k_unpad, v_unpad = (t.contiguous() for t in x_unpad.unbind(dim=1))
+
+        # Direct-index variants reuse `indices` (computed above).
+        out_unpad = out_padded.flatten(0, 1)[indices].view(-1, h, d).contiguous()
+        dout_unpad = grad_out.flatten(0, 1)[indices].view(-1, h, d).contiguous()
+        # lse_padded [b, h, s] -> [b, s, h] -> [nnz, h] -> [h, nnz].
+        lse_unpad = lse_padded.permute(0, 2, 1).contiguous().flatten(0, 1)[indices].t().contiguous()
+
+        dq_unpad = torch.empty_like(q_unpad)
+        dk_unpad = torch.empty_like(k_unpad)
+        dv_unpad = torch.empty_like(v_unpad)
+        _fa2_varlen_backward(
+            dout_unpad,
+            q_unpad,
+            k_unpad,
+            v_unpad,
+            out_unpad,
+            lse_unpad,
+            dq_unpad,
+            dk_unpad,
+            dv_unpad,
+            cu_seqlens_q=cu_seqlens,
+            cu_seqlens_k=cu_seqlens,
+            max_seqlen_q=max_s,
+            max_seqlen_k=max_s,
+            dropout_p=ctx.dropout_p,
+            softmax_scale=ctx.softmax_scale,
+            causal=ctx.causal,
+            window_size_left=-1,
+            window_size_right=-1,
+            softcap=0.0,
+            alibi_slopes=None,
+            deterministic=ctx.deterministic,
+            rng_state=None,
+        )
+
+        # Re-pad each grad and stack into dqkv.
+        def _repad(dt_unpad: torch.Tensor) -> torch.Tensor:
+            padded = pad_input(rearrange(dt_unpad, "nnz h d -> nnz (h d)"), indices, b, s)
+            return rearrange(padded, "b s (h d) -> b s h d", h=h)
+
+        dqkv = torch.stack([_repad(dq_unpad), _repad(dk_unpad), _repad(dv_unpad)], dim=2)
+        # 6 inputs total: qkv, key_padding_mask, causal, dropout_p, softmax_scale, deterministic.
+        return dqkv, None, None, None, None, None
+
+    torch.library.register_autograd(
+        "fastvideo::_flash_attn_no_pad_forward",
+        _flash_attn_no_pad_backward,
+        setup_context=_flash_attn_no_pad_setup_context,
+    )
+
+    # ---------- cross-attention: flash_attn_varlen_qk_no_pad (FA2) ----------
+
+    @torch.library.custom_op(
+        "fastvideo::_flash_attn_varlen_qk_no_pad_forward",
+        mutates_args=(),
+        device_types="cuda",
+    )
+    def _flash_attn_varlen_qk_no_pad_forward(
+        query: torch.Tensor,
+        key: torch.Tensor,
+        value: torch.Tensor,
+        query_padding_mask: torch.Tensor,
+        key_padding_mask: torch.Tensor,
+        causal: bool,
+        dropout_p: float,
+        softmax_scale: float | None,
+        deterministic: bool,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        b, sq, h, d = query.shape
+        q_unpad, q_indices, cu_seqlens_q, max_seqlen_q, _ = unpad_input(rearrange(query, "b s h d -> b s (h d)"),
+                                                                        query_padding_mask)
+        k_unpad, _, cu_seqlens_k, max_seqlen_k, _ = unpad_input(rearrange(key, "b s h d -> b s (h d)"),
+                                                                key_padding_mask)
+        v_unpad, _, _, _, _ = unpad_input(rearrange(value, "b s h d -> b s (h d)"), key_padding_mask)
+        q_unpad = rearrange(q_unpad, "nnz (h d) -> nnz h d", h=h)
+        k_unpad = rearrange(k_unpad, "nnz (h d) -> nnz h d", h=h)
+        v_unpad = rearrange(v_unpad, "nnz (h d) -> nnz h d", h=h)
+        out_unpad, lse_unpad, _ = flash_attn_varlen_func_impl(q_unpad,
+                                                              k_unpad,
+                                                              v_unpad,
+                                                              cu_seqlens_q,
+                                                              cu_seqlens_k,
+                                                              max_seqlen_q,
+                                                              max_seqlen_k,
+                                                              dropout_p=dropout_p,
+                                                              softmax_scale=softmax_scale,
+                                                              causal=causal,
+                                                              deterministic=deterministic,
+                                                              return_attn_probs=True)
+        # Pad out: [nnz_q, h, d] -> [b, sq, h, d]
+        out_padded = rearrange(pad_input(rearrange(out_unpad, "nnz h d -> nnz (h d)"), q_indices, b, sq),
+                               "b s (h d) -> b s h d",
+                               h=h)
+        # Pad lse: [h, nnz_q] -> [b, h, sq]
+        lse_padded = pad_input(lse_unpad.t().contiguous(), q_indices, b, sq).permute(0, 2, 1).contiguous()
+        return out_padded, lse_padded
+
+    @torch.library.register_fake("fastvideo::_flash_attn_varlen_qk_no_pad_forward")
+    def _flash_attn_varlen_qk_no_pad_forward_fake(
+        query: torch.Tensor,
+        key: torch.Tensor,
+        value: torch.Tensor,
+        query_padding_mask: torch.Tensor,
+        key_padding_mask: torch.Tensor,
+        causal: bool,
+        dropout_p: float,
+        softmax_scale: float | None,
+        deterministic: bool,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        del key, query_padding_mask, key_padding_mask
+        del causal, dropout_p, softmax_scale, deterministic
+        b, sq, h, _ = query.shape
+        # `out`'s head_dim comes from value (d_v), matching the real forward's
+        # out_padded ([b, sq, h, d_v]); it can differ from query's d_q.
+        out = query.new_empty(b, sq, h, value.shape[-1])
+        lse = query.new_empty(b, h, sq, dtype=torch.float32)
+        return out, lse
+
+    def _flash_attn_varlen_qk_no_pad_setup_context(ctx, inputs, output):
+        (query, key, value, query_padding_mask, key_padding_mask, causal, dropout_p, softmax_scale,
+         deterministic) = inputs
+        out, lse = output
+        ctx.save_for_backward(query, key, value, out, lse, query_padding_mask, key_padding_mask)
+        # Auxiliary output, not differentiable — see default-path note.
+        ctx.mark_non_differentiable(lse)
+        if softmax_scale is None:
+            softmax_scale = query.shape[-1]**-0.5
+        ctx.softmax_scale = softmax_scale
+        ctx.causal = causal
+        ctx.dropout_p = dropout_p
+        ctx.deterministic = deterministic
+
+    def _flash_attn_varlen_qk_no_pad_backward(ctx, grad_out, grad_lse):
+        del grad_lse
+        (query, key, value, out_padded, lse_padded, query_padding_mask, key_padding_mask) = ctx.saved_tensors
+        b, sq, h, d = query.shape
+        sk = key.shape[1]
+
+        # One `unpad_input` call per distinct mask; reuse the returned
+        # indices via direct indexing for everything else that shares
+        # the same mask (v with k_mask; out/dout/lse with q_mask; the
+        # final repad of dk/dv also reuses k_indices). Avoids ~4
+        # redundant `unpad_input` calls + their GPU→CPU `.max().item()`
+        # syncs.
+        q_unpad, q_indices, cu_seqlens_q, max_seqlen_q, _ = unpad_input(rearrange(query, "b s h d -> b s (h d)"),
+                                                                        query_padding_mask)
+        k_unpad, k_indices, cu_seqlens_k, max_seqlen_k, _ = unpad_input(rearrange(key, "b s h d -> b s (h d)"),
+                                                                        key_padding_mask)
+        q_unpad = rearrange(q_unpad, "nnz (h d) -> nnz h d", h=h).contiguous()
+        k_unpad = rearrange(k_unpad, "nnz (h d) -> nnz h d", h=h).contiguous()
+        v_unpad = value.flatten(0, 1)[k_indices].view(-1, h, d).contiguous()
+
+        # out / dout / lse follow q's shape, so index with q_indices.
+        out_unpad = out_padded.flatten(0, 1)[q_indices].view(-1, h, d).contiguous()
+        dout_unpad = grad_out.flatten(0, 1)[q_indices].view(-1, h, d).contiguous()
+        lse_unpad = lse_padded.permute(0, 2, 1).contiguous().flatten(0, 1)[q_indices].t().contiguous()
+
+        dq_unpad = torch.empty_like(q_unpad)
+        dk_unpad = torch.empty_like(k_unpad)
+        dv_unpad = torch.empty_like(v_unpad)
+        _fa2_varlen_backward(
+            dout_unpad,
+            q_unpad,
+            k_unpad,
+            v_unpad,
+            out_unpad,
+            lse_unpad,
+            dq_unpad,
+            dk_unpad,
+            dv_unpad,
+            cu_seqlens_q=cu_seqlens_q,
+            cu_seqlens_k=cu_seqlens_k,
+            max_seqlen_q=max_seqlen_q,
+            max_seqlen_k=max_seqlen_k,
+            dropout_p=ctx.dropout_p,
+            softmax_scale=ctx.softmax_scale,
+            causal=ctx.causal,
+            window_size_left=-1,
+            window_size_right=-1,
+            softcap=0.0,
+            alibi_slopes=None,
+            deterministic=ctx.deterministic,
+            rng_state=None,
+        )
+
+        # k_indices is already available from the unpad_input above —
+        # no need to recompute it for the dk/dv repad.
+        def _repad(dt_unpad: torch.Tensor, indices: torch.Tensor, batch: int, seqlen: int) -> torch.Tensor:
+            padded = pad_input(rearrange(dt_unpad, "nnz h d -> nnz (h d)"), indices, batch, seqlen)
+            return rearrange(padded, "b s (h d) -> b s h d", h=h)
+
+        dq_padded = _repad(dq_unpad, q_indices, b, sq)
+        dk_padded = _repad(dk_unpad, k_indices, b, sk)
+        dv_padded = _repad(dv_unpad, k_indices, b, sk)
+        # 9 inputs total: query, key, value, q_mask, k_mask, causal, dropout_p,
+        # softmax_scale, deterministic.
+        return dq_padded, dk_padded, dv_padded, None, None, None, None, None, None
+
+    torch.library.register_autograd(
+        "fastvideo::_flash_attn_varlen_qk_no_pad_forward",
+        _flash_attn_varlen_qk_no_pad_backward,
+        setup_context=_flash_attn_varlen_qk_no_pad_setup_context,
+    )
+
+    # ---------- public dispatchers (FA2: autograd flows through the op) -----
+
+
+    def flash_attn_no_pad_compilable(qkv,
+                                     key_padding_mask,
+                                     causal=False,
+                                     dropout_p=0.0,
+                                     softmax_scale=None,
+                                     deterministic=False):
+        """dynamo-traceable wrapper around ``flash_attn_no_pad`` (registered op,
+        full register_autograd on FA2 — both inference and training go through
+        the op, no graph break on either)."""
+        out, _ = torch.ops.fastvideo._flash_attn_no_pad_forward(qkv, key_padding_mask, causal, dropout_p, softmax_scale,
+                                                                deterministic)
+        return out
+
+    def flash_attn_varlen_qk_no_pad_compilable(query,
+                                               key,
+                                               value,
+                                               query_padding_mask,
+                                               key_padding_mask,
+                                               causal=False,
+                                               dropout_p=0.0,
+                                               softmax_scale=None,
+                                               deterministic=False):
+        """dynamo-traceable wrapper around ``flash_attn_varlen_qk_no_pad`` (registered
+        op, full register_autograd on FA2)."""
+        out, _ = torch.ops.fastvideo._flash_attn_varlen_qk_no_pad_forward(query, key, value, query_padding_mask,
+                                                                          key_padding_mask, causal, dropout_p,
+                                                                          softmax_scale, deterministic)
+        return out
+
+else:
+    # ---------- FA3 / FA4: carve-out (forward+fake only, no real backward) ---
+    # Same pattern as the parked varlen-extension and the FA3 default leg in
+    # `fastvideo/attention/backends/flash_attn.py`. Real backward for these
+    # versions is a follow-up gated on Hopper / Blackwell box validation.
+
+    @torch.library.custom_op(
+        "fastvideo::_flash_attn_no_pad_forward",
+        mutates_args=(),
+        device_types="cuda",
+    )
+    def _flash_attn_no_pad_forward(
+        qkv: torch.Tensor,
+        key_padding_mask: torch.Tensor,
+        causal: bool,
+        dropout_p: float,
+        softmax_scale: float | None,
+        deterministic: bool,
+    ) -> torch.Tensor:
+        return flash_attn_no_pad(  # type: ignore[no-untyped-call]
+            qkv,
+            key_padding_mask,
+            causal=causal,
+            dropout_p=dropout_p,
+            softmax_scale=softmax_scale,
+            deterministic=deterministic)
+
+    @torch.library.register_fake("fastvideo::_flash_attn_no_pad_forward")
+    def _flash_attn_no_pad_forward_fake(
+        qkv: torch.Tensor,
+        key_padding_mask: torch.Tensor,
+        causal: bool,
+        dropout_p: float,
+        softmax_scale: float | None,
+        deterministic: bool,
+    ) -> torch.Tensor:
+        del key_padding_mask, causal, dropout_p, softmax_scale, deterministic
+        b, s, _three, h, d = qkv.shape
+        return qkv.new_empty(b, s, h, d)
+
+    @torch.library.custom_op(
+        "fastvideo::_flash_attn_varlen_qk_no_pad_forward",
+        mutates_args=(),
+        device_types="cuda",
+    )
+    def _flash_attn_varlen_qk_no_pad_forward(
+        query: torch.Tensor,
+        key: torch.Tensor,
+        value: torch.Tensor,
+        query_padding_mask: torch.Tensor,
+        key_padding_mask: torch.Tensor,
+        causal: bool,
+        dropout_p: float,
+        softmax_scale: float | None,
+        deterministic: bool,
+    ) -> torch.Tensor:
+        return flash_attn_varlen_qk_no_pad(  # type: ignore[no-untyped-call]
+            query,
+            key,
+            value,
+            query_padding_mask=query_padding_mask,
+            key_padding_mask=key_padding_mask,
+            causal=causal,
+            dropout_p=dropout_p,
+            softmax_scale=softmax_scale,
+            deterministic=deterministic)
+
+    @torch.library.register_fake("fastvideo::_flash_attn_varlen_qk_no_pad_forward")
+    def _flash_attn_varlen_qk_no_pad_forward_fake(
+        query: torch.Tensor,
+        key: torch.Tensor,
+        value: torch.Tensor,
+        query_padding_mask: torch.Tensor,
+        key_padding_mask: torch.Tensor,
+        causal: bool,
+        dropout_p: float,
+        softmax_scale: float | None,
+        deterministic: bool,
+    ) -> torch.Tensor:
+        del key, query_padding_mask, key_padding_mask
+        del causal, dropout_p, softmax_scale, deterministic
+        b, sq, h, _ = query.shape
+        # `out`'s head_dim comes from value (d_v), matching the real forward's
+        # output ([b, sq, h, d_v]); it can differ from query's d_q.
+        return query.new_empty(b, sq, h, value.shape[-1])
+
+    def flash_attn_no_pad_compilable(qkv,
+                                     key_padding_mask,
+                                     causal=False,
+                                     dropout_p=0.0,
+                                     softmax_scale=None,
+                                     deterministic=False):
+        if torch.is_grad_enabled() and qkv.requires_grad:
+            return flash_attn_no_pad(qkv,
+                                     key_padding_mask,
+                                     causal=causal,
+                                     dropout_p=dropout_p,
+                                     softmax_scale=softmax_scale,
+                                     deterministic=deterministic)
+        return torch.ops.fastvideo._flash_attn_no_pad_forward(qkv, key_padding_mask, causal, dropout_p, softmax_scale,
+                                                              deterministic)
+
+    def flash_attn_varlen_qk_no_pad_compilable(query,
+                                               key,
+                                               value,
+                                               query_padding_mask,
+                                               key_padding_mask,
+                                               causal=False,
+                                               dropout_p=0.0,
+                                               softmax_scale=None,
+                                               deterministic=False):
+        if torch.is_grad_enabled() and (query.requires_grad or key.requires_grad or value.requires_grad):
+            return flash_attn_varlen_qk_no_pad(query,
+                                               key,
+                                               value,
+                                               query_padding_mask=query_padding_mask,
+                                               key_padding_mask=key_padding_mask,
+                                               causal=causal,
+                                               dropout_p=dropout_p,
+                                               softmax_scale=softmax_scale,
+                                               deterministic=deterministic)
+        return torch.ops.fastvideo._flash_attn_varlen_qk_no_pad_forward(query, key, value, query_padding_mask,
+                                                                        key_padding_mask, causal, dropout_p,
+                                                                        softmax_scale, deterministic)

