[ci] Cherrypick perf regression skills to component time commit

[ci] Add performance baseline reseed skill and failed-run artifacts

.agents/memory/codebase-map/README.md

@@ -119,7 +119,7 @@ FastVideo-WorldModel/
 ## Build & Test Commands
 
 ```bash
-uv pip install -e .[dev]                          # Editable install
+uv pip install -e ".[dev]"                        # Editable install
 pre-commit run --all-files                        # Lint/format/spell
 pytest tests/                                     # Top-level tests
 pytest fastvideo/tests/ -v                        # Package tests

.agents/skills/index.jsonl

@@ -6,3 +6,5 @@
 {"name": "index-related-work", "description": "Ingest a paper or repository into the related work index", "path": "index-related-work/SKILL.md", "status": "draft", "trust": "low"}
 {"name": "search-related-work", "description": "Query the related work index for relevant papers, repos, or comparisons", "path": "search-related-work/SKILL.md", "status": "draft", "trust": "low"}
 {"name": "seed-ssim-references", "description": "Run a new or updated fastvideo/tests/ssim/ test on Modal, pull generated videos, and upload them to FastVideo/ssim-reference-videos so the test has a regression baseline", "path": "seed-ssim-references/SKILL.md", "status": "draft", "trust": "low"}
+{"name": "reseed-ssim-references", "description": "Re-seed (overwrite) HF reference videos for an existing fastvideo/tests/ssim/ test and a single model id on Modal L40S. Always backs up current refs first, regenerates on Modal, pauses for the user to eyeball before-vs-after, then uploads with --force scoped to --model-id. Sister skill to seed-ssim-references; use when intentional code change has invalidated existing refs", "path": "reseed-ssim-references/SKILL.md", "status": "draft", "trust": "low"}
+{"name": "reseed-performance-baseline", "description": "Re-seed the HF performance-tracking baseline for an intentional runtime, dependency, or environment-caused benchmark shift. Use when performance CI fails because metrics such as latency, throughput, component time, or peak memory changed for an accepted reason and the rolling median baseline must be advanced by replicating one reviewed shifted source result into three success=true records, or five records when explicitly requested", "path": "reseed-performance-baseline/SKILL.md", "status": "draft", "trust": "low"}

.agents/skills/launch-experiment/SKILL.md

@@ -12,7 +12,7 @@ automates the boilerplate of setting environment variables, picking the right
 entrypoint, and applying defaults from the closest example script.
 
 ## Prerequisites
-- The repo is cloned and `fastvideo` is installed (`uv pip install -e .[dev]`).
+- The repo is cloned and `fastvideo` is installed (`uv pip install -e ".[dev]"`).
 - Dataset is preprocessed (see `docs/training/data_preprocess.md`).
 - `WANDB_API_KEY` is set in the environment (or `WANDB_MODE=offline` for local).
 - GPU resources are available (multi-GPU requires NCCL).

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

@@ -0,0 +1,426 @@
+---
+name: reseed-performance-baseline
+description: Re-seed the HF performance-tracking baseline for an intentional runtime, dependency, or environment-caused benchmark shift. Use when performance CI fails because metrics such as latency, throughput, component time, or peak memory changed for an accepted reason and the rolling median baseline in FastVideo/performance-tracking must be advanced by replicating one reviewed shifted source result into three success=true records, or five records when explicitly requested.
+---
+
+# Re-seed Performance Baseline
+
+## Purpose
+
+Replace or advance the rolling performance baseline for a single
+`(model_id, gpu_type)` pair in the HF dataset
+`FastVideo/performance-tracking`.
+
+Performance comparison uses the median of up to the last 5 successful records
+for the same model and GPU. Failed records are useful audit history, but they
+do not move the future baseline because `compare_baseline.py` loads records
+with `successful_only=True`.
+
+For a 5-record median, one shifted record is not enough to move the median if
+the other four records are from the old runtime. This skill therefore creates
+3 reviewed `success=true` records from one accepted shifted source result by
+default. If the user explicitly asks for a full reset, create 5 records.
+
+These replicated records are an intentional operator-approved baseline reset,
+not independent measurements. Mark them clearly with provenance fields so the
+HF history remains auditable.
+
+Use this skill when a performance test fails for an intentional and reviewed
+reason, such as a torch/runtime/container upgrade that legitimately increases
+peak memory or changes timings. This is the performance equivalent of
+`reseed-ssim-references`: backup first, scope tightly, require explicit human
+approval, then upload reviewed accepted baseline records.
+
+## When to use
+
+- A PR or main run failed the rolling performance comparison by more than the
+  allowed regression threshold, and maintainers agree the shift is caused by
+  an intentional runtime, dependency, hardware image, or benchmark environment
+  change rather than a FastVideo logic regression.
+- One shifted source result has been reviewed and accepted, and the operator
+  wants to replicate it into 3 successful records so the rolling median moves
+  immediately. Use 5 records only when the user explicitly asks to fully reset
+  the last-5 window.
+
+## When not to use
+
+- The benchmark failure might be a real code regression. Fix or investigate
+  the code path first.
+- The fixed benchmark thresholds in
+  `.buildkite/performance-benchmarks/tests/*.json` are too low. Those are a
+  separate gate from the rolling HF baseline and may need a code review change.
+- There is no clear source run, commit, and rationale. Baseline history is a
+  production signal; do not edit it without provenance.
+
+## Inputs
+
+| Parameter | Required | Description |
+|-----------|----------|-------------|
+| `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_result` | Yes | Path or Buildkite artifact URL for one accepted shifted performance JSON. Prefer the normalized `normalized_perf_*.json` artifact emitted by `compare_baseline.py`. |
+| `replica_count` | No | Number of success records to create from `source_result`. Default: `3`. Only use `5` if the user explicitly asks for a full reset. |
+| `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:
+
+- HF repo: `FastVideo/performance-tracking` (`HF_REPO_ID` override is
+  supported by the code, but use the default unless the user explicitly asks).
+- Local sync root: `/tmp/perf-tracking` or a timestamped local backup under
+  `performance_reseed_backup/`.
+- Baseline window: last 5 `success=true` records for the same
+  `(model_id, gpu_type)`.
+- Default reseed count: 3 replicated `success=true` records from one reviewed
+  source result. Explicit full-reset count: 5.
+
+## Steps
+
+### 1. Validate the target and source result
+
+If `source_result` is a Buildkite artifact URL, download it first into a
+local scratch directory such as `performance_reseed_source/` and use that
+downloaded JSON path for the rest of the workflow. If the agent cannot access
+the artifact because Buildkite authentication is missing, ask the user to
+download the artifact manually and provide the local path.
+
+Prefer the normalized Buildkite artifact emitted by `compare_baseline.py`:
+
+```text
+perf_reports/results/normalized_perf_*.json
+```
+
+That file is already in the HF tracking schema. Load it directly and confirm
+it has the expected baseline fields:
+
+```python
+import json
+
+with open(source_result, encoding="utf-8") as f:
+    record = json.load(f)
+```
+
+If only the older raw `fastvideo/tests/performance/results/perf_*.json`
+artifact is available, normalize it with `compare_baseline.py`'s shared helper
+before continuing. Run this from the repository root with
+`PYTHONPATH=fastvideo/tests/performance` so the script-local `hf_store` import
+resolves the same way it does in CI:
+
+```python
+import json
+from compare_baseline import normalize_performance_result
+
+with open(source_result, encoding="utf-8") as f:
+    record = normalize_performance_result(json.load(f))
+```
+
+The raw-to-normalized helper maps:
+
+- `model_id` comes from `benchmark_id`.
+- `gpu_type` comes from `device`.
+- `memory` comes from `max_peak_memory_mb`.
+- `latency` comes from `avg_generation_time_s`.
+- `throughput` comes from `throughput_fps`.
+- component timings come from the raw `text_encoder_time_s`, `dit_time_s`,
+  and `vae_decode_time_s` fields when present. If an older raw artifact lacks
+  those keys, they normalize to `None`; that source can still reseed latency,
+  throughput, and memory, but it cannot move component-time baselines.
+
+Stop if the normalized record's `model_id` or `gpu_type` does not match the
+requested `model_id` and `gpu_type`.
+
+The source record may have `success: false` when it came from a failed rolling
+baseline comparison. That is expected; only the reviewed reseed replicas become
+new `success: true` baseline records after explicit approval.
+
+Set `replica_count` to `3` by default. Set it to `5` only when the user
+explicitly asks to upload the same shifted source result 5 times for a full
+last-5 reset. Reject other counts unless the user gives a concrete reason.
+
+Check that `HF_API_KEY` is exported. The sync path may be public, but the
+upload path requires write access.
+
+### 1a. How to obtain `source_result` from CI
+
+The performance CI exports normalized source results for failed rolling
+baseline comparisons when `compare_baseline.py` ran. The preferred artifact
+comes from:
+
+```text
+perf_reports/results/normalized_perf_*.json
+```
+
+and is uploaded by Buildkite with the performance reports. The normal operator
+flow is:
+
+1. Open the failed Buildkite performance job.
+2. Download the `normalized_perf_*.json` artifact for the failed benchmark.
+3. Pass the local path or artifact URL as `source_result`.
+
+Do not scrape the Markdown performance summary to reconstruct the JSON. The
+normalized JSON artifact is the source of truth for reseed metrics and
+provenance. If only a raw `fastvideo/tests/performance/results/perf_*.json`
+artifact is present, normalize it with `normalize_performance_result()` before
+continuing. If no JSON artifact is present, the benchmark likely failed before
+writing results, so that run is not a valid source for baseline reseeding.
+
+### 2. Sync and back up existing HF records
+
+Use `fastvideo/tests/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.
+
+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)'
+```
+
+Then back up only the sanitized model directory:
+
+```bash
+SHORT_COMMIT=$(git rev-parse --short=12 HEAD)
+TIMESTAMP=$(date -u +%Y%m%d_%H%M%S)
+MODEL_SAFE=$(python - <<'PY'
+from fastvideo.tests.performance.hf_store import sanitize
+print(sanitize("<model_id>"))
+PY
+)
+BACKUP_DIR="performance_reseed_backup/${TIMESTAMP}_${SHORT_COMMIT}_${MODEL_SAFE}"
+mkdir -p "$BACKUP_DIR"
+cp -R "${PERFORMANCE_TRACKING_ROOT}/${MODEL_SAFE}" "$BACKUP_DIR/" 2>/dev/null || true
+```
+
+Write provenance next to the backup:
+
+```bash
+cat > "$BACKUP_DIR/PROVENANCE.txt" <<EOF
+model_id: <model_id>
+gpu_type: <gpu_type>
+source_result: <source_result>
+replica_count: <3_or_5>
+head_commit: $(git rev-parse HEAD)
+timestamp_utc: $(date -u +%FT%TZ)
+reason: <intent_rationale>
+EOF
+```
+
+If the backup has no prior records, this is not a destructive reseed; it is a
+first baseline seed. Continue, but report that baseline history was empty.
+
+### 3. Compute old baseline and candidate shift
+
+Load the last 5 successful records for the target:
+
+```python
+from fastvideo.tests.performance.hf_store import load_records_for_model
+
+records = load_records_for_model(
+    "/tmp/perf-tracking",
+    "<model_id>",
+    "<gpu_type>",
+    last_n=5,
+    successful_only=True,
+)
+```
+
+Print a small table showing the source result metrics, the replicated
+candidate median, and the old medians for:
+
+- `latency`
+- `throughput`
+- `memory`
+- `text_encoder_time_s`
+- `dit_time_s`
+- `vae_decode_time_s`
+
+Also print how many successful old records exist. Make clear:
+
+- 1 shifted record only seeds audit history and usually does not move the
+  median.
+- 3 replicated shifted records in a 5-record window move the median
+  immediately.
+- 5 replicated shifted records fully reset the rolling window to the source
+  result's runtime profile.
+- Replicated records are not independent measurements; they are an intentional
+  approved baseline reset and must be labeled that way.
+
+### 4. Confirm intent
+
+Require an explicit confirmation phrase before preparing the upload:
+
+> About to RE-SEED performance baseline for `<model_id>` on `<gpu_type>`.
+> This will upload `<N>` new `success=true` records to
+> `FastVideo/performance-tracking/<sanitize(model_id)>/`.
+>
+> Reason: `<intent_rationale>`
+> Source result: `<source_result>`
+> Replica count: `<replica_count>`
+> Note: these records replicate one reviewed measurement to force the rolling
+> median to the accepted runtime profile.
+> HEAD: `<git rev-parse --short=12 HEAD>`
+> Backup: `<BACKUP_DIR>`
+>
+> Reply `confirm performance reseed` to proceed, anything else to abort.
+
+Do not continue unless the user types exactly `confirm performance reseed`.
+
+### 5. Create the accepted seed records
+
+Create `replica_count` normalized records from the single source result. Use
+an explicit allowlist; do not copy the raw result JSON wholesale.
+
+Each record must include only these baseline fields plus the reseed provenance
+fields below:
+
+- `model_id`
+- `timestamp`
+- `commit_sha`
+- `gpu_type`
+- `latency`
+- `throughput`
+- `memory`
+- `text_encoder_time_s`
+- `dit_time_s`
+- `vae_decode_time_s`
+- `success: true`
+
+For normalized `normalized_perf_*.json` sources, these fields already exist.
+For older raw `perf_*.json` sources, map the raw fields exactly as
+`normalize_performance_result()` in `compare_baseline.py` does:
+
+| Normalized field | Raw source field |
+|------------------|------------------|
+| `model_id` | `benchmark_id` |
+| `gpu_type` | `device` |
+| `latency` | `avg_generation_time_s` |
+| `throughput` | `throughput_fps` |
+| `memory` | `max_peak_memory_mb` |
+| `text_encoder_time_s` | `text_encoder_time_s` |
+| `dit_time_s` | `dit_time_s` |
+| `vae_decode_time_s` | `vae_decode_time_s` |
+| `commit_sha` | `commit` |
+
+Do not upload raw-only fields such as `model_short_name`, `num_gpus`,
+`num_warmup_runs`, `num_measurement_runs`, `individual_times_s`,
+`individual_peak_memories_mb`, `thresholds`, or `pr_number`.
+
+Optional provenance fields are allowed and useful:
+
+- `baseline_reseed: true`
+- `baseline_reseed_reason`
+- `baseline_reseed_source_result`
+- `baseline_reseed_source_timestamp`
+- `baseline_reseed_replicated_source: true`
+- `baseline_reseed_batch_size`
+- `baseline_reseed_batch_index`
+- `baseline_reseed_operator`
+
+Use a fresh reseed timestamp for each replicated record, not the original
+source result timestamp. This is required because
+`load_records_for_model(..., last_n=5)` keeps the last records after loading
+the model directory; stale filenames/timestamps may not enter the last-5
+window and therefore may not move the median. Preserve the original source
+timestamp in `baseline_reseed_source_timestamp`.
+
+Use the existing filename convention from `_write_tracking_record()`:
+`<sanitize(timestamp)>_<sanitize(commit_sha)>.json` under the sanitized model
+directory, but include a deterministic suffix such as `_reseed_01`,
+`_reseed_02`, and `_reseed_03` before `.json` so the replicated files do not
+overwrite each other. For a 5-record full reset, continue through
+`_reseed_05`.
+
+If the source record already exists on HF with `success=false`, do not edit it
+in place unless the user explicitly asked for an audit-preserving correction.
+Prefer uploading new accepted seed records so failed history remains visible.
+
+### 6. Pause before upload
+
+Print:
+
+- Backup directory path.
+- HF paths that will receive the new records.
+- Old rolling medians.
+- Source metrics, replica count, and candidate median.
+- Rationale.
+
+Ask the user to reply exactly `upload`. Anything else aborts and leaves the
+prepared records plus backup on disk.
+
+### 7. Upload only the scoped records
+
+Use the shared storage helper so the path and repo type match CI:
+
+```python
+from fastvideo.tests.performance.hf_store import upload_record
+
+upload_record("<local_record_path>", record, strict=True)
+```
+
+Run it once per prepared record. Each upload goes to:
+
+```text
+FastVideo/performance-tracking/<sanitize(model_id)>/<record_filename>.json
+```
+
+Never bulk upload the whole tracking root. Never modify another model's
+directory in the same operation.
+
+### 8. Report outcome
+
+Report:
+
+- Uploaded HF paths.
+- Backup directory.
+- Old baseline window count and medians.
+- Source metrics, replica count, and candidate median.
+- Expected effect: 3 replicated shifted records move the 5-record median; 5
+  replicated shifted records fully reset the window to the accepted source
+  result.
+- Any separate threshold changes still needed in
+  `.buildkite/performance-benchmarks/tests/*.json`.
+
+Include the `intent_rationale` in the PR or follow-up comment so reviewers can
+distinguish an accepted baseline shift from a hidden regression.
+
+## Failure modes and handling
+
+- **`HF_API_KEY` unset.** Stop before upload. Do not create an untracked
+  process that appears to have reseeded but never reached HF.
+- **Source result does not match target.** Stop. The wrong benchmark or GPU
+  would poison a separate baseline.
+- **`replica_count` is 5 but the user did not explicitly ask for a full
+  reset.** Stop and use the default count of 3.
+- **The source result is noisy or suspicious.** Stop. Replicating one result
+  amplifies that measurement into the baseline, so it must be reviewed first.
+- **HF sync fails.** Stop for destructive reseeds. A stale or empty sync can
+  make the old baseline look missing.
+- **Candidate still violates fixed thresholds.** Report that this skill only
+  handles the rolling HF baseline; update benchmark JSON thresholds in code
+  review if maintainers accept the new absolute limit.
+- **The user aborts at either confirmation.** Leave the backup and prepared
+  records on disk. Nothing should be uploaded.
+- **A bad seed was uploaded.** Use the backup and HF history to identify the
+  uploaded file, then remove or supersede it with an explicitly reviewed
+  corrective record. Do not silently rewrite unrelated history.
+
+## References
+
+- `.agents/skills/reseed-ssim-references/SKILL.md` — safety pattern for
+  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,
+  `sanitize()`, and `upload_record()`.
+- `fastvideo/tests/performance/test_inference_performance.py` — source result
+  JSON schema.
+- `.buildkite/performance-benchmarks/tests/*.json` — fixed absolute benchmark
+  thresholds, separate from rolling baseline comparisons.
+
+## Changelog
+
+| Date | Change |
+|------|--------|
+| 2026-05-03 | Initial version. Sister workflow to `reseed-ssim-references`, scoped to one performance `(model_id, gpu_type)` baseline seed with backup, confirmation, provenance, and `success=true` upload. |
+| 2026-05-03 | Current policy: replicate one approved shifted source result into 3 success records by default, or 5 only when explicitly requested. Add provenance marker for replicated-source reseeds. |

.agents/skills/reseed-ssim-references/SKILL.md

@@ -0,0 +1,343 @@
+---
+name: reseed-ssim-references
+description: Re-seed HF reference videos for a single existing SSIM test on Modal L40S. Always backs up current refs locally first, regenerates on Modal, pauses for the user to eyeball before-vs-after quality, then overwrites the targeted `<model_id>` subtree on `FastVideo/ssim-reference-videos` with `--force`. Use when an intentional code change (model port fix, attention backend swap, kernel upgrade, hyperparameter change) has invalidated existing refs and they need to be regenerated. Pairs with `seed-ssim-references`, which is for first-time seeding only.
+---
+
+# Re-seed SSIM Reference Videos
+
+## Purpose
+
+Replace the existing SSIM reference videos for a single `(test_file, model_id)`
+pair on the HF dataset (`FastVideo/ssim-reference-videos`). This is **destructive**
+on HF — the old refs are overwritten — so the skill always:
+
+1. Confirms intent with a one-liner the user has to type.
+2. Downloads the existing refs as a local, timestamped backup.
+3. Regenerates on Modal L40S (same code path that CI uses).
+4. Pauses for a side-by-side eyeball of backup vs new mp4s.
+5. Uploads with `--force`, scoped to the single `--model-id`.
+6. Reminds the user to keep the backup until the PR lands.
+
+Pairs with `seed-ssim-references`, which is the inverse (first-time seeding
+only, refuses to overwrite). Re-seeding is intentionally a separate, more
+ceremonial operation because mistakenly clobbering production refs is much
+harder to recover from than failing closed.
+
+## When to use
+
+- An intentional code change (model port fix, kernel upgrade, attention
+  backend swap, hyperparameter change in the test itself) has shifted the
+  expected SSIM output and the existing refs no longer represent the new
+  ground truth.
+- A test is failing in CI **for the right reason** (the new code is correct,
+  the old refs are stale).
+
+## When not to use
+
+- A test is failing for the **wrong** reason (the port is buggy, not the
+  refs). Fix the port; re-seeding hides the bug.
+- A brand-new test that has no refs on HF yet. Use `seed-ssim-references`.
+- "Just to clean up drift" without a concrete code change to point at. The
+  PR description has to justify *why* refs changed; without a concrete
+  change, there's nothing to write.
+
+## Inputs
+
+| Parameter | Required | Description |
+|-----------|----------|-------------|
+| `test_file` | Yes | Path to the SSIM test, e.g. `fastvideo/tests/ssim/test_matrixgame_similarity.py`. Validated against `fastvideo/tests/ssim/test_*_similarity.py`. |
+| `model_id` | Yes | Single model id from the test's `*_MODEL_TO_PARAMS`, e.g. `Matrix-Game-2.0-Diffusers-Base`. Re-seed runs are **per model**. For multi-model tests, invoke the skill once per model. |
+| `intent_rationale` | Yes | One-line explanation of *why* refs are being regenerated (e.g. "Relax FA-2 head_size whitelist to include 80 — matrix_game now uses FLASH_ATTN instead of TORCH_SDPA"). Recorded in the backup directory and reused in the PR description. |
+
+Hardcoded:
+
+- Modal GPU: **L40S** (matches CI; re-seeding from another SKU produces refs
+  that L40S CI cannot match).
+- Quality tier: **`default`**. `full_quality` is a separate, deliberate
+  operation.
+- HF repo: `FastVideo/ssim-reference-videos` (override via
+  `FASTVIDEO_SSIM_REFERENCE_HF_REPO`).
+- Device folder: `L40S_reference_videos`.
+
+## Prerequisites
+
+The user has confirmed:
+
+- `modal` CLI authenticated.
+- `hf` CLI authenticated, **and** `HF_API_KEY` (or `HUGGINGFACE_HUB_TOKEN` /
+  `HF_TOKEN`) exported with **write** access to
+  `FastVideo/ssim-reference-videos`.
+- The current branch's code is the change that motivated the re-seed (i.e.
+  `git rev-parse HEAD` is the commit that intentionally invalidated refs).
+
+Fail fast if any of these are missing.
+
+## Steps
+
+### 1. Validate inputs and confirm intent
+
+- Verify `test_file` exists and matches `fastvideo/tests/ssim/test_*_similarity.py`.
+- Grep the file for `*_MODEL_TO_PARAMS` and assert `model_id` is one of its
+  keys. If the file has only a single hardcoded model, accept that model id
+  as the only valid value.
+- Print the rationale and ask the user to type **`confirm reseed`** (not just
+  `y` — make it deliberate):
+
+  > About to RE-SEED references for model `<model_id>` from test `<test_file>`.
+  > This will OVERWRITE existing refs on
+  > `FastVideo/ssim-reference-videos/reference_videos/default/L40S_reference_videos/<model_id>/`
+  > after backup + Modal regen + eyeball.
+  >
+  > Reason: `<intent_rationale>`
+  > HEAD: `<git rev-parse --short=12 HEAD>`
+  >
+  > Reply `confirm reseed` to proceed, anything else to abort.
+
+  Stop until the user types exactly `confirm reseed`. Anything else aborts
+  with no side effects.
+
+### 2. Back up existing refs
+
+Always required. The backup is the only graceful path back if anything goes
+wrong later.
+
+```bash
+SHORT_COMMIT=$(git rev-parse --short=12 HEAD)
+TIMESTAMP=$(date -u +%Y%m%d_%H%M%S)
+MODEL_SAFE=$(echo "<model_id>" | tr '/' '_')
+BACKUP_DIR="ssim_reseed_backup/${TIMESTAMP}_${SHORT_COMMIT}_${MODEL_SAFE}"
+mkdir -p "$BACKUP_DIR"
+
+hf download \
+    --repo-type dataset FastVideo/ssim-reference-videos \
+    --include "reference_videos/default/L40S_reference_videos/<model_id>/**" \
+    --local-dir "$BACKUP_DIR"
+
+mp4_count=$(find "$BACKUP_DIR" -name "*.mp4" | wc -l)
+echo "Backup mp4 count: $mp4_count"
+[ "$mp4_count" -gt 0 ] || {
+    echo "ERROR: backup is empty for <model_id>. Either the model id is wrong"
+    echo "or there are no existing refs (use seed-ssim-references instead)."
+    exit 1
+}
+
+# Provenance — used in the PR description
+cat > "$BACKUP_DIR/PROVENANCE.txt" <<EOF
+test_file: <test_file>
+model_id: <model_id>
+head_commit: $(git rev-parse HEAD)
+timestamp_utc: $(date -u +%FT%TZ)
+reason: <intent_rationale>
+EOF
+```
+
+If the `hf download` produces zero mp4s, abort — the user has either picked a
+non-existent `model_id` or there are no refs yet (in which case
+`seed-ssim-references` is the right tool).
+
+### 3. Regenerate on Modal L40S
+
+Mirror CI's exact env recipe so the regenerated refs are byte-comparable to
+what CI will produce on the same commit. Two differences from CI:
+
+1. **Pass the same env prefix CI uses** (`IMAGE_VERSION`, `BUILDKITE_*`) — see
+   `.buildkite/pipeline.yml:1-3` and `.buildkite/scripts/pr_test.sh:62-83`.
+   Without this, `ssim_test.py:17-18` resolves a different GHCR image tag
+   (default is `latest`, CI is `py3.12-latest`), and `ssim_test.py:38-46`
+   bakes different values into the image's frozen env block. **Mismatched
+   image or env is the most common source of SSIM drift between reseed and
+   CI runs.**
+2. **Do not pass `--skip-reference-download`**. Letting the test fetch the
+   existing refs and run the full SSIM compare gives "before" SSIM numbers
+   for the PR description, and the test still produces the new mp4s
+   regardless of whether the comparison passes or fails.
+
+```bash
+SUBDIR="${TIMESTAMP}_${SHORT_COMMIT}"
+
+IMAGE_VERSION="py3.12-latest" \
+BUILDKITE_REPO="$(git config --get remote.origin.url)" \
+BUILDKITE_COMMIT="$(git rev-parse HEAD)" \
+BUILDKITE_PULL_REQUEST="${BUILDKITE_PULL_REQUEST:-false}" \
+modal run fastvideo/tests/modal/ssim_test.py \
+    --git-repo="$(git config --get remote.origin.url)" \
+    --git-commit="$(git rev-parse HEAD)" \
+    --hf-api-key="$HF_API_KEY" \
+    --test-files="<test_file>" \
+    --sync-generated-to-volume \
+    --generated-volume-subdir="$SUBDIR" \
+    --no-fail-fast
+```
+
+Capture the printed `modal volume get ...` hint — its `<SUBDIR>` matches
+`$SUBDIR` and is needed for step 4. Capture the SSIM numbers from the test
+output (or from the JSON next to the generated mp4) for the PR description.
+
+### 4. Download generated videos
+
+```bash
+modal volume get --force hf-model-weights \
+    ssim_generated_videos/default/"$SUBDIR"/generated_videos \
+    ./generated_videos_modal/default
+```
+
+After this, the new mp4s live at:
+
+```
+./generated_videos_modal/default/generated_videos/L40S_reference_videos/<model_id>/<backend>/<prompt>.mp4
+```
+
+`--force` is required when `./generated_videos_modal/default` already exists
+from a prior run; safe on the first run too.
+
+### 5. PAUSE — user reviews quality side-by-side
+
+Print the diff and the comparison:
+
+```bash
+echo "=== File list diff (backup vs new) ==="
+diff -u \
+    <(find "$BACKUP_DIR/reference_videos/default/L40S_reference_videos/<model_id>" -name "*.mp4" \
+        | sed "s|$BACKUP_DIR/reference_videos/default/L40S_reference_videos/||" | sort) \
+    <(find ./generated_videos_modal/default/generated_videos/L40S_reference_videos/<model_id> -name "*.mp4" \
+        | sed "s|./generated_videos_modal/default/generated_videos/L40S_reference_videos/||" | sort) \
+    || true
+
+echo
+echo "=== SSIM numbers from this run (paste into PR) ==="
+find ./generated_videos_modal/default/generated_videos/L40S_reference_videos/<model_id> -name "*_ssim.json" -exec cat {} \;
+```
+
+Then stop and tell the user:
+
+> Old refs backed up to `$BACKUP_DIR`.
+> New videos in `./generated_videos_modal/default/generated_videos/L40S_reference_videos/<model_id>/`.
+>
+> Open both in a video player. Confirm the new videos:
+>   1. Look correct (no obvious artifacts, no black/static frames).
+>   2. Are *intentionally* different from the backup in the way described
+>      in `<intent_rationale>` (e.g. slight numerical drift only, not a
+>      different scene / different motion / corrupted output).
+>
+> Reply **`upload`** to overwrite HF, anything else to abort.
+> Aborting leaves the backup and new videos on disk for inspection — nothing
+> on HF changes.
+
+Do not proceed until the user types exactly `upload`. If they abort, leave
+everything on disk and stop here.
+
+### 6. Copy into the local reference layout
+
+Same as `seed-ssim-references` step 5:
+
+```bash
+python fastvideo/tests/ssim/reference_videos_cli.py copy-local \
+    --quality-tier default \
+    --device-folder L40S_reference_videos \
+    --generated-dir ./generated_videos_modal/default/generated_videos/L40S_reference_videos
+```
+
+Result: `fastvideo/tests/ssim/reference_videos/default/L40S_reference_videos/<model_id>/<backend>/<prompt>.mp4`.
+
+### 7. Upload with `--force`, scoped to `--model-id`
+
+The `--force` flag is what makes this skill different from `seed-ssim-references`.
+Always pair it with `--model-id` so a typo cannot accidentally overwrite a
+neighboring model's refs.
+
+```bash
+python fastvideo/tests/ssim/reference_videos_cli.py upload \
+    --quality-tier default \
+    --device-folder L40S_reference_videos \
+    --model-id "<model_id>" \
+    --force
+```
+
+The CLI's overwrite guard refuses without `--force`; with `--force` it
+overwrites only files under
+`reference_videos/default/L40S_reference_videos/<model_id>/`.
+
+### 8. Report success and retention guidance
+
+Print:
+
+- The HF path that was overwritten (`<repo>/reference_videos/default/L40S_reference_videos/<model_id>/`).
+- The local backup directory path.
+- The new SSIM numbers from step 5.
+- This restore command, in case the PR review surfaces a problem after
+  upload:
+
+  ```bash
+  python fastvideo/tests/ssim/reference_videos_cli.py upload \
+      --quality-tier default \
+      --device-folder L40S_reference_videos \
+      --model-id "<model_id>" \
+      --reference-dir "$BACKUP_DIR/reference_videos/default/L40S_reference_videos" \
+      --force
+  ```
+
+- This PR-description checklist (see `fastvideo/tests/ssim/AGENTS.md` →
+  *Updating Reference Videos*):
+  1. Source commit that produced the new refs (HEAD at re-seed time).
+  2. Test command and GPU SKU (`L40S`).
+  3. Before/after SSIM numbers.
+  4. The `<intent_rationale>` from step 1.
+  5. A note that the backup lives at `$BACKUP_DIR` and should be retained
+     until CI on the PR is green.
+
+Do **not** auto-rerun the SSIM test — the user does that as part of the PR.
+
+## Failure modes and how to handle them
+
+- **`HF_API_KEY` unset.** Stop before step 2.
+- **Backup is empty (zero mp4s).** Stop before step 3 — the model id is
+  wrong or the refs don't exist yet (use `seed-ssim-references`).
+- **Modal run fails before generation.** No mp4s on the volume. Don't
+  upload. Investigate the failure (test crash, OOM, partition exhaustion),
+  fix, then retry from step 3. Backup is still intact.
+- **Quality regressed (visual or metric).** User aborts at step 5. Backup
+  retained. New videos retained on disk for inspection. Nothing on HF
+  changed. Either fix the underlying code change or abandon the re-seed.
+- **User confirmed `upload` but later realized the new refs are wrong.**
+  Run the restore command from step 8 with the backup `--reference-dir`.
+  This is exactly why the backup exists.
+- **Multi-model test, only one model is being re-seeded.** Run the skill
+  once per model id. The `--model-id` scope on upload guarantees the others
+  are untouched.
+
+## Design notes (for future skill maintainers)
+
+- Per-`model_id` scope is mandatory. The dataset houses many model subtrees;
+  re-seeding the wrong one is hard to undo without backup.
+- `default` tier only; `full_quality` is a separate, deliberate operation
+  with different params and ~doubled runtime, and isn't what CI gates on.
+- The skill deliberately does **not** pass `--skip-reference-download` to
+  Modal so we get pre-reseed SSIM numbers for the PR. The `seed`-skill
+  passes it because no refs exist yet; for re-seed, refs do exist and
+  exposing the comparison is informative.
+- The two-token confirm (`confirm reseed`, then `upload`) is intentional.
+  Re-seeding is high-blast-radius and should not be one-keystroke.
+- The backup directory is plain mp4s + `PROVENANCE.txt`. No HF metadata is
+  preserved; the restore path uses `reference_videos_cli.py upload
+  --reference-dir` which doesn't need it.
+
+## References
+
+- `.agents/skills/seed-ssim-references/SKILL.md` — the first-time seed
+  skill this one parallels. Read it for the Modal flag rationale shared
+  between the two flows.
+- `fastvideo/tests/ssim/AGENTS.md` — directory rules, including the PR
+  expectations for any reference-video change (rationale, before/after
+  SSIM, source commit/model/backend).
+- `fastvideo/tests/ssim/reference_videos_cli.py` — `copy-local`, `upload`
+  (with `--model-id`, `--force`), `download`. The overwrite guard at
+  `upload_reference_videos` is the safety net this skill leans on.
+- `fastvideo/tests/modal/ssim_test.py` — Modal orchestrator;
+  `--sync-generated-to-volume`, `--generated-volume-subdir`,
+  `--skip-reference-download`, `--no-fail-fast`.
+
+## Changelog
+
+| Date | Change |
+|------|--------|
+| 2026-05-02 | Initial version. Sister skill to `seed-ssim-references`, scoped to single `(test_file, model_id)` re-seeds, with mandatory backup and two-token confirm. |

.agents/skills/seed-ssim-references/SKILL.md

@@ -1,26 +1,41 @@
 ---
 name: seed-ssim-references
-description: Seed HF reference videos for a single newly-added SSIM test. Runs the test on Modal L40S, downloads the generated mp4s via `modal volume get`, pauses for the user to eyeball quality, then uploads only that test's files to `FastVideo/ssim-reference-videos`. Use when a new `fastvideo/tests/ssim/test_*_similarity.py` has just been added and has no references on HF yet.
+description: Seed HF reference artefacts for a single newly-added SSIM test (pixel `.mp4` for `run_text_to_video_similarity_test`-style tests, or latent `.pt` for `run_text_to_latent_similarity_test`-style tests). Runs the test on Modal L40S, downloads the generated artefacts via `modal volume get`, pauses for the user to verify (visual eyeball for mp4, numerics dump for pt), then uploads only that test's files to `FastVideo/ssim-reference-videos`. Use when a new `fastvideo/tests/ssim/test_*_similarity.py` has just been added and has no references on HF yet.
 ---
 
-# Seed SSIM Reference Videos
+# Seed SSIM Reference Artefacts (mp4 or pt)
 
 ## Purpose
 
 A brand-new SSIM test in `fastvideo/tests/ssim/` fails forever until its
-reference videos exist on the HF dataset (`FastVideo/ssim-reference-videos`).
+reference artefacts exist on the HF dataset
+(`FastVideo/ssim-reference-videos`). The dataset hosts two kinds of artefacts
+side-by-side per `(model_id, backend, prompt)`:
+
+- **`.mp4`** — pixel ground-truth for tests that call
+  `run_text_to_video_similarity_test` / `run_image_to_video_similarity_test`
+  in `inference_similarity_utils.py`. Compared via SSIM.
+- **`.pt`** — pre-VAE latent bundle (fp16 full latent + fp32 slice +
+  metadata + `slice_spec` + `format_version`) for tests that call
+  `run_text_to_latent_similarity_test` in `latent_similarity_utils.py`.
+  Compared via cosine distance on the slice and the full tensor.
+
 This skill:
 
-1. Runs the test on Modal's L40S pool to generate the videos.
-2. Downloads them to the local repo via `modal volume get`.
-3. Pauses so the user can eyeball the mp4s and confirm quality.
-4. Uploads only the new test's files to HF, with a guard that refuses to
+1. Detects which artefact type the test produces (pixel vs latent).
+2. Runs the test on Modal's L40S pool to generate the artefacts.
+3. Downloads them to the local repo via `modal volume get`.
+4. Pauses so the user can verify quality:
+   - **mp4**: visual eyeball in a video player.
+   - **pt**: numerics dump (shape, slice stats, NaN/Inf check, metadata).
+5. Uploads only the new test's files to HF, with a guard that refuses to
    overwrite anything already present.
 
 The skill is run **manually**, once per new test. Before invoking it, the user
 has already sanity-tested the new test locally — it launches `VideoGenerator`
-and writes an mp4 without crashing. The skill does not re-test locally; it
-goes straight to Modal L40S (which is what CI uses).
+and writes an artefact without crashing (the missing-reference assertion at
+the end is expected). The skill does not re-test locally; it goes straight
+to Modal L40S (which is what CI uses).
 
 ## When to use
 
@@ -69,7 +84,7 @@ Fail fast if the token env var is missing.
 
 ## Steps
 
-### 1. Ask for the test file
+### 1. Ask for the test file, then detect artefact type
 
 If the user didn't name one, ask: *"Which SSIM test file do you want to seed
 references for? (e.g. `fastvideo/tests/ssim/test_ltx2_similarity.py`)"*.
@@ -80,6 +95,22 @@ Validate:
 - File defines a `*_MODEL_TO_PARAMS` dict — grep it to extract the set of
   model ids. Those ids drive step 5.
 
+Detect artefact type by inspecting the file's imports / helper call:
+
+- **latent** (`.pt`) — file imports `run_text_to_latent_similarity_test`
+  from `fastvideo.tests.ssim.latent_similarity_utils` (or any other helper
+  that ends with `_latent_similarity_test`).
+- **pixel** (`.mp4`) — file imports
+  `run_text_to_video_similarity_test` / `run_image_to_video_similarity_test`
+  from `fastvideo.tests.ssim.inference_similarity_utils`, OR uses the
+  legacy custom-inline helper pattern (see `test_gamecraft`,
+  `test_longcat`, etc.). Default to pixel when both heuristics fail.
+
+Record `ARTEFACT_TYPE ∈ {pixel, latent}` for use in step 4. Steps 2, 3, 5,
+and 6 are artefact-type-agnostic — `_iter_reference_files`,
+`copy_generated_to_reference`, and `upload_reference_videos` already walk
+both `.mp4` and `.pt` (see `reference_videos_cli.py`).
+
 If either check fails, stop and tell the user what's wrong.
 
 ### 2. Run the test on Modal L40S
@@ -92,9 +123,19 @@ TIMESTAMP=$(date -u +%Y%m%d_%H%M%S)
 SUBDIR="${TIMESTAMP}_${SHORT_COMMIT}"
 ```
 