fastvideo/batching/__init__.py

@@ -1,26 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-"""Dynamic generation batching helpers."""
-
-from fastvideo.batching.admission import (
-    AdmissionLimit,
-    BatchAdmissionController,
-    BatchingRule,
-    load_batching_config,
-)
-from fastvideo.batching.signature import (
-    BatchCompatibility,
-    can_dynamic_batch,
-    dynamic_batch_signature,
-    resolution_key,
-)
-
-__all__ = [
-    "AdmissionLimit",
-    "BatchAdmissionController",
-    "BatchCompatibility",
-    "BatchingRule",
-    "can_dynamic_batch",
-    "dynamic_batch_signature",
-    "load_batching_config",
-    "resolution_key",
-]

fastvideo/batching/admission.py

@@ -1,297 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-from __future__ import annotations
-
-import json
-import os
-from dataclasses import dataclass
-from difflib import get_close_matches
-from typing import Any
-
-from fastvideo.batching.signature import resolution_key
-from fastvideo.fastvideo_args import FastVideoArgs
-from fastvideo.logger import init_logger
-
-logger = init_logger(__name__)
-
-_BYTES_PER_GB = 1024**3
-_BATCHING_RULE_KEYS = frozenset({
-    "model",
-    "model_contains",
-    "resolution",
-    "device_memory_gb_min",
-    "device_memory_gb_max",
-    "offload",
-    "max_batch_size",
-    "max_cost",
-    "calibration",
-})
-
-
-@dataclass(frozen=True)
-class AdmissionLimit:
-    max_batch_size: int
-    max_cost: float | None = None
-    cap_reason: str | None = None
-
-    def reject_reason(self, *, batch_size: int, batch_cost: float) -> str | None:
-        if batch_size > self.max_batch_size:
-            return self.cap_reason or f"config_cap:{self.max_batch_size}"
-        if self.max_cost is not None and batch_cost > self.max_cost:
-            return f"cost_budget:{batch_cost:.0f}>{self.max_cost:.0f}"
-        return None
-
-    def stop_reason_for_next_cost(self, next_batch_cost: float) -> str | None:
-        if self.max_cost is not None and next_batch_cost > self.max_cost:
-            return f"cost_budget_next:{next_batch_cost:.0f}>{self.max_cost:.0f}"
-        return None
-
-
-@dataclass(frozen=True)
-class BatchingRule:
-    model: str | None = None
-    model_contains: str | None = None
-    resolution: str | None = None
-    device_memory_gb_min: float | None = None
-    device_memory_gb_max: float | None = None
-    offload: bool | None = None
-    max_batch_size: int = 1
-    max_cost: float | None = None
-    source: str = "user"
-
-    @classmethod
-    def from_dict(cls, data: dict[str, Any], *, source: str) -> BatchingRule:
-        if not isinstance(data, dict):
-            raise ValueError(f"batching config rule from {source} must be an object, got {type(data).__name__}")
-        _validate_rule_keys(data, source=source)
-        if "max_batch_size" not in data:
-            raise ValueError("batching config rule requires max_batch_size")
-
-        rule = cls(
-            model=_optional_str(data.get("model")),
-            model_contains=_optional_str(data.get("model_contains")),
-            resolution=_optional_str(data.get("resolution")),
-            device_memory_gb_min=_optional_float(data.get("device_memory_gb_min")),
-            device_memory_gb_max=_optional_float(data.get("device_memory_gb_max")),
-            offload=_optional_bool(data.get("offload")),
-            max_batch_size=int(data["max_batch_size"]),
-            max_cost=_optional_float(data.get("max_cost")),
-            source=source,
-        )
-        rule.validate()
-        return rule
-
-    def validate(self) -> None:
-        if self.model is not None and self.model_contains is not None:
-            raise ValueError("batching config rule cannot set both model and model_contains")
-        if self.model is None and self.model_contains is None:
-            raise ValueError("batching config rule requires model or model_contains")
-        if self.max_batch_size < 1:
-            raise ValueError("batching config rule max_batch_size must be >= 1")
-        if self.max_cost is not None and self.max_cost <= 0.0:
-            raise ValueError("batching config rule max_cost must be > 0")
-        if (self.device_memory_gb_min is not None and self.device_memory_gb_max is not None
-                and self.device_memory_gb_min > self.device_memory_gb_max):
-            raise ValueError("batching config rule device_memory_gb_min must be <= device_memory_gb_max")
-
-    def matches(
-        self,
-        *,
-        model_path: str,
-        resolution: str | None,
-        device_memory_gb: float | None,
-        offload: bool,
-    ) -> bool:
-        if self.model is not None and self.model != model_path:
-            return False
-        if self.model_contains is not None and self.model_contains not in model_path:
-            return False
-        if self.resolution not in (None, "*") and self.resolution != resolution:
-            return False
-        if self.offload is not None and self.offload != offload:
-            return False
-        if device_memory_gb is None:
-            return True
-        if self.device_memory_gb_min is not None and device_memory_gb < self.device_memory_gb_min:
-            return False
-        return not (self.device_memory_gb_max is not None and device_memory_gb > self.device_memory_gb_max)
-
-
-class BatchAdmissionController:
-
-    def __init__(self, fastvideo_args: FastVideoArgs, *, gpu_id: int = 0):
-        self._mode = fastvideo_args.batching_mode
-        self._user_max_batch_size = max(1, int(fastvideo_args.batching_max_size))
-        self._model_path = fastvideo_args.model_path
-        self._offload = bool(fastvideo_args.dit_cpu_offload or fastvideo_args.dit_layerwise_offload)
-        self._device_memory_gb = self._get_device_memory_gb(gpu_id)
-        self._rules = load_batching_config(fastvideo_args.batching_config)
-        self._pipeline_config = fastvideo_args.pipeline_config
-
-        if self.enabled:
-            logger.info(
-                "Batch admission enabled: user_max=%d, device_memory=%.1fGiB, rules=%d",
-                self._user_max_batch_size,
-                self._device_memory_gb or 0.0,
-                len(self._rules),
-            )
-
-    @property
-    def enabled(self) -> bool:
-        return self._mode == "dynamic" and self._user_max_batch_size > 1
-
-    def reject_reason_for_candidate(self, current_requests: list[Any], candidate_request: Any) -> str | None:
-        if not self.enabled:
-            return None
-        proposed = current_requests + [candidate_request]
-        limit = self.limit_for(proposed[0])
-        return limit.reject_reason(
-            batch_size=len(proposed),
-            batch_cost=self.estimate_batch_cost(proposed),
-        )
-
-    def batch_is_full(self, requests: list[Any]) -> bool:
-        if not self.enabled or not requests:
-            return len(requests) >= self._user_max_batch_size
-
-        limit = self.limit_for(requests[0])
-        if len(requests) >= limit.max_batch_size:
-            return True
-
-        next_cost = self.estimate_batch_cost(requests + [requests[0]])
-        return limit.max_cost is not None and next_cost > limit.max_cost
-
-    def limit_reason_for_batch(self, requests: list[Any]) -> str | None:
-        if not self.enabled or not requests:
-            return None
-
-        limit = self.limit_for(requests[0])
-        if len(requests) >= limit.max_batch_size:
-            return limit.cap_reason or f"config_cap:{limit.max_batch_size}"
-
-        next_cost = self.estimate_batch_cost(requests + [requests[0]])
-        return limit.stop_reason_for_next_cost(next_cost)
-
-    def max_admissible_batch_size(self, request: Any) -> int:
-        return self.limit_for(request).max_batch_size
-
-    def limit_for(self, request: Any) -> AdmissionLimit:
-        rules = self._matching_rules(request)
-        if not rules:
-            return AdmissionLimit(max_batch_size=self._user_max_batch_size)
-
-        config_cap = min(rule.max_batch_size for rule in rules)
-        max_batch_size = min(self._user_max_batch_size, config_cap)
-        cap_reason = f"config_cap:{max_batch_size}" if max_batch_size < self._user_max_batch_size else None
-        costs = [rule.max_cost for rule in rules if rule.max_cost is not None]
-        return AdmissionLimit(
-            max_batch_size=max(1, max_batch_size),
-            max_cost=min(costs) if costs else None,
-            cap_reason=cap_reason,
-        )
-
-    def estimate_batch_cost(self, requests: list[Any]) -> float:
-        return sum(float(self._pipeline_config.estimate_request_cost(request)) for request in requests)
-
-    def _matching_rules(self, request: Any) -> list[BatchingRule]:
-        return [
-            rule for rule in self._rules if rule.matches(
-                model_path=self._model_path,
-                resolution=resolution_key(request),
-                device_memory_gb=self._device_memory_gb,
-                offload=self._offload,
-            )
-        ]
-
-    @staticmethod
-    def _get_device_memory_gb(gpu_id: int) -> float | None:
-        try:
-            from fastvideo.platforms import current_platform
-
-            return current_platform.get_device_total_memory(gpu_id) / _BYTES_PER_GB
-        except Exception:
-            return None
-
-
-def load_batching_config(path: str | None) -> list[BatchingRule]:
-    if path is None:
-        return []
-
-    with open(path, encoding="utf-8") as f:
-        payload = json.load(f)
-
-    source = os.path.abspath(path)
-    entries = _config_entries(payload)
-    rules = [BatchingRule.from_dict(entry, source=source) for entry in entries]
-    if not rules:
-        raise ValueError(f"batching config {source} does not contain any rules")
-    return rules
-
-
-def _config_entries(payload: Any) -> list[dict[str, Any]]:
-    if isinstance(payload, dict) and payload.get("schema_version") not in (None, 1):
-        raise ValueError("batching config schema_version must be 1")
-    if isinstance(payload, dict) and isinstance(payload.get("rules"), list):
-        return payload["rules"]
-    if isinstance(payload, list):
-        return payload
-    if isinstance(payload, dict):
-        entries: list[dict[str, Any]] = []
-        for key, value in payload.items():
-            if key == "schema_version" or not isinstance(value, dict):
-                continue
-            model, _sep, resolution = key.partition("|")
-            entry = dict(value)
-            if model:
-                entry.setdefault("model", model)
-            if resolution:
-                entry.setdefault("resolution", resolution)
-            entries.append(entry)
-        return entries
-    raise ValueError("batching config must be a {'schema_version': 1, 'rules': [...]} object, "
-                     "a list of rules, or a mapping keyed by model|resolution")
-
-
-def _validate_rule_keys(data: dict[str, Any], *, source: str) -> None:
-    unknown = sorted(set(data) - _BATCHING_RULE_KEYS)
-    if not unknown:
-        return
-
-    hints = []
-    for key in unknown:
-        matches = get_close_matches(key, _BATCHING_RULE_KEYS, n=1)
-        if matches:
-            hints.append(f"{key!r} (did you mean {matches[0]!r}?)")
-        else:
-            hints.append(repr(key))
-    raise ValueError(f"batching config rule from {source} contains unknown key(s): {', '.join(hints)}")
-
-
-def _optional_str(value: Any) -> str | None:
-    if value is None:
-        return None
-    return str(value)
-
-
-def _optional_float(value: Any) -> float | None:
-    if value is None:
-        return None
-    return float(value)
-
-
-def _optional_bool(value: Any) -> bool | None:
-    if value is None:
-        return None
-    if isinstance(value, bool):
-        return value
-    if isinstance(value, int | float):
-        if value == 1.0:
-            return True
-        if value == 0.0:
-            return False
-    if isinstance(value, str):
-        lowered = value.strip().lower()
-        if lowered in ("1", "true", "yes", "y", "on"):
-            return True
-        if lowered in ("0", "false", "no", "n", "off"):
-            return False
-    raise ValueError(f"cannot parse boolean batching config value: {value!r}")

fastvideo/batching/signature.py

@@ -1,178 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-from __future__ import annotations
-
-import dataclasses
-from dataclasses import dataclass
-from enum import Enum
-from typing import Any
-
-from fastvideo.api.sampling_param import SamplingParam
-
-_SIGNATURE_EXCLUDED_FIELDS = frozenset({
-    "prompt",
-    "prompt_path",
-    "output_path",
-    "output_video_name",
-    "seed",
-    "save_video",
-    "return_frames",
-})
-
-_UNSUPPORTED_DYNAMIC_BATCH_FIELDS = frozenset({
-    "image_path",
-    "pil_image",
-    "video_path",
-    "mouse_cond",
-    "keyboard_cond",
-    "grid_sizes",
-    "pose",
-    "camera_states",
-    "camera_trajectory",
-    "action_list",
-    "action_speed_list",
-    "gt_latents",
-    "conditioning_mask",
-    "c2ws_plucker_emb",
-    "refine_from",
-    "stage1_video",
-    "trajectory_type",
-    "movement_distance",
-    "camera_rotation",
-    "ltx2_images",
-    "ltx2_conditioning_latent_stage1",
-    "ltx2_conditioning_latent_stage2",
-    "ltx2_video_conditions",
-    "init_audio",
-    "inpaint_audio",
-    "inpaint_mask",
-    "continuation_state",
-})
-
-_UNSUPPORTED_EXTRA_KEYS = frozenset({
-    "ltx2_audio_latents",
-    "ltx2_audio_clean_latent",
-    "ltx2_audio_denoise_mask",
-    "audio_num_frames",
-    "video_position_offset_sec",
-})
-
-
-@dataclass(frozen=True)
-class BatchCompatibility:
-    can_batch: bool
-    reason: str | None = None
-
-
-def resolution_key(request: Any) -> str:
-    height = _first_scalar(getattr(request, "height", None))
-    width = _first_scalar(getattr(request, "width", None))
-    num_frames = _first_scalar(getattr(request, "num_frames", None))
-    return f"{height}x{width}x{num_frames}"
-
-
-def dynamic_batch_signature(
-    request: SamplingParam,
-    *,
-    extra: dict[str, Any] | None = None,
-) -> tuple[tuple[str, Any], ...]:
-    """Build a hashable compatibility signature for a generation request."""
-    signature_items: list[tuple[str, Any]] = []
-    for field in dataclasses.fields(request):
-        if field.name in _SIGNATURE_EXCLUDED_FIELDS:
-            continue
-        signature_items.append((field.name, _freeze_signature_value(getattr(request, field.name, None))))
-    if extra:
-        signature_items.append(("extra", _freeze_signature_value(extra)))
-    return tuple(signature_items)
-
-
-def can_dynamic_batch(
-    base: SamplingParam,
-    candidate: SamplingParam,
-    *,
-    base_extra: dict[str, Any] | None = None,
-    candidate_extra: dict[str, Any] | None = None,
-) -> BatchCompatibility:
-    """Return whether two FastVideo generation requests can be merged."""
-    base_ready = _request_is_batchable(base, extra=base_extra)
-    if not base_ready.can_batch:
-        return base_ready
-    candidate_ready = _request_is_batchable(candidate, extra=candidate_extra)
-    if not candidate_ready.can_batch:
-        return candidate_ready
-
-    base_sig = dynamic_batch_signature(base, extra=base_extra)
-    candidate_sig = dynamic_batch_signature(candidate, extra=candidate_extra)
-    if base_sig == candidate_sig:
-        return BatchCompatibility(can_batch=True)
-
-    mismatch = _first_mismatch(base_sig, candidate_sig)
-    return BatchCompatibility(can_batch=False, reason=mismatch or "signature_mismatch")
-
-
-def _request_is_batchable(
-    request: SamplingParam,
-    *,
-    extra: dict[str, Any] | None = None,
-) -> BatchCompatibility:
-    if not isinstance(request.prompt, str):
-        return BatchCompatibility(can_batch=False, reason="prompt_type")
-    if request.prompt_path is not None:
-        return BatchCompatibility(can_batch=False, reason="prompt_path")
-    if request.num_videos_per_prompt != 1:
-        return BatchCompatibility(can_batch=False, reason="num_videos_per_prompt")
-    if request.return_continuation_state:
-        return BatchCompatibility(can_batch=False, reason="return_continuation_state")
-
-    for name in _UNSUPPORTED_DYNAMIC_BATCH_FIELDS:
-        value = getattr(request, name, None)
-        if _is_present(value):
-            return BatchCompatibility(can_batch=False, reason=name)
-
-    if extra:
-        unsupported = sorted(set(extra) & _UNSUPPORTED_EXTRA_KEYS)
-        if unsupported:
-            return BatchCompatibility(can_batch=False, reason=f"extra.{unsupported[0]}")
-
-    return BatchCompatibility(can_batch=True)
-
-
-def _freeze_signature_value(value: Any) -> Any:
-    if isinstance(value, str | int | float | bool | type(None)):
-        return value
-    if isinstance(value, Enum):
-        return value.value
-    if isinstance(value, dict):
-        return tuple(
-            (str(key), _freeze_signature_value(item)) for key, item in sorted(value.items(), key=lambda kv: str(kv[0])))
-    if isinstance(value, list | tuple):
-        return tuple(_freeze_signature_value(item) for item in value)
-    return repr(value)
-
-
-def _is_present(value: Any) -> bool:
-    if value is None:
-        return False
-    if value is False:
-        return False
-    return not (isinstance(value, list | tuple | dict | set) and not value)
-
-
-def _first_scalar(value: Any) -> Any:
-    if isinstance(value, list | tuple):
-        return value[0] if value else None
-    return value
-
-
-def _first_mismatch(
-    base_sig: tuple[tuple[str, Any], ...],
-    candidate_sig: tuple[tuple[str, Any], ...],
-) -> str | None:
-    if len(base_sig) != len(candidate_sig):
-        return "sampling_params"
-    for (name, base_value), (candidate_name, candidate_value) in zip(base_sig, candidate_sig, strict=True):
-        if name != candidate_name:
-            return "sampling_params"
-        if base_value != candidate_value:
-            return f"sampling_params.{name}"
-    return None