-Then launch the Modal run:
+Then launch the Modal run. The `IMAGE_VERSION` and `BUILDKITE_*` env-prefix
+**must** match what CI exports in `.buildkite/scripts/pr_test.sh`, otherwise
+`fastvideo/tests/modal/ssim_test.py` resolves a different GHCR image tag
+(default is `latest`, CI is `py3.12-latest`) and bakes different values into
+the image's frozen env block (`ssim_test.py:17-18, 38-46`). Mismatched image
+or env produces SSIM drift that doesn't show up until the same commit runs
+in CI.
 
 ```bash
+IMAGE_VERSION="py3.12-latest" \
+BUILDKITE_REPO="$(git config --get remote.origin.url)" \
+BUILDKITE_COMMIT="$(git rev-parse HEAD)" \
+BUILDKITE_PULL_REQUEST="${BUILDKITE_PULL_REQUEST:-false}" \
 modal run fastvideo/tests/modal/ssim_test.py \
     --git-repo="$(git config --get remote.origin.url)" \
     --git-commit="$(git rev-parse HEAD)" \
@@ -106,6 +147,19 @@ modal run fastvideo/tests/modal/ssim_test.py \
     --no-fail-fast
 ```
 
+Env prefix rationale (parity with CI; see `.buildkite/pipeline.yml:1-3` and
+`.buildkite/scripts/pr_test.sh:62-83`):
+- `IMAGE_VERSION=py3.12-latest`: pins the Modal image tag to the same one CI
+  uses. Without this, `ssim_test.py:17` falls back to `latest`, which on
+  GHCR is built from `Dockerfile.python3.10` — different Python, torch, and
+  flash-attn wheel than CI's `py3.12-latest` (`infra-build-image.yml:51-67`,
+  `_template-build-image.yml:65-101`).
+- `BUILDKITE_REPO`/`BUILDKITE_COMMIT`/`BUILDKITE_PULL_REQUEST`: mirror what
+  Buildkite exports. `ssim_test.py:38-46` bakes these into the image's
+  `.env(...)` block; mismatched values can perturb in-container code paths
+  that branch on PR-vs-non-PR. `false` for `BUILDKITE_PULL_REQUEST` matches
+  Buildkite's "non-PR build" sentinel.
+
 Flag rationale:
 - `--skip-reference-download`: no refs exist yet, so conftest must not try to
   pull them.
@@ -143,17 +197,59 @@ get` preserves that trailing `generated_videos/` segment.
 
 ### 4. PAUSE — user reviews quality
 
-Print the list of downloaded mp4s and their paths, then stop. Tell the user:
+Type-aware verification.
+
+**For `ARTEFACT_TYPE = pixel`** — list the downloaded mp4s and ask the user to
+open them in a video player:
 
 > "Generated videos downloaded to `./generated_videos_modal/default/generated_videos/L40S_reference_videos/`. Please open them and confirm the quality looks correct. Reply **`upload`** to continue, or anything else to abort."
 
+**For `ARTEFACT_TYPE = latent`** — `.pt` files are not human-watchable. Print
+a numerics dump for each `.pt` so the user can sanity-check shape, distribution,
+and metadata:
+
+```python
+import torch
+from pathlib import Path
+ROOT = Path("./generated_videos_modal/default/generated_videos/L40S_reference_videos")
+for p in sorted(ROOT.rglob("*.pt")):
+    d = torch.load(p, map_location="cpu", weights_only=False)
+    s = d["expected_slice"]
+    L = d["latent"].float()
+    print(f"=== {p.relative_to(ROOT)} ===")
+    print(f"  format_version: {d['format_version']}")
+    print(f"  shape:          {d['shape']}")
+    print(f"  dtype_original: {d['dtype_original']}")
+    print(f"  slice_spec:     {d['slice_spec']}")
+    print(f"  slice  shape={tuple(s.shape)}  mean={s.mean():+.4f}  std={s.std():.4f}  min={s.min():+.4f}  max={s.max():+.4f}")
+    print(f"  latent shape={tuple(L.shape)}  mean={L.mean():+.4f}  std={L.std():.4f}  min={L.min():+.4f}  max={L.max():+.4f}")
+    print(f"  finite: latent NaN={torch.isnan(L).any().item()} Inf={torch.isinf(L).any().item()}; "
+          f"slice NaN={torch.isnan(s).any().item()} Inf={torch.isinf(s).any().item()}")
+    print(f"  metadata: {d['metadata']}\n")
+```
+
+Sanity criteria:
+- `format_version == 1` (matches `LATENT_REFERENCE_FORMAT_VERSION`).
+- `shape` matches what the model produces (e.g. LTX-2 distilled =
+  `[1, 128, T_lat, H_lat, W_lat]`; Stable Audio Open 1.0 = `[1, 64, 1024]`).
+- `slice_spec.kind` matches a registered kind (`corner_3x3_first_frame`
+  for video, `audio_first_8_timesteps` for audio).
+- No `NaN`/`Inf`. `mean ≈ 0`, `std ≈ 1` (denoised latents stay close to
+  the initial Gaussian distribution; very wide deviations suggest
+  numerical drift).
+- `metadata.prompt` matches the test's prompt.
+
+Then ask:
+
+> "Numerics look right? Reply **`upload`** to continue, or anything else to abort."
+
 Do not proceed until the user explicitly says `upload`. If they abort, leave
 everything on disk so they can inspect further — no cleanup.
 
 ### 5. Copy into the local reference layout
 
-Scoped copy — only the new test's mp4s. Loop over each `<model_id>` extracted
-in step 1:
+Scoped copy — only the new test's artefacts. Single command works for both
+artefact types because `_iter_reference_files` walks `.mp4` and `.pt`:
 
 ```bash
 python fastvideo/tests/ssim/reference_videos_cli.py copy-local \
@@ -163,12 +259,13 @@ python fastvideo/tests/ssim/reference_videos_cli.py copy-local \
 ```
 
 (The `--generated-dir` points at the device-folder root inside the
-downloaded tree; `copy-local` walks all `<model>/<backend>/*.mp4`
+downloaded tree; `copy-local` walks all `<model>/<backend>/*.{mp4,pt}`
 underneath it. Since the Modal run was scoped to a single test file via
 `--test-files`, only that test's model(s) are present — so the copy is
 implicitly per-test.)
 
-Result: `fastvideo/tests/ssim/reference_videos/default/L40S_reference_videos/<model_id>/<backend>/<prompt>.mp4`.
+Result for pixel: `fastvideo/tests/ssim/reference_videos/default/L40S_reference_videos/<model_id>/<backend>/<prompt>.mp4`.
+Result for latent: same path with `.pt` extension.
 