fastvideo/configs/models/dits/__init__.py

@@ -1,6 +1,9 @@
 from fastvideo.configs.models.dits.cosmos import CosmosVideoConfig
 from fastvideo.configs.models.dits.cosmos2_5 import Cosmos25VideoConfig
+from fastvideo.configs.models.dits.dreamx_world import DreamXWorldARConfig, DreamXWorldConfig
+from fastvideo.configs.models.dits.flux import FluxDiTConfig
 from fastvideo.configs.models.dits.flux_2 import Flux2Config
+from fastvideo.configs.models.dits.glm_image import GlmImageDiTConfig
 from fastvideo.configs.models.dits.hunyuangamecraft import HunyuanGameCraftConfig
 from fastvideo.configs.models.dits.hunyuanvideo import HunyuanVideoConfig
 from fastvideo.configs.models.dits.hunyuanvideo15 import HunyuanVideo15Config
@@ -13,7 +16,8 @@ from fastvideo.configs.models.dits.hyworld import HYWorldConfig
 from fastvideo.configs.models.dits.kandinsky5 import Kandinsky5VideoConfig
 
 __all__ = [
-    "HunyuanVideoConfig", "HunyuanVideo15Config", "HunyuanGameCraftConfig", "WanVideoConfig", "CosmosVideoConfig",
-    "Cosmos25VideoConfig", "LongCatVideoConfig", "LTX2VideoConfig", "HYWorldConfig", "Kandinsky5VideoConfig",
-    "MagiHumanVideoConfig", "StableAudioConfig", "Flux2Config"
+    "HunyuanVideoConfig", "HunyuanVideo15Config", "HunyuanGameCraftConfig", "WanVideoConfig", "DreamXWorldConfig",
+    "DreamXWorldARConfig", "CosmosVideoConfig", "Cosmos25VideoConfig", "FluxDiTConfig", "Flux2Config",
+    "LongCatVideoConfig", "LTX2VideoConfig", "HYWorldConfig", "Kandinsky5VideoConfig", "MagiHumanVideoConfig",
+    "StableAudioConfig", "GlmImageDiTConfig"
 ]

fastvideo/configs/models/dits/dreamx_world.py

@@ -0,0 +1,71 @@
+# SPDX-License-Identifier: Apache-2.0
+from dataclasses import dataclass, field
+
+from fastvideo.configs.models.dits.base import DiTArchConfig, DiTConfig
+from fastvideo.configs.models.dits.wanvideo import WanVideoArchConfig
+
+
+@dataclass
+class DreamXWorldArchConfig(WanVideoArchConfig):
+    """DreamX-World DiT config with camera PRoPE control fields."""
+
+    add_control_adapter: bool = True
+    cam_method: str | None = "prope"
+    attn_compress: int = 1
+    cam_self_attn_layers: tuple[int, ...] | None = None
+
+
+@dataclass
+class DreamXWorldConfig(DiTConfig):
+    arch_config: DiTArchConfig = field(default_factory=DreamXWorldArchConfig)
+
+    prefix: str = "Wan"
+
+
+@dataclass
+class DreamXWorldARArchConfig(DreamXWorldArchConfig):
+    """DreamX-World-5B autoregressive causal DiT config."""
+
+    model_type: str = "ti2v"
+    patch_size: tuple[int, int, int] = (1, 2, 2)
+    text_len: int = 512
+    text_dim: int = 4096
+    freq_dim: int = 256
+    attn_compress: int = 4
+    cam_self_attn_layers: tuple[int, ...] | None = tuple(range(30))
+    local_attn_size: int = 12
+    sink_size: int = 3
+    num_frames_per_block: int = 3
+    rope_cache_policy: str = "block_relativistic"
+    # The official AR checkpoint (AMAP-ML/DreamX-World ``model.safetensors``)
+    # already uses FastVideo's native key names and the converter copies the
+    # tensors verbatim, so every rule is an identity. The rules enumerate the
+    # full state-dict surface of ``DreamXWorldARTransformer3DModel`` (norm1 /
+    # norm2 / head.norm are affine-free and have no parameters).
+    param_names_mapping: dict = field(
+        default_factory=lambda: {
+            r"^patch_embedding\.(.*)$": r"patch_embedding.\1",
+            r"^text_embedding\.([02])\.(.*)$": r"text_embedding.\1.\2",
+            r"^time_embedding\.([02])\.(.*)$": r"time_embedding.\1.\2",
+            r"^time_projection\.1\.(.*)$": r"time_projection.1.\1",
+            r"^blocks\.(\d+)\.self_attn\.(q|k|v|o)\.(.*)$": r"blocks.\1.self_attn.\2.\3",
+            r"^blocks\.(\d+)\.self_attn\.norm_(q|k)\.weight$": r"blocks.\1.self_attn.norm_\2.weight",
+            r"^blocks\.(\d+)\.cross_attn\.(q|k|v|o)\.(.*)$": r"blocks.\1.cross_attn.\2.\3",
+            r"^blocks\.(\d+)\.cross_attn\.norm_(q|k)\.weight$": r"blocks.\1.cross_attn.norm_\2.weight",
+            r"^blocks\.(\d+)\.cam_self_attn\.(q_proj|k_proj|v_proj|out_proj)\.(.*)$": r"blocks.\1.cam_self_attn.\2.\3",
+            r"^blocks\.(\d+)\.cam_self_attn\.norm_(q|k)\.weight$": r"blocks.\1.cam_self_attn.norm_\2.weight",
+            r"^blocks\.(\d+)\.norm3\.(.*)$": r"blocks.\1.norm3.\2",
+            r"^blocks\.(\d+)\.ffn\.([02])\.(.*)$": r"blocks.\1.ffn.\2.\3",
+            r"^blocks\.(\d+)\.modulation$": r"blocks.\1.modulation",
+            r"^head\.head\.(.*)$": r"head.head.\1",
+            r"^head\.modulation$": r"head.modulation",
+        })
+    reverse_param_names_mapping: dict = field(default_factory=dict)
+    lora_param_names_mapping: dict = field(default_factory=dict)
+
+
+@dataclass
+class DreamXWorldARConfig(DreamXWorldConfig):
+    arch_config: DiTArchConfig = field(default_factory=DreamXWorldARArchConfig)
+
+    prefix: str = "Wan"

fastvideo/configs/models/dits/flux.py

@@ -0,0 +1,27 @@
+# SPDX-License-Identifier: Apache-2.0
+
+from dataclasses import dataclass, field
+
+from fastvideo.configs.models.dits.base import DiTArchConfig, DiTConfig
+
+
+@dataclass
+class FluxTransformer2DArchConfig(DiTArchConfig):
+
+    patch_size: int = 1
+    in_channels: int = 64
+    out_channels: int | None = None
+    num_layers: int = 19
+    num_single_layers: int = 38
+    attention_head_dim: int = 128
+    num_attention_heads: int = 24
+    joint_attention_dim: int = 4096
+    pooled_projection_dim: int = 768
+    guidance_embeds: bool = True
+    axes_dims_rope: tuple[int, int, int] = (16, 56, 56)
+
+
+@dataclass
+class FluxDiTConfig(DiTConfig):
+    arch_config: DiTArchConfig = field(default_factory=FluxTransformer2DArchConfig)
+    prefix: str = "flux"

fastvideo/configs/models/dits/glm_image.py

@@ -0,0 +1,61 @@
+# SPDX-License-Identifier: Apache-2.0
+from dataclasses import dataclass, field
+
+from fastvideo.configs.models.dits.base import DiTArchConfig, DiTConfig
+
+
+def is_blocks(n: str, m) -> bool:
+    return "transformer_blocks" in n and str.isdigit(n.split(".")[-1])
+
+
+@dataclass
+class GlmImageDiTArchConfig(DiTArchConfig):
+
+    _fsdp_shard_conditions: list = field(default_factory=lambda: [is_blocks])
+
+    hidden_size: int = 4096
+    num_attention_heads: int = 32
+    attention_head_dim: int = 128
+    in_channels: int = 16
+    out_channels: int = 16
+    num_layers: int = 30
+
+    text_embed_dim: int = 1472
+    time_embed_dim: int = 512
+    condition_dim: int = 256
+
+    prior_vq_quantizer_codebook_size: int = 16384
+
+    patch_size: int = 2
+
+    max_height: int = 2048
+    max_width: int = 2048
+
+    qk_norm: str = "layer_norm"
+    eps: float = 1e-5
+
+    exclude_lora_layers: list[str] = field(
+        default_factory=lambda: ["image_projector", "glyph_projector", "prior_token_embedding"])
+
+    param_names_mapping: dict = field(
+        default_factory=lambda: {
+            r"^glyph_projector\.net\.0\.proj\.(.*)$": r"glyph_projector.fc_in.\1",
+            r"^glyph_projector\.net\.2\.(.*)$": r"glyph_projector.fc_out.\1",
+            r"^prior_projector\.net\.0\.proj\.(.*)$": r"prior_projector.fc_in.\1",
+            r"^prior_projector\.net\.2\.(.*)$": r"prior_projector.fc_out.\1",
+            r"^transformer_blocks\.(\d+)\.ff\.net\.0\.proj\.(.*)$": r"transformer_blocks.\1.ff.fc_in.\2",
+            r"^transformer_blocks\.(\d+)\.ff\.net\.2\.(.*)$": r"transformer_blocks.\1.ff.fc_out.\2",
+        })
+
+    reverse_param_names_mapping: dict = field(default_factory=dict)
+    lora_param_names_mapping: dict = field(default_factory=dict)
+
+    def __post_init__(self):
+        super().__post_init__()
+        self.num_channels_latents = self.out_channels
+
+
+@dataclass
+class GlmImageDiTConfig(DiTConfig):
+    arch_config: DiTArchConfig = field(default_factory=GlmImageDiTArchConfig)
+    prefix: str = "GlmImage"

fastvideo/configs/models/dits/kandinsky5.py

@@ -2,14 +2,24 @@
 from dataclasses import dataclass, field
 
 from fastvideo.configs.models.dits.base import DiTArchConfig, DiTConfig
+from fastvideo.platforms import AttentionBackendEnum
+
+
+def _is_kandinsky5_transformer_block(n: str, m) -> bool:
+    return ("text_transformer_blocks" in n or "visual_transformer_blocks" in n) and n.split(".")[-1].isdigit()
 
 
 @dataclass
 class Kandinsky5ArchConfig(DiTArchConfig):
-    _fsdp_shard_conditions: list = field(default_factory=lambda: [
-        lambda n, m:
-        ("text_transformer_blocks" in n or "visual_transformer_blocks" in n) and n.split(".")[-1].isdigit()
-    ])
+    _fsdp_shard_conditions: list = field(default_factory=lambda: [_is_kandinsky5_transformer_block])
+
+    # NABLA block-sparse attention for attention_type="nabla" checkpoints, plus
+    # the dense backends every DiT supports.
+    _supported_attention_backends: tuple[AttentionBackendEnum, ...] = (
+        AttentionBackendEnum.NABLA_ATTN,
+        AttentionBackendEnum.FLASH_ATTN,
+        AttentionBackendEnum.TORCH_SDPA,
+    )
 
     # Native FastVideo implementation uses the same parameter names as diffusers
     # except FFN internals: Diffusers FFN uses `in_layer/out_layer`, while

fastvideo/configs/models/dits/wanvideo.py

@@ -1,5 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 from dataclasses import dataclass, field
+from typing import Literal
 
 from fastvideo.configs.models.dits.base import DiTArchConfig, DiTConfig
 
@@ -19,6 +20,11 @@ class WanVideoArchConfig(DiTArchConfig):
             r"^condition_embedder\.text_embedder\.linear_2\.(.*)$": r"condition_embedder.text_embedder.fc_out.\1",
             r"^condition_embedder\.time_embedder\.linear_1\.(.*)$": r"condition_embedder.time_embedder.mlp.fc_in.\1",
             r"^condition_embedder\.time_embedder\.linear_2\.(.*)$": r"condition_embedder.time_embedder.mlp.fc_out.\1",
+            # AnyFlow dual-timestep checkpoints expose delta_embedder weights with the
+            # same internal layout as time_embedder. The regex is harmless on plain
+            # Wan checkpoints (no delta_embedder keys to match).
+            r"^condition_embedder\.delta_embedder\.linear_1\.(.*)$": r"condition_embedder.delta_embedder.mlp.fc_in.\1",
+            r"^condition_embedder\.delta_embedder\.linear_2\.(.*)$": r"condition_embedder.delta_embedder.mlp.fc_out.\1",
             r"^condition_embedder\.time_proj\.(.*)$": r"condition_embedder.time_modulation.linear.\1",
             r"^condition_embedder\.image_embedder\.ff\.net\.0\.proj\.(.*)$":
             r"condition_embedder.image_embedder.ff.fc_in.\1",
@@ -86,6 +92,14 @@ class WanVideoArchConfig(DiTArchConfig):
     # "relativistic" keeps long rollouts in-distribution; a no-op unless sink_size > 0 and local_attn_size > 0.
     rope_cache_policy: str = "absolute"
 
+    # AnyFlow dual-timestep conditioning. Defaults preserve bit-identity with
+    # the legacy single-timestep forward (no delta_embedder allocated, no
+    # extra computation on the embedder forward path).
+    r_embedder: bool = False
+    r_embedder_fusion: Literal["additive", "gated"] = "additive"
+    r_embedder_gate_value: float = 0.25
+    r_embedder_deltatime_type: Literal["r", "t-r"] = "r"
+
     def __post_init__(self):
         super().__post_init__()
         self.out_channels = self.out_channels or self.in_channels

fastvideo/configs/models/encoders/base.py

@@ -43,11 +43,16 @@ class TextEncoderArchConfig(EncoderArchConfig):
     require_processor: bool = False
 
     def __post_init__(self) -> None:
-        self.tokenizer_kwargs = {
+        # update_model_arch re-runs __post_init__ after pipeline configs may
+        # have customized tokenizer_kwargs (e.g. kandinsky5/gen3c/longcat set
+        # "padding"); rebuilding the dict here would silently wipe those
+        # customizations, so only fill in defaults for keys not already set.
+        defaults = {
             "truncation": True,
             "max_length": self.text_len,
             "return_tensors": "pt",
         }
+        self.tokenizer_kwargs = defaults | self.tokenizer_kwargs
 
 
 @dataclass

fastvideo/configs/models/vaes/__init__.py

@@ -2,6 +2,7 @@ from fastvideo.configs.models.vaes.cosmosvae import CosmosVAEConfig
 from fastvideo.configs.models.vaes.cosmos2_5vae import Cosmos25VAEConfig
 from fastvideo.configs.models.vaes.gamecraftvae import GameCraftVAEConfig
 from fastvideo.configs.models.vaes.gen3cvae import Gen3CVAEConfig
+from fastvideo.configs.models.vaes.glm_image import GlmImageVAEConfig
 from fastvideo.configs.models.vaes.hunyuanvae import HunyuanVAEConfig
 from fastvideo.configs.models.vaes.hunyuan15vae import Hunyuan15VAEConfig
 from fastvideo.configs.models.vaes.ltx2vae import LTX2VAEConfig
@@ -21,4 +22,5 @@ __all__ = [
     "OobleckVAEArchConfig",
     "OobleckVAEConfig",
     "Flux2VAEConfig",
+    "GlmImageVAEConfig",
 ]

fastvideo/configs/models/vaes/glm_image.py

@@ -0,0 +1,95 @@
+# SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+
+from fastvideo.configs.models.vaes.autoencoder_kl import (AutoencoderKLArchConfig, AutoencoderKLVAEConfig)
+
+_GLM_IMAGE_LATENTS_MEAN: tuple[float, ...] = (
+    -0.2080078125,
+    1.875,
+    -0.470703125,
+    -1.265625,
+    -1.421875,
+    0.77734375,
+    -0.3671875,
+    -0.9453125,
+    0.318359375,
+    0.7734375,
+    -0.1884765625,
+    -0.022216796875,
+    -0.220703125,
+    -1.59375,
+    -0.81640625,
+    -0.255859375,
+)
+_GLM_IMAGE_LATENTS_STD: tuple[float, ...] = (
+    3.0625,
+    2.203125,
+    2.265625,
+    4.84375,
+    2.5,
+    3.9375,
+    2.203125,
+    3.03125,
+    2.1875,
+    2.046875,
+    2.71875,
+    2.390625,
+    2.390625,
+    2.453125,
+    2.25,
+    2.15625,
+)
+
+
+@dataclass
+class GlmImageVAEArchConfig(AutoencoderKLArchConfig):
+    act_fn: str = "silu"
+    block_out_channels: tuple[int, ...] = (128, 512, 1024, 1024)
+    down_block_types: tuple[str, ...] = (
+        "DownEncoderBlock2D",
+        "DownEncoderBlock2D",
+        "DownEncoderBlock2D",
+        "DownEncoderBlock2D",
+    )
+    up_block_types: tuple[str, ...] = (
+        "UpDecoderBlock2D",
+        "UpDecoderBlock2D",
+        "UpDecoderBlock2D",
+        "UpDecoderBlock2D",
+    )
+    force_upcast: bool = True
+    in_channels: int = 3
+    out_channels: int = 3
+    latent_channels: int = 16
+    latents_mean: tuple[float, ...] = _GLM_IMAGE_LATENTS_MEAN
+    latents_std: tuple[float, ...] = _GLM_IMAGE_LATENTS_STD
+    layers_per_block: int = 3
+    mid_block_add_attention: bool = False
+    norm_num_groups: int = 32
+    sample_size: int = 1024
+    scaling_factor: float = 0.18215
+    shift_factor: float | None = None
+    use_quant_conv: bool = False
+    use_post_quant_conv: bool = False
+
+    temporal_compression_ratio: int = 1
+    spatial_compression_ratio: int = 8
+
+
+@dataclass
+class GlmImageVAEConfig(AutoencoderKLVAEConfig):
+    arch_config: GlmImageVAEArchConfig = field(default_factory=GlmImageVAEArchConfig)
+
+    use_tiling: bool = True
+    use_temporal_tiling: bool = False
+    use_parallel_tiling: bool = False
+
+    tile_sample_min_height: int = 512
+    tile_sample_min_width: int = 512
+    tile_sample_stride_height: int = 384
+    tile_sample_stride_width: int = 384
+
+    load_encoder: bool = True
+    load_decoder: bool = True

fastvideo/configs/pipelines/__init__.py

@@ -1,10 +1,12 @@
 from fastvideo.configs.pipelines.base import PipelineConfig
 from fastvideo.configs.pipelines.cosmos import CosmosConfig
 from fastvideo.configs.pipelines.cosmos2_5 import Cosmos25Config
+from fastvideo.configs.pipelines.dreamx_world import DreamXWorld5BARPipelineConfig, DreamXWorld5BCamPipelineConfig
 from fastvideo.configs.pipelines.hunyuan import FastHunyuanConfig, HunyuanConfig
 from fastvideo.configs.pipelines.hunyuan15 import Hunyuan15T2V480PConfig, Hunyuan15T2V720PConfig
 from fastvideo.configs.pipelines.hunyuangamecraft import HunyuanGameCraftPipelineConfig
 from fastvideo.configs.pipelines.hyworld import HYWorldConfig
+from fastvideo.configs.pipelines.kandinsky5 import Kandinsky5I2VConfig, Kandinsky5T2VConfig
 from fastvideo.configs.pipelines.matrixgame2 import MatrixGame2I2V480PConfig
 from fastvideo.configs.pipelines.matrixgame3 import MatrixGame3I2V720PConfig
 from fastvideo.pipelines.basic.ltx2.pipeline_configs import LTX2T2VConfig
@@ -16,5 +18,6 @@ __all__ = [
     "HunyuanConfig", "FastHunyuanConfig", "HunyuanGameCraftPipelineConfig", "PipelineConfig", "Hunyuan15T2V480PConfig",
     "Hunyuan15T2V720PConfig", "WanT2V480PConfig", "WanI2V480PConfig", "WanT2V720PConfig", "WanI2V720PConfig",
     "SelfForcingWanT2V480PConfig", "LucyEditDevConfig", "CosmosConfig", "Cosmos25Config", "LTX2T2VConfig",
-    "HYWorldConfig", "MatrixGame2I2V480PConfig", "MatrixGame3I2V720PConfig", "get_pipeline_config_cls_from_name"
+    "DreamXWorld5BCamPipelineConfig", "DreamXWorld5BARPipelineConfig", "HYWorldConfig", "Kandinsky5T2VConfig",
+    "Kandinsky5I2VConfig", "MatrixGame2I2V480PConfig", "MatrixGame3I2V720PConfig", "get_pipeline_config_cls_from_name"
 ]

fastvideo/configs/pipelines/base.py

@@ -276,27 +276,6 @@ class PipelineConfig:
                 f"Length of text postprocess functions ({len(self.postprocess_text_funcs)}) must be equal to length of text preprocessing functions ({len(self.preprocess_text_funcs)})"
             )
 