 ### 6. Upload to HF — scoped per model_id, with overwrite guard
 
@@ -201,33 +298,54 @@ it will auto-download the refs they just uploaded.
 ## Failure modes and how to handle them
 
 - **`HF_API_KEY` unset.** Stop before step 2. The Modal run needs it (passed
-  via `--hf-api-key`), and step 6 needs it for upload.
-- **Modal run fails before generation.** No mp4s on the volume — nothing to
-  download. Fix the test locally (`pytest fastvideo/tests/ssim/<test_file>`)
+  via `--hf-api-key`), and step 6 needs it for upload. If the user
+  ran `hf auth login` instead of exporting an env var, read the cached
+  token via `huggingface_hub.get_token()` and forward it to Modal as
+  `--hf-api-key="$CACHED_TOKEN"`.
+- **Modal run fails before generation.** No artefacts on the volume — nothing
+  to download. Fix the test locally (`pytest fastvideo/tests/ssim/<test_file>`)
   and retry from step 2.
 - **`./generated_videos_modal/default/L40S_reference_videos/` missing after
-  `modal volume get`.** The run didn't produce videos (most likely the test
-  crashed before writing, or `REQUIRED_GPUS` exceeded the partition capacity
-  — see Modal logs).
+  `modal volume get`.** The run didn't produce artefacts (most likely the
+  test crashed before writing, or `REQUIRED_GPUS` exceeded the partition
+  capacity — see Modal logs).
+- **Latent test crashed with FSDP / inference_mode error
+  (`RuntimeError: Inference tensors do not track version counter`).** The
+  test must pass `init_kwargs_override={"use_fsdp_inference": False}` when
+  `sp_size == 1` — see `test_stable_audio_similarity.py` for the pattern.
+  Fix in the test, push, retry.
 - **Upload guard fires (files already exist).** The test name / model id
   collides with something already on HF. Verify the user actually wants to
   replace existing refs; if so, re-run the upload with `--force`. If not,
   rename the model id in `*_MODEL_TO_PARAMS` and re-seed.
-- **Quality looks wrong in step 4.** Abort. The mp4s stay on disk for
+- **Quality looks wrong in step 4.** Abort. The artefacts stay on disk for
   inspection. The fix is usually in the test's params (resolution, steps,
   seed) — edit the test, then re-run the skill.
+  - For latent: also check `slice_spec.kind` matches the latent rank
+    (`corner_3x3_first_frame` requires 5-D, `audio_first_8_timesteps`
+    requires 3-D); a rank/kind mismatch raises in `_extract_expected_slice`.
 
 ## Design notes (for future skill maintainers)
 
 - The skill deliberately runs on Modal, **not** locally, because the CI
   runner is L40S. Seeding from a different GPU SKU produces refs that CI's
-  L40S runs can't match (SSIM drifts across SKUs).
+  L40S runs can't match (pixel SSIM drifts across SKUs; latent cosine has
+  tighter cross-SKU bf16 drift but the configured tolerances assume
+  same-SKU seed → same-SKU verify).
 - The skill is default-tier only. `full_quality` refs are seeded by a
   separate, deliberate operation — they double runtime and aren't what CI
   gates on.
 - The overwrite guard in `reference_videos_cli.py upload` is default-on
   specifically because this skill exists. Re-seeding is a distinct operation
   that requires explicit `--force`.
+- Both artefact types share the same Modal flow: the orchestrator sets
+  `--skip-reference-download` + `--no-fail-fast`, runs pytest, the test's
+  helper writes the artefact (`.mp4` via `imageio` for pixel,
+  `save_latent_reference` → `torch.save` for latent) BEFORE the
+  missing-reference assertion raises. `_sync_generated_videos_to_volume` in
+  `ssim_test.py` does a `shutil.copytree` of the whole `generated_videos/`
+  tree, picking up `.mp4`, `.pt`, and the `*_ssim.json` / `*_latent.json`
+  metric files alongside.
 
 ## References
 
@@ -236,10 +354,17 @@ it will auto-download the refs they just uploaded.
   `--skip-reference-download`, `--no-fail-fast`.
 - `fastvideo/tests/ssim/reference_videos_cli.py` — `copy-local`, `upload`
   (with `--model-id`, `--force`), `download`, `ensure` subcommands.
+  Extension allowlist is `REFERENCE_EXTENSIONS = VIDEO_EXTENSIONS +
+  LATENT_EXTENSIONS` (`.pt`).
 - `fastvideo/tests/ssim/README.md` — reference layout, HF repo conventions.
-- `fastvideo/tests/ssim/inference_similarity_utils.py` —
-  `run_text_to_video_similarity_test` + `_build_init_kwargs`: what each test
-  config passes to `VideoGenerator.from_pretrained`.
+- `fastvideo/tests/ssim/inference_similarity_utils.py` — pixel helpers
+  (`run_text_to_video_similarity_test`,
+  `run_image_to_video_similarity_test`, `build_init_kwargs`).
+- `fastvideo/tests/ssim/latent_similarity_utils.py` — latent helper
+  (`run_text_to_latent_similarity_test`), slice spec dispatch
+  (`_extract_expected_slice`), reference schema
+  (`save_latent_reference` / `load_latent_reference`),
+  `LATENT_REFERENCE_FORMAT_VERSION`.
 
 ## Changelog
 
@@ -248,3 +373,4 @@ it will auto-download the refs they just uploaded.
 | 2026-04-17 | Initial version (Modal sync-to-volume flow). |
 | 2026-04-21 | Rewrite: single-test scope, explicit user-review pause, per-`model_id` upload, HF overwrite guard. Dropped `scripts/seed_ssim.sh`. |
 | 2026-04-21 | Post-first-run fixes: `modal volume get` needs `--force` when parent exists; download tree has an extra `generated_videos/` level so `--generated-dir` must reflect it. |
+| 2026-05-01 | Latent (`*.pt`) artefact support: artefact-type detection in step 1, type-aware verification (visual eyeball for mp4, numerics dump for pt) in step 4, FSDP+inference_mode failure-mode added, design notes for the unified Modal flow. Triggered by PR #1253 (LTX-2 latent migration + Stable Audio latent test). |

.agents/workflows/evaluation-development.md

@@ -4,24 +4,21 @@ description: How to develop, validate, and register a new evaluation metric
 
 # Evaluation Development SOP
 
-Standard procedure for adding new video quality evaluation metrics to
-the FastVideo agent toolkit.
+Standard procedure for adding new video quality evaluation metrics to the
+FastVideo agent toolkit.
 
-## When to use
+## When to Use
 
-- You need a metric that does not exist in
-  `.agents/memory/evaluation-registry/README.md`.
+- You need a metric that doesn't exist in `.agents/memory/evaluation-registry/README.md`.
 - An existing metric needs significant changes to its methodology.
-- You are exploring a new evaluation approach.
+- You're exploring a new evaluation approach.
 
 ## Steps
 
 ### 1. Research
 
-- Search `.agents/memory/related-work/` for existing evaluation
-  approaches.
-- Check `.agents/memory/evaluation-registry/README.md` for current
-  metrics and their limitations.
+- Search `.agents/memory/related-work/` for existing evaluation approaches.
+- Check the `evaluation_registry.md` for current metrics and their limitations.
 - Review literature: FVD, CLIP-Score, human preference, etc.
 
 ### 2. Prototype
@@ -32,25 +29,21 @@ the FastVideo agent toolkit.
 
 ### 3. Validate
 
-- **Known-good test**: metric should score high on reference-quality
-  videos.
-- **Known-bad test**: metric should score low on degraded or unrelated
-  videos.
-- **Sensitivity test**: small quality differences should produce
-  meaningful score differences.
+- **Known-good test**: Metric should score high on reference-quality videos.
+- **Known-bad test**: Metric should score low on degraded/unrelated videos.
+- **Sensitivity test**: Small quality differences should produce meaningful
+  score differences.
 - Document thresholds and their justification.
 
 ### 4. Register
 
 Update `.agents/memory/evaluation-registry/README.md`:
-
 - Add the metric with status `Active`.
 - Document location, thresholds, and trust level.
 
 ### 5. Integrate
 
-Update `.agents/skills/evaluate-video-quality/SKILL.md`:
-
+Update `.agents/skills/evaluate-video-quality.md`:
 - Add the new metric as a section.
 - Include code examples and interpretation guide.
 
@@ -59,35 +52,3 @@ Update `.agents/skills/evaluate-video-quality/SKILL.md`:
 - Move the exploration log content into the skill.
 - Clean up the exploration file or mark it as `promoted`.
 - If anything went wrong during development, create a lesson.
-
-## Where the metrics live
-
-The eval suite is `fastvideo/eval/`. New metrics register themselves
-via `@register("<group>.<name>")` and are auto-discovered when
-`fastvideo.eval.metrics` is imported.
-
-- **Native metrics** (SSIM, PSNR, LPIPS, optical flow, VLM): add a
-  file under the appropriate group dir
-  (`fastvideo/eval/metrics/common/`, `optical_flow/`, `videoscore2/`,
-  `physics_iq/`).
-- **Metrics that wrap upstream research code**: follow the vbench
-  pattern in `fastvideo/eval/metrics/vbench/`. The contract is:
-  - Upstream lives as a git submodule under
-    `fastvideo/third_party/eval/<bench>/`, pinned to a SHA in repo-root
-    `.gitmodules`.
-  - The metric package's `__init__.py` inserts the submodule path on
-    `sys.path` and installs runtime compat shims (attribute-level
-    monkey-patches) for any modern-dep drift. Do not modify upstream
-    files on disk, and do not ship a `setup.sh`.
-  - See `fastvideo/eval/README.md` for the worked vbench example.
-  - Full porting guide:
-    [`docs/contributing/eval-metrics.md`](../../docs/contributing/eval-metrics.md).
-
-## Out of scope of the initial eval port
-
-The following land in follow-up PRs:
-
-- **MIND** metrics (depends on a separate `vipe` submodule).
-- **VBench-2.0** sibling package.
-- Native conversion of **FVD** under `fastvideo/eval/metrics/fvd/`.
-- The training-time `EvalCallback`.

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

@@ -29,18 +29,24 @@
     "Will Smith casually eats noodles, his relaxed demeanor contrasting with the energetic background of a bustling street food market. The scene captures a mix of humor and authenticity. Mid-shot framing, vibrant lighting."
   ],
   "run_config": {
-    "num_warmup_runs": 1,
-    "num_measurement_runs": 3,
+    "num_warmup_runs": 2,
+    "num_measurement_runs": 5,
     "required_gpus": 2
   },
   "thresholds": {
     "L40S": {
       "max_generation_time_s": 34.0,
-      "max_peak_memory_mb": 11000.0
+      "max_peak_memory_mb": 11000.0,
+      "max_text_encoder_time_s": 5.0,
+      "max_dit_time_s": 10.0,
+      "max_vae_decode_time_s": 10.0
     },
     "default": {
       "max_generation_time_s": 120.0,
-      "max_peak_memory_mb": 30000.0
+      "max_peak_memory_mb": 30000.0,
+      "max_text_encoder_time_s": 5.0,
+      "max_dit_time_s": 10.0,
+      "max_vae_decode_time_s": 10.0
     }
   }
 }

.buildkite/scripts/pr_test.sh

@@ -15,8 +15,21 @@ log "Project root: $PROJECT_ROOT"
 # Install Modal if not available
 if ! python3 -m modal --version &> /dev/null; then
     log "Modal not found, installing..."
-    python3 -m pip install modal
-    
+    if ! command -v uv &> /dev/null; then
+        log "uv not found, bootstrapping..."
+        if ! curl -LsSf https://astral.sh/uv/install.sh | sh; then
+            log "Error: Failed to bootstrap uv via astral.sh installer."
+            exit 1
+        fi
+        export PATH="$HOME/.local/bin:$PATH"
+        if ! command -v uv &> /dev/null; then
+            log "Error: uv still not on PATH after bootstrap."
+            exit 1
+        fi
+    fi
+    # --break-system-packages preserves prior `pip install --user` semantics on PEP 668 agents.
+    uv pip install --system --break-system-packages modal
+
     # Verify installation
     if ! python3 -m modal --version &> /dev/null; then
         log "Error: Failed to install modal. Please install it manually."
@@ -63,7 +76,86 @@ EFFECTIVE_PR=${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 IMAGE_VERSION=$IMAGE_VERSION"
+MODAL_ENV="BUILDKITE_REPO=$BUILDKITE_REPO BUILDKITE_COMMIT=$BUILDKITE_COMMIT BUILDKITE_PULL_REQUEST=$EFFECTIVE_PR BUILDKITE_BRANCH=${BUILDKITE_BRANCH:-} TEST_SCOPE=${TEST_SCOPE:-} IMAGE_VERSION=$IMAGE_VERSION"
+
+POST_RUN_HOOK=""
+
+upload_performance_artifacts() {
+    SHORT_SHA=${BUILDKITE_COMMIT:0:7}
+    LOCAL_DIR="downloaded_reports"
+
+    _download_reports() {
+        log "Downloading perf_reports/ from Modal Volume..."
+        mkdir -p "$LOCAL_DIR"
+        if ! modal volume get hf-model-weights "perf_reports/" "$LOCAL_DIR"; then
+            log "Error: Failed to download perf_reports/ from Modal Volume."
+            return 1
+        fi
+    }
+
+    _upload_dashboard() {
+        local target
+        target=$(find "$LOCAL_DIR" -name "dashboard_${SHORT_SHA}_*" | head -n 1)
+        log "TARGET dashboard: '$target'"
+
+        if [ -n "$target" ]; then
+            log "Found dashboard: $target. Uploading to Buildkite..."
+            buildkite-agent artifact upload "$target"
+            buildkite-agent annotate --style info --context "perf-dashboard" < "$target"
+        else
+            log "Warning: Could not find a dashboard file matching $SHORT_SHA"
+        fi
+    }
+
+    _upload_perf_summary() {
+        local target
+        target=$(find "$LOCAL_DIR" -name "perf_${SHORT_SHA}_*" | head -n 1)
+        log "TARGET perf summary: '$target'"
+
+        if [ -n "$target" ]; then
+            log "Found perf summary: $target. Uploading to Buildkite..."
+            buildkite-agent artifact upload "$target"
+            buildkite-agent annotate --style info --context "perf-summary" < "$target"
+        else
+            log "Warning: Could not find a perf summary file matching $SHORT_SHA"
+        fi
+    }
+
+    _upload_normalized_perf_results() {
+        local found=0
+        while IFS= read -r -d '' target; do
+            found=1
+            log "Found normalized performance result: $target. Uploading to Buildkite..."
+            buildkite-agent artifact upload "$target"
+        done < <(find "$LOCAL_DIR" -path "*/results/normalized_perf_*.json" -print0)
+
+        if [ "$found" -eq 0 ]; then
+            log "No normalized performance result artifacts found. This is expected when the rolling performance comparison did not run."
+        fi
+    }
+
+    _cleanup_modal_volume() {
+        log "Cleaning up perf_reports/ from Modal Volume..."
+        if modal volume rm hf-model-weights "perf_reports/" --recursive; then
+            log "Successfully deleted perf_reports/ from Modal Volume."
+        else
+            log "Warning: Failed to delete perf_reports/ from Modal Volume. Manual cleanup may be required."
+        fi
+    }
+
+    _cleanup_local() {
+        log "Cleaning up local download directory..."
+        rm -rf "$LOCAL_DIR"
+    }
+
+    # --- Main flow ---
+    _download_reports || { _cleanup_local; return 1; }
+    _upload_dashboard
+    _upload_perf_summary
+    _upload_normalized_perf_results
+    _cleanup_modal_volume
+    _cleanup_local
+}
 
 case "$TEST_TYPE" in
     "encoder")
@@ -124,8 +216,9 @@ case "$TEST_TYPE" in
         MODAL_COMMAND="$MODAL_ENV HF_API_KEY=$HF_API_KEY python3 -m modal run $MODAL_TEST_FILE::run_lora_extraction_tests"
         ;;
     "performance")
-        log "Running performance tests..."
+        log "Running performance tests on Modal..."
         MODAL_COMMAND="$MODAL_ENV HF_API_KEY=$HF_API_KEY python3 -m modal run $MODAL_TEST_FILE::run_performance_tests"
+        POST_RUN_HOOK="upload_performance_artifacts"
         ;;
     "api_server")
         log "Running API server integration tests..."
@@ -147,5 +240,10 @@ else
     log "Error: Modal test failed with exit code: $TEST_EXIT_CODE"
 fi
 
+if [ -n "$POST_RUN_HOOK" ]; then
+    log "Executing post-run hook: $POST_RUN_HOOK"
+    "$POST_RUN_HOOK"
+fi
+
 log "=== Test execution completed with exit code: $TEST_EXIT_CODE ==="
 exit $TEST_EXIT_CODE

.buildkite/scripts/pre_commit.sh

@@ -13,8 +13,21 @@ log "Project root: $PROJECT_ROOT"
 
 if ! python3 -m pre_commit --version &> /dev/null; then
     log "pre-commit not found, installing..."
-    python3 -m pip install --user pre-commit==4.0.1
-    
+    if ! command -v uv &> /dev/null; then
+        log "uv not found, bootstrapping..."
+        if ! curl -LsSf https://astral.sh/uv/install.sh | sh; then
+            log "Error: Failed to bootstrap uv via astral.sh installer."
+            exit 1
+        fi
+        export PATH="$HOME/.local/bin:$PATH"
+        if ! command -v uv &> /dev/null; then
+            log "Error: uv still not on PATH after bootstrap."
+            exit 1
+        fi
+    fi
+    # --break-system-packages preserves prior `pip install --user` semantics on PEP 668 agents.
+    uv pip install --system --break-system-packages pre-commit==4.0.1
+
     if ! python3 -m pre_commit --version &> /dev/null; then
         log "Error: Failed to install pre-commit."
         exit 1

.github/workflows/infra-docs.yml

@@ -37,10 +37,11 @@ jobs:
         with:
           python-version: '3.12'
 
+      - name: Install uv
+        uses: astral-sh/setup-uv@v3
+
       - name: Install dependencies
-        run: |
-          python -m pip install --upgrade pip
-          pip install -r requirements-mkdocs.txt
+        run: uv pip install --system -r requirements-mkdocs.txt
 
       - name: Setup Pages
         uses: actions/configure-pages@v4

.github/workflows/publish-fastvideo.yml

@@ -56,10 +56,11 @@ jobs:
         with:
           python-version: '3.10'
 
+      - name: Install uv
+        uses: astral-sh/setup-uv@v3
+
       - name: Install build dependencies
-        run: |
-          python -m pip install --upgrade pip
-          pip install build twine wheel
+        run: uv pip install --system build twine wheel
 
       - name: Build package
         run: |

.github/workflows/publish-kernel.yml

@@ -131,11 +131,13 @@ jobs:
           clang-11 --version
           nvcc --version
 
+      - name: Install uv
+        uses: astral-sh/setup-uv@v3
+
       - name: Install PyTorch ${{ matrix.torch-cuda.torch-version }}+cu${{ matrix.torch-cuda.cuda-version }}
         run: |
-          pip install --upgrade pip
-          pip install typing-extensions==4.12.2
-          pip install --no-cache-dir torch==${{ matrix.torch-cuda.torch-version }} --index-url https://download.pytorch.org/whl/${{matrix.torch-cuda.torch-cuda-short}}
+          uv pip install --system typing-extensions==4.12.2
+          uv pip install --system --no-cache-dir torch==${{ matrix.torch-cuda.torch-version }} --index-url https://download.pytorch.org/whl/${{matrix.torch-cuda.torch-cuda-short}}
           nvcc --version
           python --version
           python -c "import torch; print('PyTorch:', torch.__version__)"
@@ -145,20 +147,20 @@ jobs:
       - name: Build wheel
         run: |
           export PYTHONPATH=$GITHUB_WORKSPACE:$PYTHONPATH
-          
-          pip install setuptools ninja packaging wheel triton scikit-build-core cmake build
-          
+
+          uv pip install --system setuptools ninja packaging wheel triton scikit-build-core cmake build
+
           cd fastvideo-kernel
           git submodule update --init --recursive  # Ensure ThunderKittens submodule is initialized
           # Release builds are produced on GPU-less runners, so force-enable TK and target Hopper.
           export TORCH_CUDA_ARCH_LIST="9.0a"
           export CMAKE_ARGS="${CMAKE_ARGS:-} -DFASTVIDEO_KERNEL_BUILD_TK=ON -DCMAKE_CUDA_ARCHITECTURES=90a"
-          
+
           # Build standard wheel (no local version suffix) for PyPI
           python -m build --wheel --outdir dist
-          
+
           # Fix the wheel to be manylinux compliant
-          pip install auditwheel
+          uv pip install --system auditwheel
           # Point auditwheel at torch libs, but do not vendor them into the wheel.
           TORCH_LIB_DIR=$(python - <<'PY'
           import os
@@ -211,10 +213,13 @@ jobs:
           pattern: 'fastvideo_kernel-py*'
           merge-multiple: true
 
+      - name: Install uv
+        uses: astral-sh/setup-uv@v3
+
       - name: Build source distribution
         run: |
-          pip install build scikit-build-core cmake ninja
-          
+          uv pip install --system build scikit-build-core cmake ninja
+
           cd fastvideo-kernel
           # We don't need full CUDA/Torch to just package the source (sdist)
           python -m build --sdist --outdir dist

.gitmodules

@@ -4,6 +4,3 @@
 [submodule "fastvideo-kernel/include/cutlass"]
 	path = fastvideo-kernel/include/cutlass
 	url = https://github.com/NVIDIA/cutlass.git
-[submodule "fastvideo/third_party/eval/vbench"]
-	path = fastvideo/third_party/eval/vbench
-	url = https://github.com/Vchitect/VBench.git

.pre-commit-config.yaml

@@ -7,20 +7,13 @@ exclude: |
     fastvideo-kernel/.*|
     assets/.*|
     tests/.*|
-    demo/.*|
-    predict\.py|
     scripts/.*|
-    assets/prompts/.*|
-    fastvideo/data_preprocess/.*|
     fastvideo/dataset/.*|
     fastvideo/models/.*|
-    fastvideo/sample/.*|
-    fastvideo/train\.py|
-    fastvideo/utils/.*|
     examples/.*|
+    \.agents/.*|
     .github/workflows/publish-fastvideo.yml|
-    .github/workflows/_template-build-image.yml|
-    docs/source/inference/support_matrix.md
+    .github/workflows/_template-build-image.yml
   )
 repos:
 - repo: https://github.com/google/yapf

AGENTS.md

@@ -11,7 +11,7 @@
 - Static assets: `assets/` (including `assets/images/`, `assets/videos/`, and `assets/prompts/`) and `comfyui/assets/`.
 
 ## Build, Test, and Development Commands
-- `uv pip install -e .[dev]`: editable install with lint/test extras.
+- `uv pip install -e ".[dev]"`: editable install with lint/test extras.
 - `pre-commit install --hook-type pre-commit --hook-type commit-msg`: enable local hooks.
 - `pre-commit run --all-files`: run formatter/lint/type/spelling checks.
 - `pytest tests/`: run top-level test suite.
@@ -23,7 +23,8 @@
 - Python 3.10+; 4-space indentation; keep code and imports readable and explicit.
 - Style tools are configured in `pyproject.toml` and `.pre-commit-config.yaml`:
   - `yapf` (format), `ruff` (lint, auto-fix), `mypy` (typing), `codespell`.
-- Target line length is 80.
+- Lint via `pre-commit run --files <changed paths>` (or `pre-commit run --all-files` for a full sweep) before committing. Do not shell out to `yapf`/`ruff`/`codespell`/`mypy` directly — pre-commit chains them with the project's config and respects the `.pre-commit-config.yaml` excludes (e.g. `fastvideo/tests/` is intentionally skipped). If pre-commit reports `(no files to check)` for your paths, that exclude is deliberate — don't bypass it.
+- Target line length is 120 (configured in `pyproject.toml` for ruff, yapf, and isort).
 - Naming: `snake_case` for functions/files, `PascalCase` for classes, `UPPER_SNAKE_CASE` for constants.
 
 ## Testing Guidelines
@@ -54,3 +55,31 @@ This repository is agent-friendly. Before doing any work, read:
 If you are exploring a new procedure that has no existing SOP, document your
 progress in `.agents/exploration/` and flag it for review at the end of your
 session.
+
+## Per-Directory AGENTS.md
+
+Local guidance lives next to the code. Read the in-scope file before editing:
+
+| Directory | What it covers |
+|-----------|----------------|
+| `fastvideo/AGENTS.md` | Core package map, public API, registry-driven model dispatch |
+| `fastvideo/configs/AGENTS.md` | Arch + pipeline config dataclasses, `param_names_mapping` |
+| `fastvideo/models/AGENTS.md` | DiT / VAE / encoder / scheduler / loader layout (pre-commit excluded) |
+| `fastvideo/layers/AGENTS.md` | Tensor-parallel linear/attention layer rules for ports |
+| `fastvideo/attention/AGENTS.md` | Backend registry + env-var override |
+| `fastvideo/pipelines/AGENTS.md` | Stage ABC, `basic/<model>/`, `preprocess/`, presets |
+| `fastvideo/training/AGENTS.md` | Legacy monolithic pipelines (frozen for existing models) |
+| `fastvideo/train/AGENTS.md` | New modular trainer (methods × models × callbacks, YAML) |
+| `fastvideo/tests/AGENTS.md` | Test taxonomy, conftest, pre-commit-excluded path |
+| `fastvideo/tests/ssim/AGENTS.md` | GPU SSIM regression authoring + reference video sync |
+| `scripts/checkpoint_conversion/AGENTS.md` | Adding a converter for a new HF/official checkpoint |
+
+## Critical: Two Training Stacks Coexist
+
+- `fastvideo/training/` — legacy, monolithic per-model `*_training_pipeline.py` and
+  `*_distillation_pipeline.py`. Still authoritative for shipped models.
+- `fastvideo/train/` — new modular framework (composable methods × models × callbacks
+  driven by YAML). Preferred for new training work.
+
+Pick the matching stack before editing. Do not migrate a pipeline between them
+without an explicit ask — the conventions and config surfaces differ.

benchmarks/fvd/feature_extractors.py

@@ -128,7 +128,7 @@ class CLIPFeatureExtractor(BaseFeatureExtractor):
 
     def __init__(self, device: str = 'cuda', model_name: str = "openai/clip-vit-base-patch32"):
         if not TRANSFORMERS_AVAILABLE:
-            raise ImportError("Please install transformers: pip install transformers")
+            raise ImportError("Please install transformers: uv pip install transformers")
         super().__init__(device)
         self.processor = CLIPProcessor.from_pretrained(model_name)
         self.model = CLIPModel.from_pretrained(model_name).to(self.device)
@@ -171,7 +171,7 @@ class VideoMAEFeatureExtractor(BaseFeatureExtractor):
 
     def __init__(self, device: str = 'cuda', model_name: str = "MCG-NJU/videomae-base"):
         if not TRANSFORMERS_AVAILABLE:
-            raise ImportError("Please install transformers: pip install transformers")
+            raise ImportError("Please install transformers: uv pip install transformers")
         super().__init__(device)
         self.model = VideoMAEModel.from_pretrained(model_name).to(self.device)
         self.model.eval()

benchmarks/fvd/i3d_model.py

@@ -57,7 +57,7 @@ class I3DFeatureExtractor(nn.Module):
         except Exception as e:
             raise RuntimeError(f"Failed to load I3D model from Hugging Face Hub. Error: {e}\n"
                                f"Ensure you have internet connection and huggingface_hub installed:\n"
-                               f"pip install huggingface_hub") from e
+                               f"uv pip install huggingface_hub") from e
 
     def preprocess(self, videos: torch.Tensor) -> torch.Tensor:
         """

benchmarks/scripts/run.sh

@@ -1,7 +1,7 @@
 #!/bin/bash
 
 # 1. Install missing dependency
-pip install -q opencv-python-headless transformers huggingface_hub
+uv pip install -q opencv-python-headless transformers huggingface_hub
 
 # 2. Run FVD script
 python benchmarks/fvd/run_fvd.py

benchmarks/scripts/setup_fvd.sh

@@ -1,4 +1,4 @@
 #!/bin/bash
 
 # 1. Install missing dependency
-pip install -q opencv-python-headless
+uv pip install -q opencv-python-headless

comfyui/README.md

@@ -38,10 +38,10 @@ cp -r /path/to/FastVideo/comfyui /path/to/ComfyUI/custom_nodes/FastVideo
 
 #### Install dependencies:
 
-Currently, the only dependency is `fastvideo`, which can be installed using pip.
+Currently, the only dependency is `fastvideo`, which can be installed with `uv`.
 
 ```bash
-pip install fastvideo
+uv pip install fastvideo
 ```
 
 #### Install missing custom nodes:

docker/Dockerfile.python3.10

@@ -42,15 +42,15 @@ RUN source $HOME/.local/bin/env && \
     uv venv --python 3.10 --seed /opt/venv && \
     source /opt/venv/bin/activate && \
     uv pip install --no-cache-dir --upgrade pip && \
-    uv pip install --no-cache-dir .[dev] && \
-    uv pip install --no-cache-dir https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.7.16/flash_attn-2.8.3+cu128torch2.10-cp310-cp310-linux_x86_64.whl
+    uv pip install --no-cache-dir ".[dev]" && \
+    uv pip install --no-cache-dir https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.9.4/flash_attn-2.8.3+cu128torch2.11-cp310-cp310-linux_x86_64.whl
 
 COPY . .
 
 # Install dependencies using uv and set up shell configuration
 RUN source $HOME/.local/bin/env && \
     source /opt/venv/bin/activate && \
-    uv pip install --no-cache-dir -e .[dev] && \
+    uv pip install --no-cache-dir -e ".[dev]" && \
     git config --unset-all http.https://github.com/.extraheader || true && \
     echo 'source /opt/venv/bin/activate' >> /root/.bashrc && \
     echo 'if [ -n "$ZSH_VERSION" ] && [ -f ~/.zshrc ]; then . ~/.zshrc; elif [ -f ~/.bashrc ]; then . ~/.bashrc; fi' > /root/.profile

docker/Dockerfile.python3.11

@@ -42,15 +42,15 @@ RUN source $HOME/.local/bin/env && \
     uv venv --python 3.11 --seed /opt/venv && \
     source /opt/venv/bin/activate && \
     uv pip install --no-cache-dir --upgrade pip && \
-    uv pip install --no-cache-dir .[dev] && \
-    uv pip install --no-cache-dir https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.7.16/flash_attn-2.8.3+cu128torch2.10-cp311-cp311-linux_x86_64.whl
+    uv pip install --no-cache-dir ".[dev]" && \
+    uv pip install --no-cache-dir https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.9.4/flash_attn-2.8.3+cu128torch2.11-cp311-cp311-linux_x86_64.whl
 
 COPY . .
 
 # Install dependencies using uv and set up shell configuration
 RUN source $HOME/.local/bin/env && \
     source /opt/venv/bin/activate && \
-    uv pip install --no-cache-dir -e .[dev] && \
+    uv pip install --no-cache-dir -e ".[dev]" && \
     git config --unset-all http.https://github.com/.extraheader || true && \
     echo 'source /opt/venv/bin/activate' >> /root/.bashrc && \
     echo 'if [ -n "$ZSH_VERSION" ] && [ -f ~/.zshrc ]; then . ~/.zshrc; elif [ -f ~/.bashrc ]; then . ~/.bashrc; fi' > /root/.profile

docker/Dockerfile.python3.12

@@ -42,15 +42,15 @@ RUN source $HOME/.local/bin/env && \
     uv venv --python 3.12 --seed /opt/venv && \
     source /opt/venv/bin/activate && \
     uv pip install --no-cache-dir --upgrade pip && \
-    uv pip install --no-cache-dir .[dev] && \
-    uv pip install --no-cache-dir https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.7.16/flash_attn-2.8.3+cu128torch2.10-cp312-cp312-linux_x86_64.whl
+    uv pip install --no-cache-dir ".[dev]" && \
+    uv pip install --no-cache-dir https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.9.4/flash_attn-2.8.3+cu128torch2.11-cp312-cp312-linux_x86_64.whl
 
 COPY . .
 
 # Install dependencies using uv and set up shell configuration
 RUN source $HOME/.local/bin/env && \
     source /opt/venv/bin/activate && \
-    uv pip install --no-cache-dir -e .[dev] && \
+    uv pip install --no-cache-dir -e ".[dev]" && \
     git config --unset-all http.https://github.com/.extraheader || true && \
     echo 'source /opt/venv/bin/activate' >> /root/.bashrc && \
     echo 'if [ -n "$ZSH_VERSION" ] && [ -f ~/.zshrc ]; then . ~/.zshrc; elif [ -f ~/.bashrc ]; then . ~/.bashrc; fi' > /root/.profile

docker/Dockerfile.python3.12.cuda12.9.1

@@ -42,7 +42,7 @@ RUN source $HOME/.local/bin/env && \
     uv venv --python 3.12 --seed /opt/venv && \
     source /opt/venv/bin/activate && \
     uv pip install --no-cache-dir --upgrade pip && \
-    uv pip install --no-cache-dir .[dev] && \
+    uv pip install --no-cache-dir ".[dev]" && \
     uv pip install --no-cache-dir flash-attn==2.8.3 --no-build-isolation
 
 COPY . .
@@ -50,7 +50,7 @@ COPY . .
 # Install dependencies using uv and set up shell configuration
 RUN source $HOME/.local/bin/env && \
     source /opt/venv/bin/activate && \
-    uv pip install --no-cache-dir -e .[dev] && \
+    uv pip install --no-cache-dir -e ".[dev]" && \
     git config --unset-all http.https://github.com/.extraheader || true && \
     echo 'source /opt/venv/bin/activate' >> /root/.bashrc && \
     echo 'if [ -n "$ZSH_VERSION" ] && [ -f ~/.zshrc ]; then . ~/.zshrc; elif [ -f ~/.bashrc ]; then . ~/.bashrc; fi' > /root/.profile

docker/rocm7_1.Dockerfile

@@ -43,7 +43,7 @@ COPY . .
 # Install dependencies using uv and set up shell configuration
 RUN source $HOME/.local/bin/env && \
     source /opt/venv/bin/activate && \
-    uv pip install --no-cache-dir -e .[rocm] && \
+    uv pip install --no-cache-dir -e ".[rocm]" && \
     git config --unset-all http.https://github.com/.extraheader || true && \
     echo 'source /opt/venv/bin/activate' >> /root/.bashrc && \
     echo 'if [ -n "$ZSH_VERSION" ] && [ -f ~/.zshrc ]; then . ~/.zshrc; elif [ -f ~/.bashrc ]; then . ~/.bashrc; fi' > /root/.profile

docs/README.md

@@ -6,7 +6,7 @@ This directory contains the FastVideo documentation built with MkDocs.
 
 ```bash
 # Install dependencies
-pip install -r requirements-mkdocs.txt
+uv pip install -r requirements-mkdocs.txt
 
 # Serve docs with live reload (recommended for development)
 mkdocs serve

docs/contributing/coding_agents.md

@@ -296,8 +296,10 @@ Action:
 
 - Add or reuse a numerical parity test that loads the official model and the
   FastVideo model and compares outputs.
-- See examples in `tests/local_tests/` (e.g., `tests/local_tests/upsamplers/`)
-  and the commands in `tests/local_tests/README.md`.
+- See examples in `tests/local_tests/` organized by model family
+  (e.g., `tests/local_tests/sd35/`, `tests/local_tests/ltx2/`,
+  `tests/local_tests/stable_audio/`) and the navigation index in
+  `tests/local_tests/README.md`.
 - If there are discrepancies, add opt‑in logging to both models and compare
   activation summaries (layer output sums, per‑stage logs).
 - First align the loaded weights (validate `param_names_mapping`).
@@ -348,7 +350,8 @@ Purpose:
 
 Action:
 
-- Add a pipeline parity test under `tests/local_tests/pipelines/`.
+- Add a pipeline parity test under `tests/local_tests/<family>/`
+  (e.g., `tests/local_tests/<family>/test_<family>_pipeline_parity.py`).
 - See the [Testing Guide](testing.md) for test conventions.
 
 ### 7) Add user‑facing examples

docs/contributing/developer_env/runpod.md

@@ -99,7 +99,7 @@ cd /FastVideo
 **Install the package**
 