-    def estimate_request_cost(self, request: Any) -> float:
-        """Estimate relative memory/compute cost for batching admission.
-
-        The default is intentionally simple and model-agnostic: pixel count
-        times frame count. Pipeline subclasses can override this when they have
-        calibrated costs.
-        """
-        height = getattr(request, "height", None)
-        width = getattr(request, "width", None)
-        num_frames = getattr(request, "num_frames", None)
-        if isinstance(height, list):
-            height = height[0] if height else None
-        if isinstance(width, list):
-            width = width[0] if width else None
-        if isinstance(num_frames, list):
-            num_frames = num_frames[0] if num_frames else None
-        height = int(height or 1)
-        width = int(width or 1)
-        num_frames = int(num_frames or 1)
-        return float(max(1, height) * max(1, width) * max(1, num_frames))
-
     def dump_to_json(self, file_path: str):
         output_dict = shallow_asdict(self)
         del_keys = []

fastvideo/configs/pipelines/dreamx_world.py

@@ -0,0 +1,127 @@
+# SPDX-License-Identifier: Apache-2.0
+"""DreamX-World-5B-Cam FastVideo model configuration helpers."""
+
+from collections.abc import Callable
+from dataclasses import dataclass, field
+
+import torch
+
+from fastvideo.configs.models import DiTConfig, EncoderConfig, VAEConfig
+from fastvideo.configs.models.dits.dreamx_world import (DreamXWorldARArchConfig, DreamXWorldARConfig,
+                                                        DreamXWorldArchConfig, DreamXWorldConfig)
+from fastvideo.configs.models.encoders import BaseEncoderOutput, T5Config
+from fastvideo.configs.models.encoders.t5 import T5ArchConfig
+from fastvideo.configs.models.vaes import WanVAEConfig
+from fastvideo.configs.pipelines.base import PipelineConfig
+from fastvideo.configs.pipelines.wan import LucyEditDevConfig, t5_postprocess_text
+
+
+def make_dreamx_world_5b_cam_dit_config() -> DreamXWorldConfig:
+    """Return the DreamX-World DiT config matching DreamX-World-5B-Cam."""
+    return DreamXWorldConfig(arch_config=DreamXWorldArchConfig(
+        num_attention_heads=24,
+        attention_head_dim=128,
+        in_channels=48,
+        out_channels=48,
+        ffn_dim=14336,
+        num_layers=30,
+        cross_attn_norm=True,
+        qk_norm="rms_norm_across_heads",
+        add_control_adapter=True,
+        cam_method="prope",
+        attn_compress=1,
+        cam_self_attn_layers=None,
+    ))
+
+
+def make_dreamx_world_5b_ar_dit_config() -> DreamXWorldARConfig:
+    """Return the DreamX-World-5B autoregressive causal DiT config."""
+    return DreamXWorldARConfig(arch_config=DreamXWorldARArchConfig(
+        model_type="ti2v",
+        num_attention_heads=24,
+        attention_head_dim=128,
+        in_channels=48,
+        out_channels=48,
+        ffn_dim=14336,
+        num_layers=30,
+        cross_attn_norm=True,
+        qk_norm=True,
+        add_control_adapter=True,
+        cam_method="prope",
+        attn_compress=4,
+        cam_self_attn_layers=tuple(range(30)),
+        local_attn_size=12,
+        sink_size=3,
+        num_frames_per_block=3,
+    ))
+
+
+def make_dreamx_world_5b_cam_vae_config() -> WanVAEConfig:
+    """Return the Wan2.2 48-channel VAE config used by DreamX-World-5B-Cam."""
+    return LucyEditDevConfig().vae_config
+
+
+def make_dreamx_world_5b_cam_text_encoder_config() -> T5Config:
+    """Return the UMT5-XXL text encoder config used by DreamX-World-5B-Cam."""
+    return T5Config(
+        arch_config=T5ArchConfig(
+            vocab_size=256384,
+            d_model=4096,
+            d_kv=64,
+            d_ff=10240,
+            num_layers=24,
+            num_decoder_layers=None,
+            num_heads=64,
+            relative_attention_num_buckets=32,
+            dropout_rate=0.0,
+            text_len=512,
+            feed_forward_proj="gelu",
+            is_encoder_decoder=False,
+        ),
+        prefix="umt5",
+    )
+
+
+@dataclass
+class DreamXWorld5BCamPipelineConfig(PipelineConfig):
+    """Pipeline config for the first-scope DreamX-World-5B-Cam mode."""
+
+    dit_config: DiTConfig = field(default_factory=make_dreamx_world_5b_cam_dit_config)
+    vae_config: VAEConfig = field(default_factory=make_dreamx_world_5b_cam_vae_config)
+    text_encoder_configs: tuple[EncoderConfig,
+                                ...] = field(default_factory=lambda: (make_dreamx_world_5b_cam_text_encoder_config(), ))
+    postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], torch.Tensor],
+                                  ...] = field(default_factory=lambda: (t5_postprocess_text, ))
+    text_encoder_precisions: tuple[str, ...] = field(default_factory=lambda: ("bf16", ))
+    flow_shift: float | None = 3.0
+    ti2v_task: bool = True
+    expand_timesteps: bool = True
+    vae_tiling: bool = False
+    vae_sp: bool = False
+    vae_precision: str = "fp32"
+    vae_decode_precision: str | None = "bf16"
+    dit_precision: str = "bf16"
+
+    def __post_init__(self) -> None:
+        self.vae_config.load_encoder = True
+        self.vae_config.load_decoder = True
+        self.dit_config.expand_timesteps = self.expand_timesteps
+
+
+@dataclass
+class DreamXWorld5BARPipelineConfig(DreamXWorld5BCamPipelineConfig):
+    """Pipeline config for DreamX-World-5B autoregressive forcing."""
+
+    dit_config: DiTConfig = field(default_factory=make_dreamx_world_5b_ar_dit_config)
+    flow_shift: float | None = 5.0
+    ti2v_task: bool = True
+    is_causal: bool = True
+    dmd_denoising_steps: tuple[int, ...] = (1000, 750, 500, 250)
+    warp_denoising_step: bool = True
+    context_noise: float = 0.1
+    num_frames_per_block: int = 3
+    color_correction_strength: float = 1.0
+
+    def __post_init__(self) -> None:
+        super().__post_init__()
+        self.dit_config.expand_timesteps = True

fastvideo/configs/pipelines/flux.py

@@ -0,0 +1,74 @@
+# SPDX-License-Identifier: Apache-2.0
+
+from __future__ import annotations
+
+from collections.abc import Callable
+from dataclasses import dataclass, field
+
+import torch
+
+from fastvideo.configs.models import EncoderConfig
+from fastvideo.configs.models.dits.flux import FluxDiTConfig
+from fastvideo.configs.models.encoders import (
+    BaseEncoderOutput,
+    CLIPTextConfig,
+    T5LargeConfig,
+)
+from fastvideo.configs.models.vaes.autoencoder_kl import AutoencoderKLVAEConfig
+from fastvideo.configs.pipelines.base import PipelineConfig, preprocess_text
+
+
+def _flux_clip_pooled_postprocess(outputs: BaseEncoderOutput) -> torch.Tensor:
+    """CLIP branch for FLUX: Diffusers uses pooled prompt embeddings only."""
+    if outputs.pooler_output is None:
+        raise RuntimeError(
+            "FLUX CLIP conditioning requires pooler_output. Ensure the CLIP text encoder returns pooled features.")
+    return outputs.pooler_output
+
+
+def _flux_t5_sequence_postprocess(outputs: BaseEncoderOutput) -> torch.Tensor:
+    if outputs.last_hidden_state is None:
+        raise RuntimeError("FLUX T5 conditioning requires last_hidden_state.")
+    return outputs.last_hidden_state
+
+
+@dataclass
+class FluxPipelineConfig(PipelineConfig):
+    """Pipeline layout for Diffusers FLUX.1-dev (CLIP + T5 + packed DiT + FlowMatch)."""
+
+    scheduler_arch: str = "FlowMatchEulerDiscreteScheduler"
+    transformer_arch: str = "FluxTransformer2DModel"
+    vae_arch: str = "AutoencoderKL"
+    text_encoder_archs: tuple[str, ...] = ("CLIPTextModel", "T5EncoderModel")
+    tokenizer_archs: tuple[str, ...] = ("CLIPTokenizer", "T5TokenizerFast")
+
+    dit_config: FluxDiTConfig = field(default_factory=FluxDiTConfig)
+    vae_config: AutoencoderKLVAEConfig = field(default_factory=AutoencoderKLVAEConfig)
+
+    embedded_cfg_scale: float = 3.5
+    flow_shift: float | None = None
+
+    text_encoder_configs: tuple[EncoderConfig, ...] = field(default_factory=lambda: (CLIPTextConfig(), T5LargeConfig()))
+    preprocess_text_funcs: tuple[Callable[[str], str],
+                                 ...] = field(default_factory=lambda: (preprocess_text, preprocess_text))
+    postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], torch.Tensor], ...] = field(
+        default_factory=lambda: (_flux_clip_pooled_postprocess, _flux_t5_sequence_postprocess))
+
+    dit_precision: str = "bf16"
+    vae_precision: str = "fp32"
+    text_encoder_precisions: tuple[str, ...] = field(default_factory=lambda: ("fp32", "bf16"))
+
+    def __post_init__(self) -> None:
+        te_cfgs = list(self.text_encoder_configs)
+        if len(te_cfgs) >= 1:
+            te_cfgs[0].tokenizer_kwargs.setdefault("padding", "max_length")
+            te_cfgs[0].tokenizer_kwargs.setdefault("max_length", 77)
+            te_cfgs[0].tokenizer_kwargs.setdefault("truncation", True)
+            te_cfgs[0].tokenizer_kwargs.setdefault("return_tensors", "pt")
+        if len(te_cfgs) >= 2:
+            cap = 512
+            te_cfgs[1].tokenizer_kwargs["max_length"] = min(int(te_cfgs[1].tokenizer_kwargs.get("max_length", cap)),
+                                                            cap)
+            te_cfgs[1].tokenizer_kwargs.setdefault("padding", "max_length")
+            te_cfgs[1].tokenizer_kwargs.setdefault("truncation", True)
+            te_cfgs[1].tokenizer_kwargs.setdefault("return_tensors", "pt")