 ```bash
-uv pip install -e .[dev]
+uv pip install -e ".[dev]"
 ```
 
 The Docker image already includes Flash Attention and most heavy dependencies, so this is fast.

docs/contributing/eval-metrics.md

@@ -1,559 +0,0 @@
-# Porting Eval Metrics into `fastvideo.eval`
-
-This guide is for contributors adding new evaluation metrics to
-FastVideo's eval suite. To run the existing metrics, see
-[`fastvideo/eval/README.md`](../../fastvideo/eval/README.md).
-
-## When to use this guide
-
-Use this guide when you are:
-
-- Adding a new metric (native or wrapping a third-party library).
-- Porting a benchmark (e.g. VBench, MIND, EvalCrafter) whose Python
-  code needs to be importable from a pinned upstream.
-- Adding a new metric group (audio, vlm, etc.).
-
-## TL;DR
-
-Metrics are auto-discovered from
-`fastvideo/eval/metrics/<group>/<name>/metric.py`. Each declares itself
-with `@register("<group>.<name>")` and subclasses `BaseMetric`. Three
-recipes:
-
-1. **Native metric** (pure-PyTorch, no submodule). Drop a file,
-   declare deps, implement `compute(sample)`.
-2. **Library-wrapped metric** (CLIP, torch.hub, transformers, pyiqa).
-   Same as above, plus route the library's cache through
-   `get_cache_dir()` if it has a `download_root=` / `cache_dir=`
-   kwarg.
-3. **Upstream-submodule-wrapped metric** (vbench-style). Pin upstream
-   as a git submodule under `fastvideo/third_party/eval/<bench>/`. The
-   adapter `__init__.py` does the `sys.path` insert and any runtime
-   compat shims for modern dep versions. Patches live as Python in
-   that file rather than as on-disk patches to the submodule.
-
-The full recipes are below.
-
----
-
-## 0) Layout and auto-discovery
-
-```
-fastvideo/eval/metrics/
-├── base.py                  # BaseMetric + lifecycle contract
-├── common/                  # group: SSIM, PSNR, LPIPS
-├── optical_flow/            # group: gt_optical_flow, synthetic_optical_flow
-├── vlm/                     # group: VideoScore-2
-├── physics_iq/              # group + sub-metrics
-└── vbench/                  # group: 16 sub-metrics
-    ├── __init__.py          # sys.path bootstrap + runtime compat shims
-    ├── _grit_helper.py      # shared upstream-touching helpers
-    └── <sub_metric>/metric.py
-```
-
-Auto-discovery (`fastvideo/eval/metrics/__init__.py`) walks each group
-dir and imports every `metric.py` it finds, which fires the
-`@register` decorators. Names starting with `_` are skipped. Use that
-prefix for shared helpers or vendored code that should not register
-itself.
-
----
-
-## 1) The `BaseMetric` contract
-
-Every metric subclasses `fastvideo.eval.metrics.base.BaseMetric` and
-declares:
-
-```python
-class YourMetric(BaseMetric):
-    name: str = "common.your_metric"          # must match @register
-    requires_reference: bool = True           # needs sample["reference"]
-    higher_is_better: bool = True             # for ranking / aggregates
-    dependencies: list[str] = []              # importable module names;
-                                              # registry surfaces a clean
-                                              # ImportError if missing
-    needs_gpu: bool = False
-    backbone: str | None = None               # e.g. "clip_vit_l14"
-```
-
-You must implement:
-
-```python
-def compute(self, sample: dict) -> list[MetricResult]:
-    """sample['video'] is (1, T, C, H, W). Return a one-element list.
-
-    The leading 1 is preserved for forward-compat with batched eval;
-    today :class:`EvalWorker` always invokes metrics with B=1.
-    """
-```
-
-You may override:
-
-- `setup(self) -> None`. Eager model loading. Called once by
-  `create_evaluator`. Idempotent (re-entrant). Use the `if self._model
-  is not None: return` pattern.
-- `to(self, device)`. Move the metric and its submodels to `device`.
-
-If a required input is missing (e.g. an fps-aware metric called
-without `fps`), return `self._skip(sample, reason)` instead of
-raising.
-
----
-
-## 2) Recipe A: native metric (no external deps)
-
-Smallest case. Pixel math, simple closed-form.
-
-```python
-# fastvideo/eval/metrics/common/your_metric/metric.py
-from __future__ import annotations
-import torch
-from fastvideo.eval.metrics.base import BaseMetric
-from fastvideo.eval.registry import register
-from fastvideo.eval.types import MetricResult
-
-
-@register("common.your_metric")
-class YourMetric(BaseMetric):
-    name = "common.your_metric"
-    requires_reference = True
-    higher_is_better = True
-    needs_gpu = False
-    dependencies: list[str] = []  # nothing extra
-
-    def compute(self, sample: dict) -> list[MetricResult]:
-        gen, ref = sample["video"], sample["reference"]   # (B,T,C,H,W) each
-        per_video = ((gen - ref) ** 2).mean(dim=(1, 2, 3, 4)).sqrt()
-        return [
-            MetricResult(name=self.name, score=float(s), details={})
-            for s in per_video
-        ]
-```
-
-That is the whole recipe. Drop the file and the registry picks it up.
-
----
-
-## 3) Recipe B: library-wrapped metric (CLIP, torch.hub, transformers, pyiqa)
-
-If your metric loads a backbone from a Python package, route the
-library at the eval cache so users get one knob (`FASTVIDEO_EVAL_CACHE`)
-to redirect everything.
-
-### Cache routing rules
-
-| Library | How to route | Location after redirect |
-|---|---|---|
-| `clip.load("ViT-X")` | pass `download_root=str(get_cache_dir() / "clip")` | `${FASTVIDEO_EVAL_CACHE}/clip/` |
-| `torch.hub.load(...)` | nothing; `TORCH_HOME` is redirected at `fastvideo.eval` import time | `${FASTVIDEO_EVAL_CACHE}/torch/hub/` |
-| `transformers.from_pretrained(...)` | nothing; leave HF's default cache (`~/.cache/huggingface/hub/`) so users dedupe with other ML projects | `~/.cache/huggingface/hub/` |
-| `huggingface_hub.snapshot_download` / `hf_hub_download` | use `ensure_checkpoint(...)` (it wraps these with filelock) | same as above |
-| `pyiqa.create_metric(...)` | no env var or kwarg honored; document in metric docstring | pyiqa-internal |
-| `lpips`, `ptlflow` | torch.hub-based, auto-redirected | `${FASTVIDEO_EVAL_CACHE}/torch/hub/` |
-| Raw URL (no HF Hub) | use `ensure_checkpoint(name, source="https://...")` | `${FASTVIDEO_EVAL_CACHE}/models/<name>` |
-| Dataset asset (raw video/mask/image) auto-fetched from a public bucket | download into `get_cache_dir() / "datasets" / "<bench>"`, mirroring upstream's relative layout. Vendor any small manifest (CSV/JSON ≤1 MB) under the metric folder so the dataset can be used without external setup. | `${FASTVIDEO_EVAL_CACHE}/datasets/<bench>/` |
-
-### Dataset assets: vendor the manifest, auto-fetch the rest
-
-If your metric ships with its own paired-reference dataset (Physics-IQ
-is the canonical example), follow this layout:
-
-- **Manifest** (CSV/JSON ≤1 MB): vendor it under
-  `fastvideo/eval/metrics/<bench>/_vendored/<manifest>.<ext>`, with a
-  sibling `_vendored/LICENSE` recording attribution and provenance.
-  The `_vendored/` subdir is the project-wide convention for
-  upstream-provenance files: it is auto-skipped by metric discovery
-  (the `_` prefix) and by codespell (one `*/_vendored/*` glob in
-  `[tool.codespell].skip`), so dropping in a new vendored file
-  requires no further config. Read the manifest from the dataset
-  module via a `Path(__file__)`-relative resolver. Mirror
-  `_VENDORED_DESCRIPTIONS_CSV` in
-  `fastvideo/eval/datasets/physics_iq.py`.
-- **Heavy assets** (videos, masks, images): do not vendor. Auto-fetch
-  on first miss into `get_cache_dir() / "datasets" / "<bench>"`,
-  mirroring upstream's relative directory layout one-for-one so a
-  pre-downloaded mirror at any path works as a drop-in
-  `dataset_root=`. Use atomic `.part` then final-rename to be safe
-  under concurrent SLURM ranks.
-- **Bucket override**: expose `FASTVIDEO_<BENCH>_BUCKET_URL` so users
-  with internal mirrors can redirect.
-- **Opt-out**: accept `auto_download: bool = True` in the dataset
-  constructor; on `False`, raise `FileNotFoundError` instead of
-  fetching. This covers air-gapped runs and CI.
-
-The end-state is `get_dataset("<bench>")` with no kwargs.
-
-### Example: CLIP backbone + LAION head
-
-```python
-# fastvideo/eval/metrics/your_group/your_metric/metric.py
-from __future__ import annotations
-import torch
-import torch.nn as nn
-from fastvideo.eval.metrics.base import BaseMetric
-from fastvideo.eval.registry import register
-from fastvideo.eval.types import MetricResult
-
-
-@register("your_group.your_metric")
-class YourMetric(BaseMetric):
-    name = "your_group.your_metric"
-    requires_reference = False
-    needs_gpu = True
-    dependencies = ["clip"]      # "openai-clip" PyPI; importable as `clip`
-
-    def __init__(self) -> None:
-        super().__init__()
-        self._clip = None
-        self._head = None
-
-    def setup(self) -> None:
-        if self._clip is not None:
-            return
-        import clip
-        from fastvideo.eval.models import ensure_checkpoint, get_cache_dir
-
-        # Backbone: route CLIP's cache through our root.
-        self._clip, _ = clip.load(
-            "ViT-L/14",
-            device=self.device,
-            download_root=str(get_cache_dir() / "clip"),
-        )
-        self._clip.eval()
-
-        # URL-fetched head: ensure_checkpoint downloads to
-        # ${FASTVIDEO_EVAL_CACHE}/models/ with filelock + atomic rename.
-        ckpt = ensure_checkpoint(
-            "your_head.pth",
-            source="https://example.com/path/to/your_head.pth",
-        )
-        self._head = nn.Linear(768, 1)
-        self._head.load_state_dict(
-            torch.load(ckpt, map_location="cpu", weights_only=True)
-        )
-        self._head.to(self.device).eval()
-
-    def to(self, device):
-        super().to(device)
-        if self._clip is not None:
-            self._clip = self._clip.to(self.device)
-        if self._head is not None:
-            self._head = self._head.to(self.device)
-        return self
-
-    def compute(self, sample: dict) -> list[MetricResult]:
-        ...
-```
-
-### Do not redirect other `~/.cache/...` dirs
-
-If a third-party library hard-codes `~/.cache/<lib>/` and offers no
-override, document the exception in the metric's docstring. Forcing
-redirection by setting `os.environ` or patching `os.path.expanduser`
-is fragile and breaks user expectations of where the library's cache
-lives.
-
----
-
-## 4) Recipe C: upstream-submodule-wrapped metric (vbench pattern)
-
-Use this when the upstream benchmark ships Python code (`vbench/`,
-`MIND/`, etc.) that is not pip-installable cleanly. See
-`fastvideo/eval/metrics/vbench/__init__.py` for the worked example.
-
-### 4.1 Pin the upstream as a submodule
-
-```bash
-git submodule add <upstream-url> fastvideo/third_party/eval/<bench>
-cd fastvideo/third_party/eval/<bench>
-git checkout <pinned-sha>
-cd -
-git add .gitmodules fastvideo/third_party/eval/<bench>
-```
-
-The submodule pulls under the standard `git submodule update --init
---recursive` flow that users already run for kernel deps.
-
-### 4.2 Bootstrap on `sys.path`
-
-```python
-# fastvideo/eval/metrics/<bench>/__init__.py
-from __future__ import annotations
-import sys
-from pathlib import Path
-
-# fastvideo/eval/metrics/<bench>/__init__.py → ../../../../third_party/eval/<bench>
-_UPSTREAM = Path(__file__).resolve().parents[3] / "third_party" / "eval" / "<bench>"
-if _UPSTREAM.is_dir() and str(_UPSTREAM) not in sys.path:
-    sys.path.insert(0, str(_UPSTREAM))
-```
-
-We do not `pip install` the upstream because its egg-link/.pth would
-just re-do this `sys.path.insert`, and skipping the install also skips
-the upstream's `setup.py` (which often gates on a specific CUDA
-version).
-
-### 4.3 Modern-dep compat: runtime shims
-
-Upstream code pinned to e.g. `transformers==4.33.2`, `numpy<2`
-typically breaks against modern versions in 3-4 known places (API
-renames). Fix those at import time, in the same `__init__.py`:
-
-```python
-def _install_compat_shims() -> None:
-    # Example: transformers.modeling_utils API moved.
-    try:
-        import transformers.modeling_utils as _mu
-        import transformers.pytorch_utils as _pu
-        for _n in ("apply_chunking_to_forward",
-                   "find_pruneable_heads_and_indices",
-                   "prune_linear_layer"):
-            if not hasattr(_mu, _n) and hasattr(_pu, _n):
-                setattr(_mu, _n, getattr(_pu, _n))
-    except ImportError:
-        pass
-
-    # Example: numpy.lib.function_base.disp removed in numpy>=2.
-    try:
-        import types, numpy.lib as _nl
-        if not hasattr(_nl, "function_base"):
-            _stub = types.ModuleType("numpy.lib.function_base")
-            _stub.disp = lambda *a, **k: None
-            sys.modules["numpy.lib.function_base"] = _stub
-            _nl.function_base = _stub
-    except ImportError:
-        pass
-
-_install_compat_shims()
-```
-
-For function-level patches that cannot be expressed as attribute
-writes (e.g. wrapping a model factory function), use a
-`sys.meta_path` finder that wraps the loader. See
-`_install_modeling_finetune_hook()` in
-`fastvideo/eval/metrics/vbench/__init__.py` for the pattern (about 30
-lines).
-
-Why shims rather than `git apply` patches: patches go stale when the
-upstream SHA changes; shims are versioned Python code in our repo,
-they are grep-able, and they only run if the targeted module is
-imported.
-
-### 4.4 Per-sub-metric files
-
-Each sub-metric is a normal `BaseMetric` subclass that imports from
-the upstream:
-
-```python
-# fastvideo/eval/metrics/<bench>/<sub>/metric.py
-from fastvideo.eval.metrics.base import BaseMetric
-from fastvideo.eval.registry import register
-from fastvideo.eval.types import MetricResult
-
-
-@register("<bench>.<sub>")
-class YourSubMetric(BaseMetric):
-    ...
-
-    def setup(self) -> None:
-        if self._model is not None:
-            return
-        # The sys.path bootstrap fired when fastvideo.eval.metrics.<bench>
-        # was imported (which auto-discovery does before importing this
-        # sub-package). Upstream imports just work:
-        from <bench>.something import SomeModel
-        ...
-```
-
-### 4.5 Conditional registration when the upstream is missing
-
-If a user installed `fastvideo[eval]` but did not run `git submodule
-update --init`, `<bench>.*` metrics should not register. The
-auto-discovery walker imports each sub-package's `metric` module; have
-that import bail out cleanly:
-
-```python
-# fastvideo/eval/metrics/<bench>/__init__.py — at the bottom
-_AVAILABLE = (_UPSTREAM / "<bench>" / "__init__.py").is_file()
-```
-
-```python
-# fastvideo/eval/metrics/<bench>/<sub>/__init__.py
-from fastvideo.eval.metrics.<bench> import _AVAILABLE
-if _AVAILABLE:
-    from .metric import YourSubMetric  # noqa
-```
-
-`fastvideo eval list` then reflects what the user actually has rather
-than what they could have.
-
-### 4.6 Leave the upstream alone unless it blocks the metric
-
-The upstream is pinned. If a metric works against the pinned SHA,
-leave the upstream files untouched. If it actively breaks against
-modern deps (the import-drift cases above), shim it. Avoid
-fastvideo-side forks of upstream code; they make patches go stale and
-parity drift.
-
----
-
-## 5) Model checkpoints: `ensure_checkpoint`
-
-Use `ensure_checkpoint(name, source, filename=None)` for any
-non-package weights. It resolves a local path, downloading on miss,
-with filelock safety across processes and SLURM ranks.
-
-| `source` form | What happens |
-|---|---|
-| `"/abs/path/to/file.pth"` | passthrough, returned unchanged |
-| `"https://..."` | downloaded to `${FASTVIDEO_EVAL_CACHE}/models/<name>` via `huggingface_hub.http_get`, atomic rename, filelock |
-| `"org/repo"` (no `filename`) | `snapshot_download(repo_id)` → `~/.cache/huggingface/hub/` |
-| `"org/repo"` (with `filename`) | `hf_hub_download(repo_id, filename)` → `~/.cache/huggingface/hub/` |
-
-`name` is only used as the local filename for URL sources. HF sources
-ignore it (HF manages its own cache key by content hash).
-
-```python
-from fastvideo.eval.models import ensure_checkpoint
-
-# URL: name matters
-ckpt = ensure_checkpoint(
-    "amt-s.pth",
-    source="https://huggingface.co/lalala125/AMT/resolve/main/amt-s.pth",
-)
-
-# HF single file: name is decorative
-ckpt = ensure_checkpoint(
-    "raft-things.pth",                      # ignored; HF cache uses repo+sha
-    source="OpenGVLab/VBench_Used_Models",
-    filename="raft-things.pth",
-)
-```
-
----
-
-## 6) Declaring `dependencies`
-
-Set `dependencies = ["pkg1", "pkg2"]` on your metric class with
-importable module names (not PyPI distribution names). The registry
-checks each via `importlib.util.find_spec` at instantiation time and
-raises a clean `ImportError` pointing the user at the right install
-extra:
-
-```python
-class YourMetric(BaseMetric):
-    dependencies = ["clip", "timm"]   # importable as `import clip`, `import timm`
-```
-
-If a dep is in `[project.optional-dependencies.eval-<group>]`, you do
-not need to do anything more. If it is a new dep, add it to that
-group in `pyproject.toml`.
-
----
-
-## 7) Common gotchas
-
-- The standard `git submodule update --init --recursive` is enough
-  for a benchmark; do not write a `setup.sh`. Modern-dep compat goes
-  into your `__init__.py` as runtime shims.
-- Do not modify upstream files on disk. The submodule should always
-  match its pinned SHA. Compat lives in our `__init__.py`.
-- Do not pip-install the upstream. The egg-link is a glorified
-  `sys.path.insert`, which we do directly in `__init__.py`.
-- Do not call `torch.hub.set_dir(...)` from your metric. It is done
-  globally in `fastvideo/eval/__init__.py`.
-- Do not put cache-redirection env vars in your metric's `setup()`.
-  By the time `setup()` runs, the library has likely already cached
-  the default-location decision. Set env vars at package-init time.
-- Skip rather than raise when an input is missing. Use
-  `self._skip(sample, reason)` for any expected-missing input. It
-  returns a list of `MetricResult(score=None)` so other metrics in
-  the same evaluator continue.
-- Watch for upstream re-registration conflicts. If the upstream uses
-  a global registry (detectron2's `META_ARCH_REGISTRY`, MMCV, etc.),
-  loading the same model twice in the same process will throw. The
-  evaluator already loads each metric once; if you write a custom
-  setup-then-call pattern, mirror that single-load discipline.
-
----
-
-## 8) Training-time eval: keep evaluators hot, free caches between calls
-
-When wiring eval into a training loop, the working pattern is:
-
-1. Construct the `Evaluator` once and attach it to the pipeline
-   (`self._eval = create_evaluator(...)`). Do not recreate it per
-   validation round; that re-pays the model load cost.
-2. Save validation videos to disk (the diffusion path already does
-   this). Pass paths to `evaluator.evaluate`, not in-memory tensors
-   that share GPU memory with the training model.
-3. Run validation only on rank 0 of each sequence-parallel group.
-   Gather paths from other ranks and let rank 0 score everything.
-4. After every `evaluate(...)` call, call
-   `evaluator.release_cuda_memory()` in a `finally` block. That runs
-   `gc.collect()` + `torch.cuda.empty_cache()` +
-   `torch.cuda.ipc_collect()`. The eval model stays loaded; only
-   transient activation buffers from the just-finished call get
-   freed:
-
-   ```python
-   for video_path in batch:
-       try:
-           scores = self._eval.evaluate(video=load_video(video_path))
-       finally:
-           self._eval.release_cuda_memory()
-   ```
-
-5. If memory pressure spikes (rare on H200), call
-   `evaluator.unload()` to drop every metric reference and let the
-   GPU memory be GC'd. `unload` is reversible:
-   `evaluator.reload()` rebuilds the same metrics with the original
-   config (re-paying the model load cost). Calling `evaluate`
-   between `unload` and `reload` raises a clear `RuntimeError`.
-
-For most metrics (sub-1 GB backbones, e.g. CLIP/DINO/RAFT/AMT) the
-eval model can stay co-resident with the training model in
-`transformer.eval()` mode without any swap. For larger ones
-(VideoScore2 at 14 GB), measure first; if it fits on the rank-0 GPU
-during validation (training model in eval mode means no
-grads/optimizer updates), keep it hot. If not, `unload` between
-rounds.
-
-## 9) Local verification
-
-Native and library-wrapped metrics: a single-GPU smoke is enough.
-
-```python
-import torch
-from fastvideo.eval import create_evaluator
-
-ev = create_evaluator(metrics=["<group>.<your_metric>"], device="cuda")
-video = torch.randn(1, 49, 3, 256, 256, device="cuda").clamp(0, 1)
-print(ev.evaluate(video=video))
-```
-
-Submodule-wrapped metrics: also do a parity check against the
-upstream once. Clone upstream into a separate venv, run the same
-video through both, and expect an exact match on bit-deterministic
-metrics and ≤1% drift on backbone-heavy ones (driven by
-transformers/torch version differences).
-
-For quick parity in CI: pin a tiny test video, record expected
-scores ± tolerance, and add a calibration test under
-`fastvideo/tests/eval/`.
-
----
-
-## 10) When not to add a metric
-
-- **Set-vs-set distribution metrics** (FVD, FID-style) do not fit
-  `BaseMetric.compute(sample)` cleanly; they need a population.
-  Adding them requires a stateful accumulator interface that does
-  not exist yet. Open an issue first.
-- **Metrics requiring a single-GPU model larger than available
-  memory.** Eval is not the place for tensor-parallel sharding;
-  metrics are expected to fit on one GPU.
-- **Metrics that need `mmcv` with a conflicting CUDA ABI.** Document
-  the affected sub-metrics as unsupported and skip them. Building
-  isolation infrastructure (subprocess engine, per-metric venv) is
-  out of scope.

docs/contributing/overview.md

@@ -49,7 +49,7 @@ git clone https://github.com/hao-ai-lab/FastVideo.git && cd FastVideo
 Install FastVideo in editable mode and set up hooks:
 
 ```bash
-uv pip install -e .[dev]
+uv pip install -e ".[dev]"
 
 # Optional: FlashAttention (builds native kernels)
 uv pip install flash-attn --no-build-isolation -v

docs/design/inference_schema_parity_inventory.yaml

@@ -306,6 +306,30 @@ surfaces:
         sources: [fastvideo.configs.pipelines.wan.MatrixGameI2V480PConfig]
       num_frames_per_block:
         sources: [fastvideo.configs.pipelines.wan.MatrixGameI2V480PConfig]
+      audio_channels:
+        sources:
+          - fastvideo.configs.pipelines.stable_audio.StableAudioT2AConfig
+          - fastvideo.configs.pipelines.stable_audio.StableAudioOpenSmallConfig
+      audio_end_in_s:
+        sources:
+          - fastvideo.configs.pipelines.stable_audio.StableAudioT2AConfig
+          - fastvideo.configs.pipelines.stable_audio.StableAudioOpenSmallConfig
+      audio_start_in_s:
+        sources:
+          - fastvideo.configs.pipelines.stable_audio.StableAudioT2AConfig
+          - fastvideo.configs.pipelines.stable_audio.StableAudioOpenSmallConfig
+      max_audio_duration_s:
+        sources:
+          - fastvideo.configs.pipelines.stable_audio.StableAudioT2AConfig
+          - fastvideo.configs.pipelines.stable_audio.StableAudioOpenSmallConfig
+      sample_size:
+        sources:
+          - fastvideo.configs.pipelines.stable_audio.StableAudioT2AConfig
+          - fastvideo.configs.pipelines.stable_audio.StableAudioOpenSmallConfig
+      sampling_rate:
+        sources:
+          - fastvideo.configs.pipelines.stable_audio.StableAudioT2AConfig
+          - fastvideo.configs.pipelines.stable_audio.StableAudioOpenSmallConfig
     compatibility_only:
       batch_size: "Gen3C inference-only tuning field pending typed batching design."
       gradient_checkpointing: "Gen3C inference-only compatibility field pending typed batching design."
@@ -380,6 +404,13 @@ surfaces:
       ltx2_stg_scale_audio: request.extensions.ltx2.stg_scale_audio
       ltx2_stg_blocks_video: request.extensions.ltx2.stg_blocks_video
       ltx2_stg_blocks_audio: request.extensions.ltx2.stg_blocks_audio
+      audio_start_in_s: request.extensions.stable_audio.audio_start_in_s
+      audio_end_in_s: request.extensions.stable_audio.audio_end_in_s
+      init_audio: request.extensions.stable_audio.init_audio
+      init_audio_strength: request.extensions.stable_audio.init_audio_strength
+      init_noise_level: request.extensions.stable_audio.init_noise_level
+      inpaint_audio: request.extensions.stable_audio.inpaint_audio
+      inpaint_mask: request.extensions.stable_audio.inpaint_mask
     internal_only:
       data_type: "Derived from the request shape and not a public input."
 

docs/design/training_architecture.md

@@ -243,7 +243,7 @@ for step in range(start_step, max_steps):
 
 ```bash
 # Install
-uv pip install -e .[dev]
+uv pip install -e ".[dev]"
 
 # Run DMD2 distillation on Wan 2.1
 torchrun --nproc_per_node=8 -m fastvideo.train.entrypoint.train \

docs/getting_started/installation.md

@@ -27,7 +27,7 @@ uv pip install fastvideo
 conda create -n fastvideo python=3.12 -y
 conda activate fastvideo
 
-pip install fastvideo
+uv pip install fastvideo
 ```
 
 ### From source
@@ -41,11 +41,11 @@ uv pip install -e .
 uv pip install flash-attn --no-build-isolation -v
 ```
 
-Alternative with Conda environment:
+Alternative with Conda environment (still drives installs through `uv`):
 
 ```bash
-pip install -e .
-pip install flash-attn --no-build-isolation -v
+uv pip install -e .
+uv pip install flash-attn --no-build-isolation -v
 ```
 
 ## Hardware Requirements

docs/getting_started/installation/gpu.md

@@ -58,14 +58,16 @@ uv pip install flash-attn --no-build-isolation -v
 
 #### With Conda environment (alternative)
 
+`uv` works inside an active conda env too, so prefer `uv pip` for the actual install:
+
 ```bash
-pip install fastvideo
+uv pip install fastvideo
 ```
 
 Also optionally install FlashAttention:
 
 ```bash
-pip install flash-attn --no-build-isolation -v
+uv pip install flash-attn --no-build-isolation -v
 ```
 
 ### Installation from Source
@@ -87,7 +89,7 @@ uv pip install -e .
 Alternative with Conda environment:
 
 ```bash
-pip install -e .
+uv pip install -e .
 ```
 
 ### Optional Dependencies
@@ -101,7 +103,7 @@ uv pip install flash-attn --no-build-isolation -v
 Alternative with Conda environment:
 
 ```bash
-pip install flash-attn --no-build-isolation -v
+uv pip install flash-attn --no-build-isolation -v
 ```
 
 ## Set up using Docker

docs/getting_started/installation/mps.md

@@ -57,8 +57,10 @@ uv pip install fastvideo
 
 #### With Conda environment (alternative)
 
+`uv` works inside an active conda env too, so prefer `uv pip` for the actual install:
+
 ```bash
-pip install fastvideo
+uv pip install fastvideo
 ```
 
 ### Installation from Source
@@ -80,7 +82,7 @@ uv pip install -e .
 Alternative with Conda environment:
 
 ```bash
-pip install -e .
+uv pip install -e .
 ```
 
 ## Development Environment Setup

docs/inference/gen3c.md

@@ -19,7 +19,7 @@
 - Install MoGe:
 
 ```bash
-pip install git+https://github.com/microsoft/MoGe.git
+uv pip install git+https://github.com/microsoft/MoGe.git
 ```
 
 - If you hit `ImportError: libGL.so.1` (common on Ubuntu/headless nodes), you can try installing OpenCV runtime libs:

docs/inference/optimizations.md

@@ -54,7 +54,7 @@ FASTVIDEO_ATTENTION_BACKEND=SAGE_ATTN python example.py
 We recommend always installing [Flash Attention 2](https://github.com/Dao-AILab/flash-attention):
 
 ```bash
-pip install flash-attn==2.7.4.post1 --no-build-isolation
+uv pip install flash-attn==2.7.4.post1 --no-build-isolation
 ```
 
 And if using a Hopper+ GPU (ie H100), installing [Flash Attention 3](https://github.com/Dao-AILab/flash-attention?tab=readme-ov-file#flashattention-3-beta-release) by compiling it from source (takes about 10 minutes for me):
@@ -63,7 +63,7 @@ And if using a Hopper+ GPU (ie H100), installing [Flash Attention 3](https://git
 git clone https://github.com/Dao-AILab/flash-attention.git && cd flash-attention
 
 cd hopper
-pip install ninja
+uv pip install ninja
 python setup.py install
 ```
 
@@ -98,7 +98,7 @@ To use [SageAttention](https://github.com/thu-ml/SageAttention) 2.1.1, please co
 ```bash
 git clone https://github.com/thu-ml/SageAttention.git
 cd sageattention
-python setup.py install  # or pip install -e .
+python setup.py install  # or uv pip install -e .
 ```
 
 ### Sage Attention 3

examples/distill/Wan2.1-T2V/Wan-Syn-Data-480P/README.md

@@ -4,7 +4,7 @@ These are end-to-end example scripts for distilling Wan2.1 T2V 1.3B model using
 ### 0. Make sure you have installed VSA
 
 ```bash
-pip install vsa
+uv pip install vsa
 ```
 
 ### 1. Download dataset:

examples/distill/Wan2.2-TI2V-5B-Diffusers/Data-free/README.md

@@ -4,7 +4,7 @@ These are end-to-end example scripts for distilling Wan2.2 TI2V 5B model DMD+VSA
 ### 0. Make sure you have installed VSA
 
 ```bash
-pip install vsa
+uv pip install vsa
 ```
 
 ### Data-free Distillation

examples/distill/Wan2.2-TI2V-5B-Diffusers/crush_smol/README.md

@@ -4,7 +4,7 @@ These are end-to-end example scripts for distilling Wan2.2 TI2V 5B model DMD+VSA
 ### 0. Make sure you have installed VSA
 
 ```bash
-pip install vsa
+uv pip install vsa
 ```
 
 ### 1. Download dataset:

examples/inference/basic/basic_gen3c.py

@@ -7,7 +7,7 @@ and the GEN3C diffusion model.
 
 Requirements:
   1. Install MoGe:
-     pip install git+https://github.com/microsoft/MoGe.git
+     uv pip install git+https://github.com/microsoft/MoGe.git
      If you hit `ImportError: libGL.so.1`, install:
      sudo apt-get update && sudo apt-get install -y libgl1 libglib2.0-0 libsm6 libxext6 libxrender1
   2. Download and convert weights:

examples/inference/basic/basic_stable_audio.py

@@ -0,0 +1,77 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Stable Audio Open 1.0 — text-to-audio (baseline) example.
+
+User story (game-audio designer, prototyping):
+    "I'm prototyping a level and I need 6 seconds of background
+    ambience — gentle wind, distant thunder, a hint of birdsong. I
+    don't want to dig through a sound library; I want to type what I
+    hear in my head and get a wav back. If it's wrong I'll iterate
+    on the prompt. This is the first stop."
+
+User story (musician sketching ideas):
+    "I want to bounce a 30s lo-fi drum loop to use as a placeholder
+    bed while I build the rest of the track. Type prompt, get audio,
+    drop into the DAW. The actual production beat I'll record
+    myself, but I need *something* to write the chords against."
+
+User story (researcher exploring the model):
+    "First time touching Stable Audio Open — what does it sound
+    like at default settings? This is the smallest amount of code
+    that goes from prompt to mp4."
+
+How it works:
+    Pure text-to-audio (T2A). The pipeline runs:
+        T5 + NumberConditioner -> StableAudioDiT -> Oobleck VAE
+    via the `dpmpp-3m-sde` k-diffusion sampler. All components are
+    FastVideo-native — no diffusers / transformers model imports at
+    runtime (see REVIEW item 30). Mirrors upstream
+    `stable_audio_tools.inference.generation.generate_diffusion_cond`
+    bit-for-bit (~0.2% abs_mean drift on 25 steps).
+
+Tunable knobs (the "creative dials"):
+    audio_end_in_s
+        1–6   — quick ideation (sub-10s wall clock at 100 steps)
+        10–30 — full musical phrase / loop length (the README example
+                uses 30s)
+        47.5  — model maximum (full sample_size = 2097152 / 44100 Hz)
+    num_inference_steps
+        25  — fast preview, occasional artifacts
+        100 — preset default (matches the HF model card)
+        250 — diminishing returns past here
+    guidance_scale
+        3   — looser, more variation per seed
+        7   — preset default; matches README
+        12+ — sharper but can sound "fried"
+
+Prerequisites:
+  1. Accept the terms on https://huggingface.co/stabilityai/stable-audio-open-1.0
+     and export your HF token in the shell:
+         export HF_TOKEN=hf_...
+  2. Install optional inference deps (one-time):
+         uv pip install k_diffusion einops_exts alias_free_torch torchsde
+"""
+from fastvideo import VideoGenerator
+
+PROMPT = "Lo-fi hip hop instrumental with vinyl crackle and gentle piano."
+
+
+def main() -> None:
+    generator = VideoGenerator.from_pretrained(
+        "FastVideo/stable-audio-open-1.0-Diffusers",
+        num_gpus=1,
+    )
+    output_path = "outputs_audio/stable_audio_basic/output_stable_audio.wav"
+    generator.generate_video(
+        prompt=PROMPT,
+        output_path=output_path,
+        save_video=True,
+        # 6-second clip; the model max is ~47.5s.
+        audio_end_in_s=6.0,
+        # The registered preset gives 100 steps + CFG=7.0 by default;
+        # override num_inference_steps / guidance_scale here for QA.
+    )
+    generator.shutdown()
+
+
+if __name__ == "__main__":
+    main()

examples/inference/basic/basic_stable_audio_a2a.py

@@ -0,0 +1,77 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Stable Audio Open 1.0 — audio-to-audio variation example.
+
+User story (musician, late at night):
+    "I generated this 12-second lo-fi loop earlier and I love the chord
+    progression and overall vibe, but the snare hit at 0:08 sounds wrong
+    and the rhythm feels stiff. I don't want to start over from scratch
+    and lose what's working — I want the model to keep the harmony and
+    mood but reroll the percussion + groove."
+
+User story (sound designer, on a deadline):
+    "I have one good 'sword clang' SFX. The art director wants 8 sibling
+    variations that all feel like the same sword from different angles —
+    same metal, same weight, slightly different impact. I'd rather
+    refine my one good take than text-prompt my way through 50 misses."
+
+Pass `init_audio=path/to/clip` (any wav/mp3/mp4/m4a/flac the standard
+deps decode) and the model will use it as a starting point for the
+text prompt instead of pure noise.
+
+Picking `init_audio_strength` (0.0 to 1.0):
+
+    Higher = closer to the source clip. Lower = more transformation.
+    (Same convention as the "Input Audio Strength" slider in
+    Stability's commercial Stable Audio web UI, so values transfer
+    directly.)
+
+      | strength | what you get                                       |
+      |----------|----------------------------------------------------|
+      |   1.00   | Output ≈ reference. No transformation.             |
+      |   0.85   | Texture micro-variation only.                      |
+      |   0.70   | Light reroll, same instruments.                    |
+      |   0.60   | Default. Instrument identity is replaceable        |
+      |          | (cello can take over from piano on the same notes).|
+      |   0.50   | Heavy — only melody / chord progression survives.  |
+      |   0.30   | Reference acts as a loose mood prompt.             |
+      |   0.00   | Plain T2A — reference ignored.                     |
+
+    Rule of thumb by intent:
+      * "Fix one part of this clip"            -> 0.75 .. 0.85
+      * "Same notes, different instrument"     -> 0.55 .. 0.65
+      * "Same chord progression, new content"  -> 0.40 .. 0.55
+      * "Use this as a loose mood prompt"      -> 0.20 .. 0.35
+
+    If the reference timbre is bleeding through more than you want,
+    lower it; if the structure is gone, raise it.
+
+Prerequisites: same as `basic_stable_audio.py`.
+"""
+from fastvideo import VideoGenerator
+
+PROMPT = "Change the piano to a cello playing the same notes"
+# Path to any audio-bearing file (wav, mp3, mp4, m4a, flac, ...).
+# Set to `None` to skip A2A and run plain T2A.
+INIT_AUDIO_PATH: str | None = None
+# Reference fidelity in [0, 1] -- higher = closer to source.
+INIT_AUDIO_STRENGTH = 0.6
+
+
+def main() -> None:
+    generator = VideoGenerator.from_pretrained(
+        "FastVideo/stable-audio-open-1.0-Diffusers",
+        num_gpus=1,
+    )
+    generator.generate_video(
+        prompt=PROMPT,
+        output_path="outputs_audio/stable_audio_a2a/output_a2a.wav",
+        save_video=True,
+        audio_end_in_s=6.0,
+        init_audio=INIT_AUDIO_PATH,
+        init_audio_strength=INIT_AUDIO_STRENGTH,
+    )
+    generator.shutdown()
+
+
+if __name__ == "__main__":
+    main()

examples/inference/basic/basic_stable_audio_inpaint.py

@@ -0,0 +1,84 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Stable Audio Open 1.0 — inpainting / outpainting (loop extension) example.
+
+User story (loop extension — the killer app):
+    "I have a 6-second drum loop my client likes. They want it as
+    background bed for a 30-second ad. I need it to loop seamlessly,
+    but a hard cut every 6s sounds bad. Let me extend it to 30s,
+    keeping the first 6s exactly as-is and letting the model continue
+    the groove for the remaining 24s."
+
+User story (audio repair):
+    "There's a microphone bump at 0:14 in this 30-second field
+    recording — really obvious in headphones. Mask out 0:13 to 0:15
+    and let the model regenerate plausible ambience that blends in.
+    Everything else stays exactly as I recorded it."
+
+User story (transition smoothing):
+    "I have two 10-second clips I want to crossfade. Mask out a 1s
+    overlap region in the middle and let the model invent a coherent
+    transition between the two."
+
+How it works (RePaint-style blending):
+    Stable Audio Open 1.0 wasn't trained as an inpainting model
+    (`model_type=diffusion_cond`, not `diffusion_cond_inpaint`), so we
+    can't use the upstream's mask-conditioned approach directly. We
+    use the RePaint trick instead, which works on any v-prediction
+    diffusion model:
+
+      1. Encode the reference clip into latent space.
+      2. At every denoising step `i`, replace the kept region of the
+         in-flight latent (where mask == 1) with the reference
+         re-noised to the next timestep's sigma. Only the unkept
+         region (mask == 0) is freely denoised.
+      3. After the loop, the kept region is exactly the reference;
+         the unkept region is freshly generated content.
+
+    This is approximate compared to a properly trained inpainting
+    checkpoint — the seam between kept/unkept can have slight EQ
+    discontinuity — but it works on the existing public model.
+
+Tunable: the mask is a 1-D tensor in {0, 1} at the model's sample
+rate. Conventions:
+    1.0 = keep this sample from the reference
+    0.0 = regenerate this sample
+
+Prerequisites: same as `basic_stable_audio.py`.
+"""
+import os
+
+from fastvideo import VideoGenerator
+
+PROMPT = "Steady lo-fi hip hop drum loop with vinyl crackle."
+# Required: path to the reference audio file (wav, mp3, mp4, m4a, flac,
+# ...) you want to extend or repair. The pipeline raises if a mask is
+# passed without a reference, so this must be a real path.
+REFERENCE_AUDIO_PATH = "path/to/your/loop.wav"
+KEEP_SECONDS = 6.0       # first KEEP_SECONDS preserved exactly
+TOTAL_SECONDS = 12.0     # extend the loop to this duration
+
+
+def main() -> None:
+    if not os.path.isfile(REFERENCE_AUDIO_PATH):
+        raise FileNotFoundError(
+            f"REFERENCE_AUDIO_PATH={REFERENCE_AUDIO_PATH!r} does not exist. "
+            "Edit this script to point at a real audio file (wav/mp3/mp4/"
+            "m4a/flac) before running.")
+    generator = VideoGenerator.from_pretrained(
+        "FastVideo/stable-audio-open-1.0-Diffusers",
+        num_gpus=1,
+    )
+    generator.generate_video(
+        prompt=PROMPT,
+        output_path="outputs_audio/stable_audio_inpaint/output_inpaint.wav",
+        save_video=True,
+        audio_end_in_s=TOTAL_SECONDS,
+        inpaint_audio=REFERENCE_AUDIO_PATH,
+        # Tuple form: keep first KEEP_SECONDS, regenerate the rest.
+        inpaint_mask=(KEEP_SECONDS, TOTAL_SECONDS),
+    )
+    generator.shutdown()
+
+
+if __name__ == "__main__":
+    main()

examples/inference/basic/basic_stable_audio_small.py

@@ -0,0 +1,53 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Stable Audio Open Small — fast / lightweight T2A example.
+
+User story (interactive UI builder):
+    "I'm building a sound-design UI where the user types a prompt and
+    we want sub-2-second feedback so the experience feels like
+    autocomplete, not a render queue. The full Stable Audio Open 1.0
+    takes ~8s on a single GPU; the small variant takes a fraction of
+    that — quality is lower but completely usable for real-time
+    iteration."
+
+User story (overnight batch jobs):
+    "I'm generating 10,000 short SFX variants for a procedural game.
+    Wall-clock matters more than per-clip polish — give me the small
+    model so I can fit the run in one night instead of a week."
+
+How it works:
+    The small variant is a separate Stability AI checkpoint
+    (`stabilityai/stable-audio-open-small`) that ships the same Oobleck
+    VAE as the 1.0 base model but a smaller / faster DiT (`embed_dim=1024`,
+    `depth=16`, `qk_norm="ln"`) and only one duration conditioner
+    (`seconds_total`, no `seconds_start`). FastVideo loads from the
+    converted Diffusers-format repo `FastVideo/stable-audio-open-small-Diffusers`
+    via the standard component loader; per-variant arch fields come
+    from `transformer/config.json` and `conditioner/config.json`.
+
+Prerequisites: same as `basic_stable_audio.py`. The converted repo is
+public so no gated-access flow is required.
+"""
+from fastvideo import VideoGenerator
+
+PROMPT = "Lo-fi hip hop instrumental with vinyl crackle and gentle piano."
+
+
+def main() -> None:
+    generator = VideoGenerator.from_pretrained(
+        "FastVideo/stable-audio-open-small-Diffusers",
+        num_gpus=1,
+    )
+    output_path = "outputs_audio/stable_audio_small/output_stable_audio_small.wav"
+    generator.generate_video(
+        prompt=PROMPT,
+        output_path=output_path,
+        save_video=True,
+        # Small variant trains on a ~11.9s window — keep `audio_end_in_s`
+        # at or below that.
+        audio_end_in_s=6.0,
+    )
+    generator.shutdown()
+
+
+if __name__ == "__main__":
+    main()

examples/inference/eval/basic_ltx2_eval.py

@@ -1,100 +0,0 @@
-"""Generate one LTX2 video and score it with VBench metrics.
-
-The generation block is the same as
-``examples/inference/basic/basic_ltx2.py`` — same prompt, same model,
-same shape, same num_frames. After ``shutdown()`` the script loads the
-mp4 back, builds a single :class:`fastvideo.eval.Evaluator`, and runs
-the prompt-aware VBench subset that's meaningful for an arbitrary
-text→video sample.
-
-The first run downloads CLIP / DINO / RAFT / AMT / ViCLIP / MUSIQ
-weights to ``~/.cache/fastvideo/eval/`` (~few GB total).
-
-GPU memory caveat
------------------
-Scoring 1088×1920×121 with all 8 metrics needs a dedicated GPU (~80 GB).
-On a shared GPU, ``vbench.motion_smoothness`` (AMT correlation volume)
-will OOM — its memory autoscale reads ``total_memory`` rather than
-``mem_get_info()`` free memory and therefore underestimates the
-required scale-down. Drop ``motion_smoothness`` from ``METRICS`` if
-sharing, or run on a smaller-resolution generation.
-"""
-import torch
-
-from fastvideo import VideoGenerator
-from fastvideo.eval import Evaluator
-from fastvideo.eval.io import build_eval_kwargs
-
-PROMPT = (
-    "A warm sunny backyard. The camera starts in a tight cinematic close-up "
-    "of a woman and a man in their 30s, facing each other with serious "
-    "expressions. The woman, emotional and dramatic, says softly, \"That's "
-    "it... Dad's lost it. And we've lost Dad.\" The man exhales, slightly "
-    "annoyed: \"Stop being so dramatic, Jess.\" A beat. He glances aside, "
-    "then mutters defensively, \"He's just having fun.\" The camera slowly "
-    "pans right, revealing the grandfather in the garden wearing enormous "
-    "butterfly wings, waving his arms in the air like he's trying to take "
-    "off. He shouts, \"Wheeeew!\" as he flaps his wings with full commitment. "
-    "The woman covers her face, on the verge of tears. The tone is deadpan, "
-    "absurd, and quietly tragic."
-)
-
-# VBench sub-metrics meaningful for an arbitrary text→video sample
-# (just the generated frames, optionally fps + the source prompt).
-# Structured-prompt metrics (vbench.color, vbench.multiple_objects,
-# vbench.scene, ...) are excluded — they need prompts built to a
-# specific schema.
-METRICS = [
-    "vbench.aesthetic_quality",       # CLIP + LAION aesthetic head
-    "vbench.subject_consistency",     # DINO frame-to-first cosine
-    "vbench.background_consistency",  # DINO on background patches
-    "vbench.imaging_quality",         # pyiqa MUSIQ
-    "vbench.temporal_flickering",     # pixel-wise frame deltas
-    "vbench.motion_smoothness",       # AMT frame interpolator residual
-    "vbench.dynamic_degree",          # RAFT optical-flow magnitude (needs fps)
-    "vbench.overall_consistency",     # ViCLIP video↔prompt similarity
-]
-
-
-def main() -> None:
-    # ----- generation (matches examples/inference/basic/basic_ltx2.py) -----
-    generator = VideoGenerator.from_pretrained(
-        "Davids048/LTX2-Base-Diffusers",
-        num_gpus=1,
-    )
-
-    output_path = "outputs_video/ltx2_basic/output_ltx2_base_t2v_1088_1920_1.1.mp4"
-    generator.generate_video(
-        prompt=PROMPT,
-        output_path=output_path,
-        save_video=True,
-        num_frames=121,
-        height=1088,
-        width=1920,
-    )
-    generator.shutdown()
-    # Free residual CUDA memory the generator left behind so the
-    # evaluator can grab the largest possible workspace for AMT/RAFT.
-    torch.cuda.empty_cache()
-
-    # ----- scoring -----
-    print(f"\n[eval] building evaluator: {METRICS}")
-    evaluator = Evaluator(metrics=METRICS)
-
-    # LTX2 outputs at 24 fps by default.
-    sample = build_eval_kwargs({"prompt": PROMPT}, output_path, fps=24.0)
-    print(f"[eval] running ({sample['video'].shape[1]} frames @ 24 fps)...")
-    results = evaluator.evaluate(**sample)
-
-    print("\n=== VBench scores ===")
-    for name in METRICS:
-        r = results[name]
-        if r.score is None:
-            reason = r.details.get("skipped", "no score")
-            print(f"  {name}: SKIPPED ({reason})")
-        else:
-            print(f"  {name}: {r.score:.4f}")
-
-
-if __name__ == "__main__":
-    main()

examples/inference/eval/bench_physics_iq.py

@@ -1,157 +0,0 @@
-"""End-to-end Physics-IQ: dataset → generate → score → aggregate.
-
-Generates one video per take-1 scenario with LTX2 (using the scenario
-caption as the prompt), scores each generated video against the take-1
-reference and the take-2 "physical-variance" reference, and prints
-aggregate scores using :meth:`PhysicsIQMetric.aggregate_components` —
-the official scoring recipe from the upstream benchmark.
-
-Reference videos / masks / switch-frames auto-fetch on first miss into
-``${FASTVIDEO_EVAL_CACHE}/datasets/physics_iq/``; pass ``--dataset-root``
-to point at a pre-downloaded mirror instead.
-
-Quick smoke run on 4 scenarios across 2 GPUs::
-
-    python examples/inference/eval/bench_physics_iq.py \\
-        --limit 4 --num-gpus 2 \\
-        --videos-dir outputs_video/physics_iq_smoke
-
-Re-score existing generations without regenerating::
-
-    python examples/inference/eval/bench_physics_iq.py \\
-        --videos-dir outputs_video/physics_iq_smoke \\
-        --skip-generation
-"""
-from __future__ import annotations
-
-import argparse
-import json
-from pathlib import Path
-
-from fastvideo.eval import create_evaluator, get_metric
-from fastvideo.eval.datasets import get_dataset
-
-
-def _expected_filename(row: dict) -> str:
-    """Filename Physics-IQ expects for the generated video for *row*.
-
-    Uses the dataset's own ``expected_gen_filename`` annotation so the
-    output filenames match the benchmark's manifest convention.
-    """
-    return row["auxiliary_info"]["expected_gen_filename"]
-
-
-def _generate_videos(rows: list[dict], videos_dir: Path,
-                     model: str, num_gpus: int,
-                     num_frames: int, height: int, width: int) -> None:
-    from fastvideo import VideoGenerator
-
-    videos_dir.mkdir(parents=True, exist_ok=True)
-    todo = [(row, videos_dir / _expected_filename(row)) for row in rows]
-    todo = [(row, out) for (row, out) in todo if not out.is_file()]
-    if not todo:
-        print(f"[gen] all {len(rows)} videos already present; skipping.")
-        return
-
-    print(f"[gen] {len(todo)}/{len(rows)} scenarios to render with {model} "
-          f"({num_frames}x{height}x{width})...")
-    gen = VideoGenerator.from_pretrained(model, num_gpus=num_gpus)
-    try:
-        for row, out_path in todo:
-            gen.generate_video(
-                prompt=row["prompt"], output_path=str(out_path), save_video=True,
-                num_frames=num_frames, height=height, width=width,
-            )
-    finally:
-        gen.shutdown()
-
-
-def main() -> None:
-    p = argparse.ArgumentParser(description=__doc__,
-                                formatter_class=argparse.RawDescriptionHelpFormatter)
-    p.add_argument("--dataset-root", type=Path, default=None,
-                   help="Path to a pre-downloaded Physics-IQ release. "
-                        "Defaults to ${FASTVIDEO_EVAL_CACHE}/datasets/physics_iq, "
-                        "auto-fetching missing assets from the public bucket.")
-    p.add_argument("--videos-dir", type=Path,
-                   default=Path("outputs_video/bench_physics_iq"),
-                   help="Where to read/write generated videos.")
-    p.add_argument("--limit", type=int, default=None,
-                   help="Truncate to first N scenarios for smoke runs.")
-    p.add_argument("--num-gpus", type=int, default=1)
-    p.add_argument("--model", default="Davids048/LTX2-Base-Diffusers",
-                   help="HF repo id of the text→video generator to use.")
-    p.add_argument("--num-frames", type=int, default=121)
-    p.add_argument("--height", type=int, default=1088)
-    p.add_argument("--width", type=int, default=1920)
-    p.add_argument("--skip-generation", action="store_true",
-                   help="Re-score existing videos under --videos-dir.")
-    p.add_argument("--scores-out", type=Path, default=None,
-                   help="Where to write per-scenario scores (JSON). "
-                        "Defaults to <videos-dir>/scores.json.")
-    args = p.parse_args()
-
-    # 1. Walk the Physics-IQ corpus. Pass --limit to the dataset
-    # constructor so auto-download only fetches the assets we'll use.
-    ds = get_dataset("physics_iq", dataset_root=args.dataset_root, limit=args.limit)
-    rows = list(ds)
-    print(f"[load] Physics-IQ: {len(rows)} scenarios from {ds.dataset_dir}")
-
-    # 2. Generate (or reuse) one mp4 per scenario.
-    if not args.skip_generation:
-        _generate_videos(
-            rows, args.videos_dir, args.model, args.num_gpus,
-            args.num_frames, args.height, args.width,
-        )
-
-    # 3. Score each scenario. The metric reads file paths directly out
-    #    of the row dict (reference, reference_take2, masks), so we
-    #    just attach the generated video path and forward.
-    evaluator = create_evaluator(metrics=["physics_iq"], num_gpus=args.num_gpus)
-
-    samples: list[dict] = []
-    matched: list[dict] = []
-    for row in rows:
-        video_path = args.videos_dir / _expected_filename(row)
-        if not video_path.is_file():
-            print(f"[eval] missing {video_path}; skipping.")
-            continue
-        # The physics_iq metric accepts file paths via its polymorphic
-        # input handling — no need to load the tensors here.
-        samples.append({"video": str(video_path), **row})
-        matched.append(row)
-
-    all_results = evaluator.evaluate(samples=samples)
-    evaluator.shutdown()
-
-    # 4. Aggregate per the upstream scoring recipe.
-    metric = get_metric("physics_iq")
-    components = metric.aggregate_components(
-        [r["physics_iq"] for r in all_results]
-    )
-
-    print()
-    print("=== Physics-IQ aggregate ===")
-    for name, value in components.items():
-        print(f"  {name:24s}  {value:.4f}")
-
-    detailed = [
-        {
-            "scenario": row["auxiliary_info"]["scenario_id"],
-            "view": row["view"],
-            "scenario_name": row["auxiliary_info"]["scenario_name"],
-            "score": results["physics_iq"].score,
-        }
-        for row, results in zip(matched, all_results)
-    ]
-    out = args.scores_out or (args.videos_dir / "scores.json")
-    out.parent.mkdir(parents=True, exist_ok=True)
-    out.write_text(json.dumps(
-        {"aggregate": components, "per_scenario": detailed},
-        indent=2,
-    ))
-    print(f"\n[done] per-scenario scores → {out}")
-
-
-if __name__ == "__main__":
-    main()