fastvideo/configs/pipelines/glm_image.py

@@ -0,0 +1,46 @@
+# SPDX-License-Identifier: Apache-2.0
+from collections.abc import Callable
+from dataclasses import dataclass, field
+
+import torch
+
+from fastvideo.configs.models import DiTConfig, EncoderConfig, VAEConfig
+from fastvideo.configs.models.dits.glm_image import GlmImageDiTConfig
+from fastvideo.configs.models.encoders import BaseEncoderOutput, T5Config
+from fastvideo.configs.models.vaes.glm_image import GlmImageVAEConfig
+from fastvideo.configs.pipelines.base import PipelineConfig
+
+
+def glm_image_t5_postprocess(outputs: BaseEncoderOutput) -> torch.Tensor:
+    mask: torch.Tensor = outputs.attention_mask
+    hidden_state: torch.Tensor = outputs.last_hidden_state
+    seq_lens = mask.gt(0).sum(dim=1).long()
+
+    assert torch.isnan(hidden_state).sum() == 0, "T5 hidden states contain NaN"
+
+    max_len = 512
+    prompt_embeds = [u[:min(v, max_len)] for u, v in zip(hidden_state, seq_lens, strict=True)]
+    prompt_embeds_tensor: torch.Tensor = torch.stack(
+        [torch.cat([u, u.new_zeros(max_len - u.size(0), u.size(1))]) for u in prompt_embeds], dim=0)
+
+    return prompt_embeds_tensor
+
+
+@dataclass
+class GlmImageConfig(PipelineConfig):
+
+    dit_config: DiTConfig = field(default_factory=GlmImageDiTConfig)
+    dit_precision: str = "bf16"
+
+    vae_config: VAEConfig = field(default_factory=GlmImageVAEConfig)
+    vae_precision: str = "fp32"
+    vae_tiling: bool = True
+    vae_sp: bool = False
+
+    text_encoder_configs: tuple[EncoderConfig, ...] = field(default_factory=lambda: (T5Config(), ))
+    text_encoder_precisions: tuple[str, ...] = field(default_factory=lambda: ("fp32", ))
+    postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], torch.Tensor],
+                                  ...] = field(default_factory=lambda: (glm_image_t5_postprocess, ))
+
+    flow_shift: float | None = 1.0
+    embedded_cfg_scale: float = 7.5

fastvideo/configs/pipelines/kandinsky5.py

@@ -0,0 +1,122 @@
+# SPDX-License-Identifier: Apache-2.0
+
+from __future__ import annotations
+
+from collections.abc import Callable
+from dataclasses import dataclass, field
+from typing import Any
+
+import torch
+
+from fastvideo.configs.models import DiTConfig, EncoderConfig, VAEConfig
+from fastvideo.configs.models.dits import Kandinsky5VideoConfig
+from fastvideo.configs.models.encoders import BaseEncoderOutput, CLIPTextConfig
+from fastvideo.configs.models.encoders.reason1 import Reason1Config
+from fastvideo.configs.models.vaes import HunyuanVAEConfig
+from fastvideo.configs.pipelines.base import PipelineConfig, preprocess_text
+
+# Byte-exact copy of the upstream Kandinsky5/diffusers template, including the
+# "promt"/"scren" typos: the checkpoints were trained with this exact system  # codespell:ignore promt,scren
+# prompt, and ENCODE_START_IDX below is the tokenized length of everything
+# before the user prompt. Fixing the typos shifts user content to index 127
+# and mis-conditions every generation.
+KANDINSKY5_PROMPT_TEMPLATE = "\n".join([
+    "<|im_start|>system\nYou are a promt engineer. Describe the video in detail.",  # codespell:ignore promt
+    "Describe how the camera moves or shakes, describe the zoom and view angle, whether it follows the objects.",
+    "Describe the location of the video, main characters or objects and their action.",
+    "Describe the dynamism of the video and presented actions.",
+    "Name the visual style of the video: whether it is a professional footage, user generated content, some kind of animation, video game or scren content.",  # codespell:ignore scren
+    "Describe the visual effects, postprocessing and transitions if they are presented in the video.",
+    "Pay attention to the order of key actions shown in the scene.<|im_end|>",
+    "<|im_start|>user\n{}<|im_end|>",
+])
+KANDINSKY5_PROMPT_TEMPLATE_ENCODE_START_IDX = 129
+
+
+def kandinsky5_qwen_preprocess_text(prompt: str) -> str:
+    if not prompt.strip():
+        prompt = "."
+    return KANDINSKY5_PROMPT_TEMPLATE.format(prompt)
+
+
+def kandinsky5_qwen_postprocess_text(outputs: BaseEncoderOutput,
+                                     mask: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
+    if outputs.hidden_states is None:
+        raise RuntimeError("Kandinsky5 Qwen prompt embeddings require hidden_states.")
+    hidden_states = outputs.hidden_states[-1]
+    prompt_embeds = hidden_states[:, KANDINSKY5_PROMPT_TEMPLATE_ENCODE_START_IDX:]
+    mask = mask[:, KANDINSKY5_PROMPT_TEMPLATE_ENCODE_START_IDX:]
+    if prompt_embeds.shape[1] == 0:
+        prompt_embeds = hidden_states[:, -1:]
+        mask = torch.ones((mask.shape[0], 1), dtype=mask.dtype, device=mask.device)
+    return prompt_embeds, mask
+
+
+def kandinsky5_clip_postprocess_text(outputs: BaseEncoderOutput) -> torch.Tensor:
+    if outputs.pooler_output is None:
+        raise RuntimeError("Kandinsky5 CLIP pooled output is required.")
+    return outputs.pooler_output
+
+
+@dataclass
+class Kandinsky5T2VConfig(PipelineConfig):
+    """Kandinsky-5.0 Lite text-to-video pipeline configuration."""
+
+    dit_config: DiTConfig = field(default_factory=Kandinsky5VideoConfig)
+    vae_config: VAEConfig = field(default_factory=HunyuanVAEConfig)
+
+    text_encoder_configs: tuple[EncoderConfig, ...] = field(default_factory=lambda: (Reason1Config(), CLIPTextConfig()))
+    preprocess_text_funcs: tuple[Callable[[str], Any], ...] = field(
+        default_factory=lambda: (kandinsky5_qwen_preprocess_text, preprocess_text))
+    postprocess_text_funcs: tuple[Callable[..., Any], ...] = field(
+        default_factory=lambda: (kandinsky5_qwen_postprocess_text, kandinsky5_clip_postprocess_text))
+
+    dit_precision: str = "bf16"
+    vae_precision: str = "bf16"
+    text_encoder_precisions: tuple[str, ...] = field(default_factory=lambda: ("bf16", "bf16"))
+    text_encoder_max_lengths: tuple[int, ...] = field(
+        default_factory=lambda: (KANDINSKY5_PROMPT_TEMPLATE_ENCODE_START_IDX + 512, 77))
+
+    flow_shift: float | None = 5.0
+    vae_tiling: bool = True
+
+    def __post_init__(self) -> None:
+        if len(self.text_encoder_configs) != 2:
+            raise ValueError(f"Kandinsky5 pipeline requires exactly 2 text encoders (qwen and clip), "
+                             f"but got {len(self.text_encoder_configs)} encoder(s).")
+        if len(self.text_encoder_precisions) != 2:
+            raise ValueError("Kandinsky5 pipeline requires exactly 2 text encoder precisions, "
+                             f"but got {len(self.text_encoder_precisions)}.")
+        if len(self.text_encoder_max_lengths) != 2:
+            raise ValueError("Kandinsky5 pipeline requires exactly 2 text encoder max lengths, "
+                             f"but got {len(self.text_encoder_max_lengths)}.")
+
+        self.vae_config.load_encoder = False
+        self.vae_config.load_decoder = True
+
+        qwen_cfg = self.text_encoder_configs[0]
+        qwen_cfg.arch_config.output_hidden_states = True
+        qwen_cfg.arch_config.tokenizer_kwargs.update({
+            "padding": True,
+            "truncation": True,
+            "return_tensors": "pt",
+        })
+
+        clip_cfg = self.text_encoder_configs[1]
+        clip_cfg.arch_config.tokenizer_kwargs.update({
+            "padding": "max_length",
+            "max_length": 77,
+            "truncation": True,
+            "add_special_tokens": True,
+            "return_tensors": "pt",
+        })
+
+
+@dataclass
+class Kandinsky5I2VConfig(Kandinsky5T2VConfig):
+    """Kandinsky-5.0 image-to-video pipeline configuration."""
+
+    def __post_init__(self) -> None:
+        super().__post_init__()
+        # I2V needs the VAE encoder to encode the conditioning image.
+        self.vae_config.load_encoder = True

fastvideo/dataset/parquet_dataset_map_style.py

@@ -2,6 +2,7 @@
 import os
 import pickle
 import random
+from collections.abc import Sequence
 from typing import Any
 
 import pyarrow as pa
@@ -94,8 +95,56 @@ class DP_SP_BatchSampler(Sampler[list[int]]):
         return len(self.sp_group_local_indices) // self.batch_size
 
 
-def get_parquet_files_and_length(path: str):
-    dataset_root = os.path.realpath(os.path.expanduser(path))
+def _parse_data_path_specs(path: str | Sequence[str] | dict[str, int]) -> list[tuple[str, int]]:
+    """Parse one or more dataset roots with old-framework repeat counts."""
+    if isinstance(path, dict):
+        return [
+            (str(root), int(repeat))
+            for root, repeat in path.items()
+            if int(repeat) > 0
+        ]
+    if isinstance(path, Sequence) and not isinstance(path, str):
+        return [(str(root), 1) for root in path]
+
+    specs: list[tuple[str, int]] = []
+    for part in str(path).split(","):
+        part = part.strip()
+        if not part:
+            continue
+        if ":" in part:
+            dir_path, count_str = part.rsplit(":", 1)
+            count = int(count_str)
+        else:
+            dir_path, count = part, 1
+        if count > 0:
+            specs.append((dir_path.strip(), count))
+    return specs
+
+
+def get_parquet_files_and_length(path: str | Sequence[str] | dict[str, int]):
+    specs = _parse_data_path_specs(path)
+    if len(specs) != 1 or specs[0][1] != 1:
+        all_file_names: list[str] = []
+        all_lengths: list[int] = []
+        for root, repeat in specs:
+            file_names, lengths = get_parquet_files_and_length(str(root))
+            for _ in range(repeat):
+                all_file_names.extend(file_names)
+                all_lengths.extend(lengths)
+        if not all_file_names:
+            raise FileNotFoundError(
+                "No parquet files found under dataset paths: "
+                f"{path}. "
+                "Please verify these paths point to preprocessed parquet data."
+            )
+        file_lengths = sorted(
+            zip(all_file_names, all_lengths, strict=True),
+            key=lambda x: x[0],
+        )
+        file_names_sorted, lengths_sorted = zip(*file_lengths, strict=True)
+        return file_names_sorted, lengths_sorted
+
+    dataset_root = os.path.realpath(os.path.expanduser(specs[0][0]))
     # Check if cached info exists
     cache_dir = os.path.join(dataset_root, "map_style_cache")
     cache_file = os.path.join(cache_dir, "file_info.pkl")
@@ -268,7 +317,7 @@ class LatentsParquetMapStyleDataset(Dataset):
 
     def __init__(
         self,
-        path: str,
+        path: str | Sequence[str] | dict[str, int],
         batch_size: int,
         parquet_schema: pa.Schema,
         cfg_rate: float = 0.0,
@@ -358,7 +407,6 @@ class LatentsParquetMapStyleDataset(Dataset):
     def __len__(self):
         return sum(self.lengths)
 