examples/inference/eval/bench_vbench.py

@@ -1,155 +0,0 @@
-"""End-to-end VBench: dataset → generate → score → aggregate.
-
-Iterates the VBench prompt corpus, generates one video per prompt with
-LTX2, scores each generated video against the requested ``vbench.*``
-sub-metrics, and prints per-metric averages over the run.
-
-Re-running with ``--skip-generation`` reuses any mp4 already on disk
-under ``--videos-dir``, so you can iterate on metric selection without
-re-paying the generation cost.
-
-Example — quick smoke run on 4 prompts from the ``aesthetic_quality``
-dimension across 2 GPUs::
-
-    python examples/inference/eval/bench_vbench.py \\
-        --dimensions aesthetic_quality \\
-        --limit 4 --num-gpus 2 \\
-        --videos-dir outputs_video/vbench_smoke
-
-Full benchmark on a single dimension::
-
-    python examples/inference/eval/bench_vbench.py \\
-        --dimensions subject_consistency --num-gpus 8
-"""
-from __future__ import annotations
-
-import argparse
-import json
-import re
-from collections import defaultdict
-from pathlib import Path
-
-from fastvideo.eval import create_evaluator
-from fastvideo.eval.datasets import get_dataset
-
-
-def _slugify(prompt: str, max_len: int = 100) -> str:
-    """Filesystem-safe filename stem; mirrors VBench's official convention."""
-    s = re.sub(r'[\\/:*?"<>|]', "", prompt[:max_len]).strip().strip(".")
-    return re.sub(r"\s+", " ", s) or "output"
-
-
-def _generate_videos(prompts: list[str], videos_dir: Path,
-                     model: str, num_gpus: int,
-                     num_frames: int, height: int, width: int) -> None:
-    from fastvideo import VideoGenerator
-
-    videos_dir.mkdir(parents=True, exist_ok=True)
-    todo = [(p, videos_dir / f"{_slugify(p)}.mp4") for p in prompts]
-    todo = [(p, out) for (p, out) in todo if not out.is_file()]
-    if not todo:
-        print(f"[gen] all {len(prompts)} videos already present; skipping.")
-        return
-
-    print(f"[gen] {len(todo)}/{len(prompts)} prompts to render with {model} "
-          f"({num_frames}x{height}x{width})...")
-    gen = VideoGenerator.from_pretrained(model, num_gpus=num_gpus)
-    try:
-        for prompt, out_path in todo:
-            gen.generate_video(
-                prompt=prompt, output_path=str(out_path), save_video=True,
-                num_frames=num_frames, height=height, width=width,
-            )
-    finally:
-        gen.shutdown()
-
-
-def main() -> None:
-    p = argparse.ArgumentParser(description=__doc__,
-                                formatter_class=argparse.RawDescriptionHelpFormatter)
-    p.add_argument("--dimensions", default="aesthetic_quality,subject_consistency",
-                   help="Comma-separated VBench dimensions (or 'all').")
-    p.add_argument("--limit", type=int, default=None,
-                   help="Truncate to first N prompts for smoke runs.")
-    p.add_argument("--videos-dir", type=Path,
-                   default=Path("outputs_video/bench_vbench"))
-    p.add_argument("--num-gpus", type=int, default=1)
-    p.add_argument("--model", default="Davids048/LTX2-Base-Diffusers",
-                   help="HF repo id of the text→video generator to use.")
-    p.add_argument("--num-frames", type=int, default=121)
-    p.add_argument("--height", type=int, default=1088)
-    p.add_argument("--width", type=int, default=1920)
-    p.add_argument("--fps", type=float, default=24.0,
-                   help="Frame-rate annotation passed to fps-aware metrics.")
-    p.add_argument("--skip-generation", action="store_true",
-                   help="Re-score existing videos under --videos-dir without "
-                        "regenerating.")
-    p.add_argument("--scores-out", type=Path, default=None,
-                   help="Where to dump per-prompt scores as JSON. "
-                        "Defaults to <videos-dir>/scores.json.")
-    args = p.parse_args()
-
-    # 1. Pull prompts from VBench.
-    dims_arg: list[str] | str = (
-        args.dimensions if args.dimensions == "all"
-        else [d.strip() for d in args.dimensions.split(",") if d.strip()]
-    )
-    ds = get_dataset("vbench", dimensions=dims_arg)
-    rows = list(ds)[: args.limit]
-    print(f"[load] VBench: {len(rows)} prompts across {ds.dimensions}")
-
-    # 2. Generate (or reuse) one mp4 per prompt.
-    if not args.skip_generation:
-        _generate_videos(
-            [row["prompt"] for row in rows],
-            args.videos_dir, args.model, args.num_gpus,
-            args.num_frames, args.height, args.width,
-        )
-
-    # 3. Score each video against the requested vbench sub-metrics.
-    metric_names = sorted(set(f"vbench.{d}" for d in ds.dimensions))
-    print(f"[eval] metrics: {metric_names}")
-    evaluator = create_evaluator(metrics=metric_names, num_gpus=args.num_gpus)
-
-    samples: list[dict] = []
-    matched_rows: list[dict] = []
-    for row in rows:
-        video_path = args.videos_dir / f"{_slugify(row['prompt'])}.mp4"
-        if not video_path.is_file():
-            print(f"[eval] missing {video_path}; skipping this row.")
-            continue
-        # Pass the path; the worker decodes lazily so memory stays bounded.
-        samples.append({
-            "video": str(video_path),
-            "fps": args.fps,
-            **row,                                       # prompt / aux / dims
-        })
-        matched_rows.append(row)
-
-    all_results = evaluator.evaluate(samples=samples)
-    evaluator.shutdown()
-
-    # 4. Aggregate per-metric.
-    by_metric: dict[str, list[float]] = defaultdict(list)
-    detailed: list[dict] = []
-    for row, results in zip(matched_rows, all_results):
-        scores = {name: r.score for name, r in results.items()}
-        detailed.append({"prompt": row["prompt"], "scores": scores})
-        for name, score in scores.items():
-            if score is not None:
-                by_metric[name].append(score)
-
-    print()
-    print("=== per-metric averages ===")
-    for name in sorted(by_metric):
-        avg = sum(by_metric[name]) / len(by_metric[name])
-        print(f"  {name:42s}  {avg:.4f}   (n={len(by_metric[name])})")
-
-    out = args.scores_out or (args.videos_dir / "scores.json")
-    out.parent.mkdir(parents=True, exist_ok=True)
-    out.write_text(json.dumps(detailed, indent=2))
-    print(f"\n[done] per-prompt scores → {out}")
-
-
-if __name__ == "__main__":
-    main()

examples/inference/eval/eval_ltx2_vbench.py

@@ -1,167 +0,0 @@
-"""End-to-end: generate a video with LTX2 and score it with VBench.
-
-Pipeline:
-    prompt → LTX2-Base → mp4 → fastvideo.eval → vbench scores
-
-Run::
-
-    pip install -e .[eval]
-    git submodule update --init fastvideo/third_party/eval/vbench
-
-    python examples/inference/eval/eval_ltx2_vbench.py
-    # or with 4 GPUs and the distilled checkpoint:
-    python examples/inference/eval/eval_ltx2_vbench.py \
-        --model FastVideo/LTX2-Distilled-Diffusers --num-gpus 4
-
-The default metric set covers the vbench sub-metrics that are
-meaningful for an arbitrary text→video sample — i.e. those that need
-only the generated video (and optionally fps + the source prompt).
-Structured-prompt metrics like ``vbench.color``, ``vbench.scene``,
-``vbench.multiple_objects`` etc. are *not* on by default — they only
-make sense when the prompt is built to a specific schema, and they
-require GRiT/detectron2 setup. Pass them via ``--metrics`` if you have
-a matching prompt.
-
-First-time runs download CLIP, DINO, RAFT, AMT, ViCLIP, and MUSIQ
-weights to ``~/.cache/fastvideo/eval/models/`` and
-``~/.cache/torch/hub/`` (~few GB total). Subsequent runs are fast.
-"""
-from __future__ import annotations
-
-import argparse
-import json
-from pathlib import Path
-
-from fastvideo import VideoGenerator
-from fastvideo.eval import create_evaluator
-from fastvideo.eval.io import load_video
-
-
-PROMPT = (
-    "A warm sunny backyard. The camera starts in a tight cinematic close-up "
-    "of a woman and a man in their 30s, facing each other with serious "
-    "expressions. The woman, emotional and dramatic, says softly, \"That's "
-    "it... Dad's lost it. And we've lost Dad.\" The man exhales, slightly "
-    "annoyed: \"Stop being so dramatic, Jess.\" A beat. He glances aside, "
-    "then mutters defensively, \"He's just having fun.\" The camera slowly "
-    "pans right, revealing the grandfather in the garden wearing enormous "
-    "butterfly wings, waving his arms in the air like he's trying to take "
-    "off. He shouts, \"Wheeeew!\" as he flaps his wings with full commitment. "
-    "The woman covers her face, on the verge of tears. The tone is deadpan, "
-    "absurd, and quietly tragic."
-)
-
-DEFAULT_METRICS = [
-    # No-input metrics: just need the generated frames.
-    "vbench.aesthetic_quality",       # CLIP + LAION aesthetic head
-    "vbench.subject_consistency",     # DINO frame-to-first cosine
-    "vbench.background_consistency",  # DINO on background patches
-    "vbench.imaging_quality",         # pyiqa MUSIQ
-    "vbench.temporal_flickering",     # pixel-wise frame deltas
-    "vbench.motion_smoothness",       # AMT frame interpolator residual
-    # Need fps annotation:
-    "vbench.dynamic_degree",          # RAFT optical-flow magnitude
-    # Need the source prompt:
-    "vbench.overall_consistency",     # ViCLIP video↔prompt similarity
-]
-
-
-def parse_args() -> argparse.Namespace:
-    p = argparse.ArgumentParser(description=__doc__,
-                                formatter_class=argparse.RawDescriptionHelpFormatter)
-    p.add_argument("--model", default="Davids048/LTX2-Base-Diffusers",
-                   help="HF repo id of the LTX2 checkpoint.")
-    p.add_argument("--num-gpus", type=int, default=1)
-    p.add_argument("--output", default="outputs_video/ltx2_eval/clip.mp4",
-                   help="Where to save the generated mp4.")
-    p.add_argument("--num-frames", type=int, default=121)
-    p.add_argument("--height", type=int, default=1088)
-    p.add_argument("--width", type=int, default=1920)
-    p.add_argument("--prompt", default=PROMPT)
-    p.add_argument("--fps", type=float, default=24.0,
-                   help="Frame-rate annotation passed to fps-aware metrics "
-                        "(e.g. vbench.dynamic_degree). LTX2 outputs at 24 fps "
-                        "by default.")
-    p.add_argument("--metrics", default=",".join(DEFAULT_METRICS),
-                   help="Comma-separated metric names. Pass 'all' for every "
-                        "registered metric, or e.g. 'vbench' for the whole group.")
-    p.add_argument("--scores-out", default="outputs_video/ltx2_eval/scores.json")
-    p.add_argument("--skip-generation", action="store_true",
-                   help="Reuse an existing --output video instead of regenerating.")
-    return p.parse_args()
-
-
-def generate(args: argparse.Namespace) -> Path:
-    out = Path(args.output)
-    if args.skip_generation and out.is_file():
-        print(f"[gen] reusing existing video at {out}")
-        return out
-    out.parent.mkdir(parents=True, exist_ok=True)
-
-    print(f"[gen] loading {args.model} ({args.num_gpus} GPU)...")
-    generator = VideoGenerator.from_pretrained(args.model, num_gpus=args.num_gpus)
-    try:
-        print(f"[gen] generating to {out}...")
-        generator.generate_video(
-            prompt=args.prompt,
-            output_path=str(out),
-            save_video=True,
-            num_frames=args.num_frames,
-            height=args.height,
-            width=args.width,
-        )
-    finally:
-        generator.shutdown()
-    return out
-
-
-def evaluate_video(video_path: Path, prompt: str, fps: float,
-                   metric_names) -> dict:
-    print(f"[eval] loading video from {video_path}...")
-    video = load_video(str(video_path))                  # (T, C, H, W) in [0, 1]
-    video = video.unsqueeze(0)                            # → (1, T, C, H, W)
-
-    print(f"[eval] building evaluator: {metric_names}")
-    evaluator = create_evaluator(metrics=metric_names, device="cuda")
-
-    print(f"[eval] running ({video.shape[1]} frames @ {fps} fps)...")
-    results = evaluator.evaluate(
-        video=video,
-        text_prompt=[prompt],
-        fps=fps,
-    )
-
-    if isinstance(results, list):
-        results = results[0]                              # batch of 1
-
-    return {
-        name: {"score": r.score, "details": r.details}
-        for name, r in results.items()
-    }
-
-
-def main() -> None:
-    args = parse_args()
-    if args.metrics.strip() == "all":
-        metric_names = "all"
-    else:
-        metric_names = [m.strip() for m in args.metrics.split(",") if m.strip()]
-
-    video_path = generate(args)
-    scores = evaluate_video(video_path, args.prompt, args.fps, metric_names)
-
-    print("\n=== VBench scores ===")
-    for name, payload in scores.items():
-        print(f"  {name}: {payload['score']}")
-
-    out = Path(args.scores_out)
-    out.parent.mkdir(parents=True, exist_ok=True)
-    out.write_text(json.dumps(
-        {"video": str(video_path), "prompt": args.prompt, "scores": scores},
-        indent=2,
-    ))
-    print(f"[done] scores written to {out}")
-
-
-if __name__ == "__main__":
-    main()

examples/inference/eval/score_folder.py

@@ -1,95 +0,0 @@
-"""Score a folder of videos in parallel across multiple GPUs.
-
-Uses :meth:`Evaluator.evaluate(samples=[...])`, which round-robins each
-sample dict across the GPU replicas the evaluator was built with.
-
-Example::
-
-    python examples/inference/eval/score_folder.py \\
-        --videos generated/ \\
-        --metrics vbench.aesthetic_quality,vbench.subject_consistency \\
-        --num-gpus 4 \\
-        --output scores.json
-
-Pair each generated video with a same-name reference video (e.g.
-``ref/<stem>.mp4``) by passing ``--reference-dir``::
-
-    python examples/inference/eval/score_folder.py \\
-        --videos generated/ --reference-dir ref/ \\
-        --metrics common.psnr,common.ssim,common.lpips \\
-        --num-gpus 4
-"""
-from __future__ import annotations
-
-import argparse
-import json
-from pathlib import Path
-
-from fastvideo.eval import create_evaluator
-
-
-def _list_videos(directory: Path) -> list[Path]:
-    exts = {".mp4", ".avi", ".mov", ".mkv", ".gif"}
-    return sorted(p for p in directory.iterdir() if p.suffix.lower() in exts)
-
-
-def main() -> None:
-    p = argparse.ArgumentParser(description=__doc__,
-                                formatter_class=argparse.RawDescriptionHelpFormatter)
-    p.add_argument("--videos", type=Path, required=True,
-                   help="Directory of generated videos.")
-    p.add_argument("--reference-dir", type=Path, default=None,
-                   help="Directory of reference videos with matching stems.")
-    p.add_argument("--metrics", default="vbench.aesthetic_quality")
-    p.add_argument("--num-gpus", type=int, default=1)
-    p.add_argument("--fps", type=float, default=None,
-                   help="Frame-rate annotation for fps-aware metrics.")
-    p.add_argument("--output", type=Path, default=Path("scores.json"))
-    args = p.parse_args()
-
-    video_paths = _list_videos(args.videos)
-    if not video_paths:
-        raise SystemExit(f"No videos under {args.videos}")
-    print(f"Found {len(video_paths)} videos in {args.videos}")
-
-    metrics: list[str] | str = (
-        args.metrics if args.metrics == "all"
-        else [m.strip() for m in args.metrics.split(",") if m.strip()]
-    )
-    evaluator = create_evaluator(metrics=metrics, num_gpus=args.num_gpus)
-
-    # Build per-video sample dicts holding *paths*, not pre-loaded
-    # tensors. Each path is decoded inside the worker thread that picks
-    # up its sample, so peak resident memory is bounded by num_gpus
-    # rather than scaling with the size of the folder.
-    samples: list[dict] = []
-    for vp in video_paths:
-        sample: dict = {"video": str(vp)}
-        if args.reference_dir is not None:
-            ref_path = args.reference_dir / vp.name
-            if not ref_path.is_file():
-                raise FileNotFoundError(f"Missing reference for {vp.name} at {ref_path}")
-            sample["reference"] = str(ref_path)
-        if args.fps is not None:
-            sample["fps"] = args.fps
-        samples.append(sample)
-
-    print(f"Scoring with {len(evaluator.metric_names)} metric(s) "
-          f"on {evaluator.num_gpus} GPU(s)...")
-    all_results = evaluator.evaluate(samples=samples)
-    evaluator.shutdown()
-
-    payload = [
-        {
-            "video": str(vp),
-            "scores": {name: r.score for name, r in results.items()},
-        }
-        for vp, results in zip(video_paths, all_results)
-    ]
-    args.output.parent.mkdir(parents=True, exist_ok=True)
-    args.output.write_text(json.dumps(payload, indent=2))
-    print(f"Wrote {args.output}")
-
-
-if __name__ == "__main__":
-    main()

examples/inference/eval/score_video.py

@@ -1,68 +0,0 @@
-"""Score one video on one GPU.
-
-Smallest possible use of ``fastvideo.eval``: load an mp4, build an
-:class:`Evaluator` for the requested metric set, run it.
-
-Examples::
-
-    # Reference-free (just the generated video):
-    python examples/inference/eval/score_video.py \\
-        --video clip.mp4 \\
-        --metrics vbench.aesthetic_quality,vbench.imaging_quality
-
-    # Reference-paired (compare against ground truth):
-    python examples/inference/eval/score_video.py \\
-        --video gen.mp4 --reference ref.mp4 \\
-        --metrics common.psnr,common.ssim,common.lpips
-"""
-from __future__ import annotations
-
-import argparse
-import json
-
-from fastvideo.eval import create_evaluator
-from fastvideo.eval.io import load_video
-
-
-def main() -> None:
-    p = argparse.ArgumentParser(description=__doc__,
-                                formatter_class=argparse.RawDescriptionHelpFormatter)
-    p.add_argument("--video", required=True, help="Path to the generated mp4.")
-    p.add_argument("--reference", default=None,
-                   help="Optional path to a reference mp4 (for paired metrics).")
-    p.add_argument("--metrics", default="common.psnr,common.ssim",
-                   help="Comma-separated metric names, or a group name like 'vbench'.")
-    p.add_argument("--device", default="cuda:0")
-    p.add_argument("--text-prompt", default=None,
-                   help="Text prompt for prompt-aware metrics "
-                        "(vbench.overall_consistency, etc.).")
-    p.add_argument("--fps", type=float, default=None,
-                   help="Frame-rate annotation for fps-aware metrics "
-                        "(vbench.dynamic_degree, etc.).")
-    args = p.parse_args()
-
-    metrics: list[str] | str = (
-        args.metrics if args.metrics in ("all",)
-        else [m.strip() for m in args.metrics.split(",") if m.strip()]
-    )
-    evaluator = create_evaluator(metrics=metrics, device=args.device)
-
-    sample: dict = {"video": load_video(args.video)}
-    if args.reference is not None:
-        sample["reference"] = load_video(args.reference)
-    if args.text_prompt is not None:
-        sample["text_prompt"] = args.text_prompt
-    if args.fps is not None:
-        sample["fps"] = args.fps
-
-    results = evaluator.evaluate(**sample)
-    evaluator.shutdown()
-
-    print(json.dumps(
-        {name: r.score for name, r in results.items()},
-        indent=2,
-    ))
-
-
-if __name__ == "__main__":
-    main()

examples/train/configs/fine_tuning/cosmos/t2v.yaml

@@ -0,0 +1,73 @@
+# Cosmos Predict2 2B T2V finetune config.
+#
+# Data must be preprocessed with Cosmos VAE + T5 text encoder
+# into parquet format before training.
+
+models:
+  student:
+    _target_: fastvideo.train.models.cosmos.CosmosModel
+    init_from: nvidia/Cosmos-Predict2-2B-Video2World
+    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: 8
+    hsdp_shard_dim: 1
+
+  data:
+    data_path: data/cosmos_preprocessed
+    dataloader_num_workers: 4
+    train_batch_size: 1
+    training_cfg_rate: 0.0
+    seed: 1000
+    # Cosmos VAE: 4x temporal, 8x spatial compression.
+    # 93 frames -> 24 latent frames, 480x832 -> 60x104
+    num_latent_t: 24
+    num_height: 480
+    num_width: 832
+    num_frames: 93
+
+  optimizer:
+    learning_rate: 1.0e-5
+    betas: [0.9, 0.999]
+    weight_decay: 0.01
+    lr_scheduler: constant
+    lr_warmup_steps: 0
+
+  loop:
+    max_train_steps: 5000
+    gradient_accumulation_steps: 1
+
+  checkpoint:
+    output_dir: outputs/cosmos_finetune
+    training_state_checkpointing_steps: 500
+    checkpoints_total_limit: 3
+    resume_from_checkpoint: latest
+
+  tracker:
+    project_name: fastvideo_cosmos
+    run_name: cosmos_finetune
+
+  model:
+    enable_gradient_checkpointing_type: full
+
+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.cosmos.cosmos_pipeline.Cosmos2VideoToWorldPipeline
+    dataset_file: data/cosmos_preprocessed/validation_prompts.json
+    every_steps: 100
+    sampling_steps: [50]
+    guidance_scale: 6.0
+
+pipeline:
+  flow_shift: 1.0

examples/train/configs/overfit_cosmos25_t2w.yaml

@@ -0,0 +1,79 @@
+# Cosmos-Predict2.5-2B Text-to-World overfitting test config.
+#
+# Overfits on a few short videos (480x832, 93 frames) to verify the
+# Cosmos 2.5 training plugin works end-to-end.
+#
+# Preprocess data first:
+#   CUDA_VISIBLE_DEVICES=0 python fastvideo/pipelines/preprocess/preprocess_cosmos25_overfit.py
+#
+# Run:
+#   bash examples/train/run.sh examples/train/configs/overfit_cosmos25_t2w.yaml
+
+models:
+  student:
+    _target_: fastvideo.train.models.cosmos.CosmosModel
+    init_from: KyleShao/Cosmos-Predict2.5-2B-Diffusers
+    trainable: true
+    enable_gradient_checkpointing_type: full
+    flow_shift: 1.0
+
+method:
+  _target_: fastvideo.train.methods.fine_tuning.finetune.FineTuneMethod
+
+training:
+  distributed:
+    num_gpus: 1
+    sp_size: 1
+    tp_size: 1
+    hsdp_replicate_dim: 1
+    hsdp_shard_dim: 1
+
+  data:
+    data_path: data/cosmos25_overfit_preprocessed
+    dataloader_num_workers: 0
+    train_batch_size: 1
+    training_cfg_rate: 0.0
+    seed: 42
+    num_latent_t: 24
+    num_height: 480
+    num_width: 832
+    num_frames: 93
+
+  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: 300
+    gradient_accumulation_steps: 1
+
+  checkpoint:
+    output_dir: outputs/cosmos25_overfit
+    training_state_checkpointing_steps: 50
+    checkpoints_total_limit: 2
+
+  tracker:
+    project_name: fastvideo_cosmos25
+    run_name: cosmos25_overfit
+
+  model:
+    precondition_outputs: false
+    enable_gradient_checkpointing_type: full
+
+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.cosmos.cosmos2_5_pipeline.Cosmos2_5Pipeline
+    dataset_file: data/cosmos25_overfit_preprocessed/validation_prompts.json
+    every_steps: 150
+    sampling_steps: [35]
+    guidance_scale: 7.0
+
+pipeline:
+  flow_shift: 1.0

fastvideo-kernel/CMakeLists.txt

@@ -1,7 +1,7 @@
 cmake_minimum_required(VERSION 3.26 FATAL_ERROR)
 project(fastvideo-kernel LANGUAGES CXX)
 
-# Prefer environment variable (used by CI or pip install git+repo_addr) if CMake var is not explicitly set.
+# Prefer environment variable (used by CI or uv pip install git+repo_addr) if CMake var is not explicitly set.
 if(NOT DEFINED GPU_BACKEND AND DEFINED ENV{GPU_BACKEND})
     set(GPU_BACKEND "$ENV{GPU_BACKEND}")
 endif()

fastvideo-kernel/python/fastvideo_kernel/vmoba.py

@@ -11,7 +11,7 @@ except ImportError:
 
     def _unsupported(*args, **kwargs):
         raise ImportError(
-            "flash-attn is not installed. Please install it, e.g., `pip install flash-attn`."
+            "flash-attn is not installed. Please install it, e.g., `uv pip install flash-attn`."
         )
 