-
 # ────────────────────────────────────────────────────────────────────────────
 # 3.  Loader helper – everything else stays just like your original trainer
 # ────────────────────────────────────────────────────────────────────────────

fastvideo/distributed/parallel_state.py

@@ -23,6 +23,7 @@ If you only need to use the distributed environment without model parallelism,
  you can skip the model parallel initialization and destruction steps.
 """
 import contextlib
+import gc
 import os
 import pickle
 import weakref
@@ -1006,6 +1007,13 @@ def cleanup_dist_env_and_memory(shutdown_ray: bool = False):
     if shutdown_ray:
         import ray  # Lazy import Ray
         ray.shutdown()
+    # Actually free GPU memory, as the name promises. FSDP-wrapped modules sit
+    # in reference cycles (module <-> FSDP state <-> hooks), so dropping the
+    # last user reference does not free their parameters until a gc pass runs.
+    # Without this, back-to-back model-load tests in one pytest session OOM.
+    gc.collect()
+    if torch.cuda.is_available():
+        torch.cuda.empty_cache()
 
 
 def get_same_node_ranks(pg: ProcessGroup | StatelessProcessGroup, source_rank: int = 0) -> list[int]:

fastvideo/entrypoints/openai/api_server.py

@@ -10,7 +10,6 @@ from fastapi.middleware.cors import CORSMiddleware
 
 from fastvideo.api.presets import validate_preset_selection
 from fastvideo.api.schema import GenerationRequest
-from fastvideo.entrypoints.openai.batching import VideoBatchScheduler
 from fastvideo.entrypoints.openai.state import (
     DEFAULT_OUTPUT_DIR,
     clear_state,
@@ -60,29 +59,11 @@ async def lifespan(app: FastAPI) -> AsyncIterator[None]:
     generator = VideoGenerator.from_fastvideo_args(args)
     logger.info("Model loaded successfully.")
 
-    video_batch_scheduler: VideoBatchScheduler | None = None
-    if args.batching_mode == "dynamic" and args.batching_max_size > 1:
-        video_batch_scheduler = VideoBatchScheduler(generator, args)
-        await video_batch_scheduler.start()
-        logger.info(
-            "Started dynamic video batch scheduler: max_size=%d delay_ms=%.2f",
-            args.batching_max_size,
-            args.batching_delay_ms,
-        )
-
-    set_state(
-        generator,
-        args,
-        output_dir,
-        default_request=default_request,
-        video_batch_scheduler=video_batch_scheduler,
-    )
+    set_state(generator, args, output_dir, default_request=default_request)
 
     yield  # server is running
 
     logger.info("Shutting down — releasing model resources ...")
-    if video_batch_scheduler is not None:
-        await video_batch_scheduler.stop()
     generator.shutdown()
     clear_state()
     logger.info("Shutdown complete.")

fastvideo/entrypoints/openai/batching.py

@@ -1,183 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-from __future__ import annotations
-
-import asyncio
-import time
-from collections import deque
-from dataclasses import dataclass
-from typing import Any
-
-from fastvideo.api.sampling_param import SamplingParam
-from fastvideo.batching.signature import can_dynamic_batch
-from fastvideo.fastvideo_args import FastVideoArgs
-from fastvideo.logger import init_logger
-
-logger = init_logger(__name__)
-
-
-@dataclass
-class _VideoBatchJob:
-    request_id: str
-    kwargs: dict[str, Any]
-    future: asyncio.Future
-    enqueue_time: float
-
-
-class VideoBatchScheduler:
-    """Async FIFO scheduler for OpenAI-compatible video generation."""
-
-    def __init__(self, generator: Any, fastvideo_args: FastVideoArgs) -> None:
-        self._generator = generator
-        self._fastvideo_args = fastvideo_args
-        self._queue: asyncio.Queue[_VideoBatchJob | None] = asyncio.Queue()
-        self._pending: deque[_VideoBatchJob] = deque()
-        self._task: asyncio.Task | None = None
-        self._stopped = False
-
-    @property
-    def enabled(self) -> bool:
-        return self._fastvideo_args.batching_mode == "dynamic" and self._fastvideo_args.batching_max_size > 1
-
-    async def start(self) -> None:
-        if self._task is not None:
-            return
-        self._task = asyncio.create_task(self._run(), name="fastvideo-video-batch-scheduler")
-
-    async def stop(self) -> None:
-        self._stopped = True
-        await self._queue.put(None)
-        if self._task is not None:
-            await self._task
-            self._task = None
-
-    async def submit(self, request_id: str, kwargs: dict[str, Any]) -> Any:
-        if self._stopped:
-            raise RuntimeError("Video batch scheduler is stopped; cannot submit new requests")
-        loop = asyncio.get_running_loop()
-        future = loop.create_future()
-        await self._queue.put(
-            _VideoBatchJob(
-                request_id=request_id,
-                kwargs=dict(kwargs),
-                future=future,
-                enqueue_time=time.perf_counter(),
-            ))
-        return await future
-
-    async def _run(self) -> None:
-        while not self._stopped:
-            job = await self._get_next_job()
-            if job is None:
-                break
-            batch = await self._collect_batch(job)
-            await self._dispatch(batch)
-
-        while self._pending:
-            pending = self._pending.popleft()
-            if not pending.future.done():
-                pending.future.set_exception(RuntimeError("Video batch scheduler stopped before dispatch"))
-
-    async def _get_next_job(self) -> _VideoBatchJob | None:
-        if self._pending:
-            return self._pending.popleft()
-        return await self._queue.get()
-
-    async def _collect_batch(self, first: _VideoBatchJob) -> list[_VideoBatchJob]:
-        batch = [first]
-        max_size = self._fastvideo_args.batching_max_size
-        delay_s = max(0.0, self._fastvideo_args.batching_delay_ms / 1000.0)
-        deadline = first.enqueue_time + delay_s
-
-        while len(batch) < max_size:
-            if delay_s > 0:
-                timeout = deadline - time.perf_counter()
-                if timeout <= 0:
-                    break
-                try:
-                    candidate = await asyncio.wait_for(self._get_next_job(), timeout=timeout)
-                except TimeoutError:
-                    break
-            else:
-                # delay=0 means "don't wait": greedily drain whatever is
-                # already queued so max_size still coalesces.
-                if self._pending:
-                    candidate = self._pending.popleft()
-                else:
-                    try:
-                        candidate = self._queue.get_nowait()
-                    except asyncio.QueueEmpty:
-                        break
-            if candidate is None:
-                await self._queue.put(None)
-                break
-            if self._jobs_are_compatible(batch[0], candidate):
-                batch.append(candidate)
-                continue
-            self._pending.appendleft(candidate)
-            break
-        return batch
-
-    async def _dispatch(self, batch: list[_VideoBatchJob]) -> None:
-        loop = asyncio.get_running_loop()
-        request_ids = [job.request_id for job in batch]
-        queue_wait_ms = (time.perf_counter() - min(job.enqueue_time for job in batch)) * 1000.0
-        if self._fastvideo_args.enable_batching_metrics:
-            logger.info(
-                "Dispatching video batch: request_ids=%s size=%d queue_wait_ms=%.2f",
-                request_ids,
-                len(batch),
-                queue_wait_ms,
-            )
-
-        try:
-            results = await loop.run_in_executor(
-                None,
-                lambda: self._generator.generate_video_batch([job.kwargs for job in batch]),
-            )
-        except Exception as exc:
-            for job in batch:
-                if not job.future.done():
-                    job.future.set_exception(exc)
-            return
-
-        if len(results) != len(batch):
-            error = RuntimeError(f"Video batch returned {len(results)} results for {len(batch)} requests")
-            for job in batch:
-                if not job.future.done():
-                    job.future.set_exception(error)
-            return
-
-        for job, result in zip(batch, results, strict=True):
-            if not job.future.done():
-                job.future.set_result(result)
-
-    def _jobs_are_compatible(self, base: _VideoBatchJob, candidate: _VideoBatchJob) -> bool:
-        try:
-            base_sampling, base_extra = self._sampling_param_from_kwargs(base.kwargs)
-            candidate_sampling, candidate_extra = self._sampling_param_from_kwargs(candidate.kwargs)
-        except Exception:
-            return False
-        return can_dynamic_batch(
-            base_sampling,
-            candidate_sampling,
-            base_extra=base_extra,
-            candidate_extra=candidate_extra,
-        ).can_batch
-
-    def _sampling_param_from_kwargs(self, kwargs: dict[str, Any]) -> tuple[SamplingParam, dict[str, Any]]:
-        sampling_param = SamplingParam.from_pretrained(self._fastvideo_args.model_path)
-        updates = dict(kwargs)
-        prompt = updates.pop("prompt", None)
-        extra: dict[str, Any] = {}
-        for key in (
-                "ltx2_audio_latents",
-                "ltx2_audio_clean_latent",
-                "ltx2_audio_denoise_mask",
-                "audio_num_frames",
-                "video_position_offset_sec",
-        ):
-            if key in updates:
-                extra[key] = updates.pop(key)
-        sampling_param.update(updates)
-        sampling_param.prompt = prompt
-        return sampling_param, extra

fastvideo/entrypoints/openai/state.py

@@ -11,7 +11,6 @@ from typing import TYPE_CHECKING
 
 if TYPE_CHECKING:
     from fastvideo.api.schema import GenerationRequest
-    from fastvideo.entrypoints.openai.batching import VideoBatchScheduler
     from fastvideo.entrypoints.video_generator import VideoGenerator
     from fastvideo.fastvideo_args import FastVideoArgs
 
@@ -21,7 +20,6 @@ _generator: VideoGenerator | None = None
 _fastvideo_args: FastVideoArgs | None = None
 _output_dir: str = DEFAULT_OUTPUT_DIR
 _default_request: GenerationRequest | None = None
-_video_batch_scheduler: VideoBatchScheduler | None = None
 
 
 def get_generator() -> VideoGenerator:
@@ -46,31 +44,23 @@ def get_default_request() -> GenerationRequest | None:
     return _default_request
 
 
-def get_video_batch_scheduler() -> VideoBatchScheduler | None:
-    """Return the video batch scheduler when dynamic batching is enabled."""
-    return _video_batch_scheduler
-
-
 def set_state(
     generator: VideoGenerator,
     fastvideo_args: FastVideoArgs,
     output_dir: str,
     default_request: GenerationRequest | None = None,
-    video_batch_scheduler: VideoBatchScheduler | None = None,
 ) -> None:
     """Set all server state at once (called from lifespan)."""
-    global _generator, _fastvideo_args, _output_dir, _default_request, _video_batch_scheduler
+    global _generator, _fastvideo_args, _output_dir, _default_request
     _generator = generator
     _fastvideo_args = fastvideo_args
     _output_dir = output_dir
     _default_request = default_request
-    _video_batch_scheduler = video_batch_scheduler
 
 
 def clear_state() -> None:
     """Clear server state on shutdown."""
-    global _generator, _fastvideo_args, _default_request, _video_batch_scheduler
+    global _generator, _fastvideo_args, _default_request
     _generator = None
     _fastvideo_args = None
     _default_request = None
-    _video_batch_scheduler = None