     _flash_attn_varlen_forward = _unsupported

fastvideo/AGENTS.md

@@ -0,0 +1,67 @@
+# `fastvideo/` — Core Package
+
+**Generated:** 2026-05-02
+
+Inference + training framework for video DiTs. Public API entry: `from fastvideo import VideoGenerator, PipelineConfig, SamplingParam`.
+
+## Public Surface (`__init__.py`)
+
+```python
+VideoGenerator   # entrypoints/video_generator.py — high-level inference handle
+PipelineConfig   # configs/pipelines/base.py — pipeline wiring dataclass
+SamplingParam    # api/sampling_param.py — runtime sampling knobs
+```
+
+CLI entry: `fastvideo` script → `entrypoints/cli/main.py` (subcommands: `generate`, `serve`, `bench`).
+
+## Layout
+
+```
+fastvideo/
+├── api/             # Schema + presets for the OpenAI-compatible serving layer
+├── attention/       # Backends + selector (FlashAttn / SageAttn / SDPA / VSA / VMoBA / SLA)
+├── configs/         # Per-model arch configs + per-pipeline configs (registry-driven)
+├── dataset/         # Dataloaders (pre-commit excluded — minimal lint surface)
+├── distributed/     # SP/TP groups, device communicators, init helpers
+├── entrypoints/     # cli/, openai/, streaming/, video_generator.py
+├── hooks/           # Runtime hook system for pipelines
+├── layers/          # Tensor-parallel linears + attention wrappers (port targets)
+├── models/          # DiT / VAE / encoder / scheduler / loader (pre-commit excluded)
+├── pipelines/       # basic/<model>/, preprocess/, stages/, training/
+├── platforms/       # CUDA/ROCm capability + AttentionBackendEnum
+├── third_party/     # Vendored externals (lint excluded; do not reformat)
+├── train/           # NEW modular trainer — methods × models × callbacks
+├── training/        # LEGACY monolithic *_training/distillation_pipeline.py
+├── worker/          # Multi-process / Ray executors
+├── workflow/        # Preprocessing workflow base class
+├── registry.py      # Pipeline-config + model-class lookup (canonical)
+├── envs.py          # Env-var declarations
+├── fastvideo_args.py# Runtime arg dataclass passed through pipelines
+└── utils.py         # FlexibleArgumentParser, qualname resolver, etc.
+```
+
+## Where to Look
+
+| Task | Location |
+|------|----------|
+| Add a new pipeline class | `pipelines/basic/<model>/` + `configs/pipelines/<model>.py` + register in `registry.py` |
+| Add a new model component | `models/<role>/<model>.py` + `configs/models/<role>/<model>.py` |
+| Wire an existing model into a new pipeline | `pipelines/basic/<model>/presets.py` + reuse stages from `pipelines/stages/` |
+| Add a converter | `scripts/checkpoint_conversion/<model>_to_*.py` (separate dir, separate AGENTS.md) |
+| Add an attention backend | `attention/backends/<name>.py` + register in selector |
+| Add a runtime CLI flag | `fastvideo_args.py` (avoid `argparse` ad-hoc inside stages) |
+
+## Conventions Specific Here
+
+- `PipelineStage` subclasses (`pipelines/stages/`) own one verb each (encode, schedule, denoise, decode). Compose, don't fork.
+- Every pipeline reads from a `PipelineConfig` subclass and a `SamplingParam`. Never read raw env vars inside a stage — go through `fastvideo.envs`.
+- Logger setup: `from fastvideo.logger import init_logger; logger = init_logger(__name__)`. Do not call `logging.getLogger` directly.
+- Imports between `train/` and `training/` are **forbidden** — they are independent stacks.
+
+## Pre-Commit Exclusions (do not assume linted)
+
+These dirs are listed in `.pre-commit-config.yaml` `exclude`:
+
+- `fastvideo/third_party/`, `fastvideo/dataset/`, `fastvideo/models/`
+
+Editing files there will NOT trigger yapf/ruff/mypy/codespell. Format manually if a sibling file shows clear style; do not introduce new violations.

fastvideo/api/results.py

@@ -15,6 +15,7 @@ class GenerationResult:
     samples: Any | None = None
     frames: Any | None = None
     audio: Any | None = None
+    audio_sample_rate: int | None = None
     size: tuple[int, int, int] | None = None
     generation_time: float | None = None
     logging_info: Any | None = None
@@ -44,6 +45,7 @@ class GenerationResult:
                 "samples",
                 "frames",
                 "audio",
+                "audio_sample_rate",
                 "size",
                 "generation_time",
                 "logging_info",
@@ -62,6 +64,7 @@ class GenerationResult:
             samples=result.get("samples"),
             frames=result.get("frames"),
             audio=result.get("audio"),
+            audio_sample_rate=result.get("audio_sample_rate"),
             size=result.get("size"),
             generation_time=result.get("generation_time"),
             logging_info=result.get("logging_info"),
@@ -80,6 +83,7 @@ class GenerationResult:
             "samples": self.samples,
             "frames": self.frames,
             "audio": self.audio,
+            "audio_sample_rate": self.audio_sample_rate,
             "size": self.size,
             "generation_time": self.generation_time,
             "logging_info": self.logging_info,

fastvideo/api/sampling_param.py

@@ -112,6 +112,33 @@ class SamplingParam:
     ltx2_stg_blocks_video: list[int] = field(default_factory=lambda: [29])
     ltx2_stg_blocks_audio: list[int] = field(default_factory=lambda: [29])
 
+    # Stable Audio (T2A): clip start/end in seconds. Honored by
+    # `StableAudioConditioningStage` + `StableAudioDecodingStage`. Other
+    # families ignore them.
+    audio_start_in_s: float | None = None
+    audio_end_in_s: float | None = None
+
+    # Stable Audio audio-to-audio (variation):
+    #   `init_audio` -- a path or `[B, C, samples]` waveform at the model
+    #                   sample rate; the pipeline encodes it via the VAE
+    #                   and uses it as the starting latent.
+    #   `init_audio_strength` -- 0..1, higher = closer to the reference
+    #                            (matches the convention of Stability's
+    #                            commercial Stable Audio 2.0 UI). 1.0 ~=
+    #                            VAE round-trip, 0.0 ~= plain T2A.
+    #   `init_noise_level` -- legacy raw `sigma_max` override (0.3..500,
+    #                         higher = more freedom). Kept for callers
+    #                         that already use it; prefer `init_audio_strength`.
+    init_audio: Any = None
+    init_audio_strength: float | None = None
+    init_noise_level: float | None = None
+
+    # Stable Audio inpainting (RePaint-style): `inpaint_audio` is the
+    # reference clip, `inpaint_mask` is a [samples] tensor in {0, 1} where
+    # 1 means *keep the reference* and 0 means *regenerate*.
+    inpaint_audio: Any = None
+    inpaint_mask: Any = None
+
     # Continuation state carried across streaming/multi-segment calls.
     continuation_state: ContinuationState | None = None
     # When True, the pipeline returns a ContinuationState on the result so

fastvideo/attention/AGENTS.md

@@ -0,0 +1,58 @@
+# `fastvideo/attention/` — Attention Backends
+
+**Generated:** 2026-05-02
+
+Backend registry + selector wrapping FlashAttn / SageAttn / SageAttn3 / SDPA / VSA / VMoBA / SLA / BSA.
+
+## Layout
+
+```
+attention/
+├── __init__.py            # Exports DistributedAttention, LocalAttention, get_attn_backend
+├── layer.py               # DistributedAttention, DistributedAttention_VSA, LocalAttention
+├── selector.py            # get_attn_backend (cached) + env-var override
+├── backends/
+│   ├── abstract.py        #   AttentionBackend / AttentionMetadata / AttentionMetadataBuilder
+│   ├── flash_attn.py      #   FA2/FA3
+│   ├── sage_attn.py       #   SageAttention v1
+│   ├── sage_attn3.py      #   SageAttention v3
+│   ├── sdpa.py            #   torch SDPA fallback
+│   ├── video_sparse_attn.py  # VSA (paper: Video Sparse Attention)
+│   ├── vmoba.py           #   Video-MoBA
+│   ├── sla.py             #   Sliding-window (STA)
+│   └── bsa_attn.py        #   Block-sparse
+└── utils/
+    ├── flash_attn_cute.py
+    └── flash_attn_no_pad.py
+```
+
+## Selection Order
+
+`get_attn_backend()` resolves via:
+
+1. Env-var override `FASTVIDEO_ATTENTION_BACKEND` (see `STR_BACKEND_ENV_VAR` in `fastvideo/utils.py`).
+2. Per-platform default from `fastvideo/platforms/`.
+3. Heuristic fallback to SDPA.
+
+The result is `@lru_cache`d. Tests that need a specific backend must use the
+`global_force_attn_backend(...)` context manager from `selector.py`, never set
+the env var mid-process.
+
+## Adding a Backend
+
+1. Subclass `AttentionBackend` in `backends/<name>.py`.
+2. Implement `AttentionMetadata` + `AttentionMetadataBuilder` for the new path.
+3. Register the enum value in `fastvideo/platforms/interface.py` (`AttentionBackendEnum`).
+4. Wire string → class resolution in `selector.py`.
+5. Verify the new backend works with `DistributedAttention` (sequence parallel)
+   and `LocalAttention` (single-rank). If it cannot support SP, document the
+   gap in the backend file's module docstring.
+
+## Anti-Patterns
+
+- Calling `torch.nn.functional.scaled_dot_product_attention` directly inside a
+  model's forward — go through `DistributedAttention` / `LocalAttention`.
+- Reading `os.environ[STR_BACKEND_ENV_VAR]` from arbitrary call sites. Use
+  `get_env_variable_attn_backend()`.
+- Caching backend instances per-module. The selector cache is process-wide; do
+  not duplicate it.

fastvideo/attention/backends/flash_attn.py

@@ -2,7 +2,6 @@
 
 import torch
 import torch.nn.functional as F
-from flash_attn import flash_attn_func as flash_attn_2_func
 from dataclasses import dataclass
 
 try:
@@ -18,6 +17,7 @@ except ImportError:
         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"
 

fastvideo/attention/backends/sla.py

@@ -405,7 +405,7 @@ class SageSLAAttentionImpl(AttentionImpl, nn.Module):
 
         if not SAGESLA_ENABLED:
             raise ImportError("SageSLA requires spas_sage_attn. "
-                              "Install with: pip install git+https://github.com/thu-ml/SpargeAttn.git")
+                              "Install with: uv pip install git+https://github.com/thu-ml/SpargeAttn.git")
 
         assert head_size in [64, 128], f"SageSLA requires head_size in [64, 128], got {head_size}"
 

fastvideo/configs/AGENTS.md

@@ -0,0 +1,53 @@
+# `fastvideo/configs/` — Config-Driven Model Registry
+
+**Generated:** 2026-05-02
+
+Two layers of dataclass configs feed every pipeline: **arch configs** (what the model is) and **pipeline configs** (how to run it).
+
+## Layout
+
+```
+configs/
+├── configs.py                  # Dataset / loader enums (DatasetType, VideoLoaderType)
+├── utils.py                    # update_config_from_args, shallow_asdict helpers
+├── backend/                    # Attention backend defaults
+├── models/
+│   ├── base.py                 # ModelConfig ABC
+│   ├── dits/                   # DiTConfig per model (wanvideo, ltx2, hunyuan, ...)
+│   ├── vaes/                   # VAEConfig per model
+│   ├── encoders/               # EncoderConfig (t5, clip, llama, qwen2_5, gemma, siglip, ...)
+│   ├── upsamplers/             # UpsamplerConfig (hunyuan15)
+│   └── audio/                  # Audio-model configs (ltx2_audio_vae, ...)
+├── pipelines/
+│   ├── base.py                 # PipelineConfig ABC + (de)serialization
+│   └── <model>.py              # Concrete configs (HunyuanConfig, WanT2V480PConfig, ...)
+└── *.json                      # Frozen reference configs for shipped models
+```
+
+## How Configs Hook Into the Registry
+
+`fastvideo/registry.py` imports every concrete `PipelineConfig` and exposes
+`get_pipeline_config_cls_from_name(...)`. Adding a new pipeline config requires:
+
+1. Subclass `PipelineConfig` in `pipelines/<model>.py`.
+2. Reference its component arch configs (DiT / VAE / encoder / upsampler).
+3. Add the import + name mapping in `fastvideo/registry.py`.
+
+Configs that do not appear in `registry.py` are unreachable from `VideoGenerator`.
+
+## Arch vs Pipeline — Where Does This Field Go?
+
+| Field type | Lives on |
+|-----------|----------|
+| Architecture constants (hidden dim, num heads, layer count) | `configs/models/<role>/<model>.py` |
+| Default sampling params (steps, cfg, shift, fps) | `configs/pipelines/<model>.py` |
+| Runtime overrides (precision, sp_size, tp_size, attention backend) | `configs/pipelines/base.py` defaults + CLI flags via `fastvideo_args.py` |
+| `param_names_mapping` for HF → FastVideo state-dict | Arch config (lives with the model definition) |
+
+If a knob is tunable per inference call → `SamplingParam`, not `PipelineConfig`.
+
+## Anti-Patterns
+
+- Hard-coding architecture constants inside model classes — always read from the arch config.
+- Using `argparse` directly here. Configs deserialize from dicts via `update_config_from_args`.
+- Importing from `fastvideo.pipelines` here. Configs are the lower layer; the dependency is one-way.

fastvideo/configs/models/base.py

@@ -48,8 +48,6 @@ class ModelConfig:
         for key, value in source_model_dict.items():
             if key in valid_fields:
                 setattr(arch_config, key, value)
-            else:
-                raise AttributeError(f"{type(arch_config).__name__} has no field '{key}'")
 
         if hasattr(arch_config, "__post_init__"):
             arch_config.__post_init__()

fastvideo/configs/models/dits/__init__.py

@@ -5,11 +5,13 @@ from fastvideo.configs.models.dits.hunyuanvideo import HunyuanVideoConfig
 from fastvideo.configs.models.dits.hunyuanvideo15 import HunyuanVideo15Config
 from fastvideo.configs.models.dits.longcat import LongCatVideoConfig
 from fastvideo.configs.models.dits.ltx2 import LTX2VideoConfig
+from fastvideo.configs.models.dits.stable_audio import StableAudioConfig
 from fastvideo.configs.models.dits.wanvideo import WanVideoConfig
 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"
+    "Cosmos25VideoConfig", "LongCatVideoConfig", "LTX2VideoConfig", "HYWorldConfig", "Kandinsky5VideoConfig",
+    "StableAudioConfig"
 ]

fastvideo/configs/models/dits/cosmos.py

@@ -50,7 +50,8 @@ class CosmosArchConfig(DiTArchConfig):
         })
 
     # Cosmos-specific config parameters based on transformer_cosmos.py
-    in_channels: int = 16
+    # in_channels includes the condition_mask channel (16 latent + 1 cond = 17)
+    in_channels: int = 17
     out_channels: int = 16
     num_attention_heads: int = 16
     attention_head_dim: int = 128

fastvideo/configs/models/dits/stable_audio.py

@@ -0,0 +1,76 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Config for the Stable Audio Open 1.0 DiT.
+
+Note: the SA pipeline bypasses the standard `ComposedPipelineBase`
+component loader because the published HF repo ships a single monolithic
+`model.safetensors` (no Diffusers-style `model_index.json` or
+per-subfolder layout). The arch fields and `param_names_mapping` here
+document the architecture and key remap so the same conventions used by
+the rest of the DiT family apply (FSDP shard conditions, supported
+attention backends, future loader integrations) — they are not currently
+consumed by `fastvideo/models/loader/fsdp_load.py` for SA.
+"""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+
+from fastvideo.configs.models.dits.base import DiTArchConfig, DiTConfig
+from fastvideo.platforms import AttentionBackendEnum
+
+
+def _is_transformer_layer(n: str, m) -> bool:
+    # Matches `transformer.layers.{i}` in the SA DiT module tree.
+    parts = n.split(".")
+    return (len(parts) >= 3 and parts[-3] == "transformer" and parts[-2] == "layers" and parts[-1].isdigit())
+
+
+@dataclass
+class StableAudioArchConfig(DiTArchConfig):
+    _fsdp_shard_conditions: list = field(default_factory=lambda: [_is_transformer_layer])
+
+    # SA's checkpoint is `stable_audio_tools` raw format (not Diffusers),
+    # so the only remaps are: strip the `model.model.` host-pipeline
+    # prefix, and rename `nn.LayerNorm`'s `gamma`/`beta` to torch's
+    # canonical `weight`/`bias`. Linear / cross-attention naming already
+    # matches FastVideo's conventions, so no further remap is needed.
+    param_names_mapping: dict = field(
+        default_factory=lambda: {
+            r"^model\.model\.(.*?)\.gamma$": r"\1.weight",
+            r"^model\.model\.(.*?)\.beta$": r"\1.bias",
+            r"^model\.model\.(.*)$": r"\1",
+        })
+
+    # SA only supports backends compatible with single-GPU LocalAttention.
+    _supported_attention_backends: tuple[AttentionBackendEnum, ...] = (
+        AttentionBackendEnum.FLASH_ATTN,
+        AttentionBackendEnum.TORCH_SDPA,
+    )
+
+    # Architecture constants (from the published `model_config.json` for
+    # `stabilityai/stable-audio-open-1.0`).
+    io_channels: int = 64
+    embed_dim: int = 1536
+    depth: int = 24
+    num_attention_heads: int = 24
+    cond_token_dim: int = 768
+    global_cond_dim: int = 1536
+    project_cond_tokens: bool = False
+    project_global_cond: bool = True
+    # Set to "ln" to wrap attention Q/K in LayerNorm (used by
+    # `stable-audio-open-small`; absent in the 1.0 base).
+    qk_norm: str | None = None
+
+    def __post_init__(self) -> None:
+        super().__post_init__()
+        self.hidden_size = self.embed_dim
+        self.in_channels = self.io_channels
+        self.out_channels = self.io_channels
+        self.num_channels_latents = self.io_channels
+        self.attention_head_dim = self.embed_dim // self.num_attention_heads
+
+
+@dataclass
+class StableAudioConfig(DiTConfig):
+    arch_config: DiTArchConfig = field(default_factory=StableAudioArchConfig)
+
+    prefix: str = "StableAudio"

fastvideo/configs/models/encoders/__init__.py

@@ -7,9 +7,12 @@ from fastvideo.configs.models.encoders.qwen2_5 import Qwen2_5_VLConfig
 from fastvideo.configs.models.encoders.siglip import SiglipVisionConfig
 from fastvideo.configs.models.encoders.reason1 import Reason1ArchConfig, Reason1Config
 from fastvideo.configs.models.encoders.gemma import LTX2GemmaConfig
+from fastvideo.configs.models.encoders.stable_audio_conditioner import (StableAudioConditionerArchConfig,
+                                                                        StableAudioConditionerConfig)
 
 __all__ = [
     "EncoderConfig", "TextEncoderConfig", "ImageEncoderConfig", "BaseEncoderOutput", "CLIPTextConfig",
     "CLIPVisionConfig", "WAN2_1ControlCLIPVisionConfig", "LlamaConfig", "T5Config", "T5LargeConfig", "Qwen2_5_VLConfig",
-    "Reason1ArchConfig", "Reason1Config", "LTX2GemmaConfig", "SiglipVisionConfig"
+    "Reason1ArchConfig", "Reason1Config", "LTX2GemmaConfig", "SiglipVisionConfig", "StableAudioConditionerArchConfig",
+    "StableAudioConditionerConfig"
 ]

fastvideo/configs/models/encoders/stable_audio_conditioner.py

@@ -0,0 +1,80 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Config for the Stable Audio Open 1.0 multi-conditioner.
+
+The conditioner bundles three sub-conditioners — a T5 text encoder
+(prompt) and two NumberConditioners (`seconds_start` / `seconds_total`)
+— into the (cross_attn_cond, cross_attn_mask, global_embed) triple the
+DiT consumes. The architecture is fully specified by the official
+`stable_audio_tools` `MultiConditioner` config; the constants here
+mirror that.
+"""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+
+from fastvideo.configs.models.base import ArchConfig
+from fastvideo.configs.models.encoders.base import (EncoderArchConfig, EncoderConfig)
+
+
+def _default_configs() -> list[dict]:
+    """Default = `stable-audio-open-1.0`'s three sub-conditioners."""
+    return [
+        {
+            "id": "prompt",
+            "type": "t5",
+            "config": {
+                "t5_model_name": "t5-base",
+                "max_length": 128
+            }
+        },
+        {
+            "id": "seconds_start",
+            "type": "number",
+            "config": {
+                "min_val": 0,
+                "max_val": 512
+            }
+        },
+        {
+            "id": "seconds_total",
+            "type": "number",
+            "config": {
+                "min_val": 0,
+                "max_val": 512
+            }
+        },
+    ]
+
+
+@dataclass
+class StableAudioConditionerArchConfig(EncoderArchConfig):
+    architectures: list[str] = field(default_factory=lambda: ["StableAudioMultiConditioner"])
+
+    # Shared embedding width across all sub-conditioners (T5 last-hidden
+    # dim and NumberEmbedder feature dim both = `cond_dim`).
+    cond_dim: int = 768
+
+    # Sub-conditioner identifiers. Order in `cross_attention_cond_ids`
+    # is the concat order for the cross-attn token sequence; order in
+    # `global_cond_ids` is the concat order for the global FiLM-style
+    # embedding.
+    cross_attention_cond_ids: tuple[str, ...] = ("prompt", "seconds_start", "seconds_total")
+    global_cond_ids: tuple[str, ...] = ("seconds_start", "seconds_total")
+
+    # Per-sub-conditioner spec list (mirrors upstream
+    # `model_config.json.model.conditioning.configs`). Each entry is
+    # `{"id": ..., "type": "t5"|"number", "config": {...}}`. The default
+    # matches `stable-audio-open-1.0`; SA-small overrides via the
+    # `conditioner/config.json` shipped in the converted repo.
+    configs: list = field(default_factory=_default_configs)
+
+    # Match official `stable_audio_tools/models/conditioners.py:334`:
+    # T5 is loaded directly in fp16.
+    t5_dtype: str = "float16"
+
+
+@dataclass
+class StableAudioConditionerConfig(EncoderConfig):
+    arch_config: ArchConfig = field(default_factory=StableAudioConditionerArchConfig)
+
+    prefix: str = "stable_audio_conditioner"

fastvideo/configs/models/encoders/t5.py

@@ -41,6 +41,14 @@ class T5ArchConfig(TextEncoderArchConfig):
     text_len: int = 512
     dtype: str | None = None
     gradient_checkpointing: bool = False
+    # Extra fields present in upstream HF T5Config but unused by FastVideo's
+    # encoder. Declared here so `update_model_arch` doesn't reject them when
+    # loading repos like `stabilityai/stable-audio-open-1.0` that ship the
+    # full HF config.
+    n_positions: int = 512
+    decoder_start_token_id: int = 0
+    output_past: bool = True
+    task_specific_params: dict | None = None
     stacked_params_mapping: list[tuple[str, str, str]] = field(default_factory=lambda: [
         # (param_name, shard_name, shard_id)
         (".qkv_proj", ".q", "q"),

fastvideo/configs/models/vaes/__init__.py

@@ -5,6 +5,7 @@ from fastvideo.configs.models.vaes.gen3cvae import Gen3CVAEConfig
 from fastvideo.configs.models.vaes.hunyuanvae import HunyuanVAEConfig
 from fastvideo.configs.models.vaes.hunyuan15vae import Hunyuan15VAEConfig
 from fastvideo.configs.models.vaes.ltx2vae import LTX2VAEConfig
+from fastvideo.configs.models.vaes.oobleck import OobleckVAEArchConfig, OobleckVAEConfig
 from fastvideo.configs.models.vaes.wanvae import WanVAEConfig
 
 __all__ = [
@@ -16,4 +17,6 @@ __all__ = [
     "Gen3CVAEConfig",
     "Hunyuan15VAEConfig",
     "LTX2VAEConfig",
+    "OobleckVAEArchConfig",
+    "OobleckVAEConfig",
 ]

fastvideo/configs/models/vaes/oobleck.py

@@ -0,0 +1,68 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Config for the Stable Audio Open 1.0 "Oobleck" VAE.
+
+Mirrors the per-channel `vae/config.json` shipped in
+`stabilityai/stable-audio-open-1.0` 1:1 (see
+`fastvideo/models/vaes/oobleck.py::OobleckVAE.from_pretrained`, which
+constructs the VAE from these fields). Inherits the FastVideo VAEConfig
+base so the standard `load_encoder` / `load_decoder` flags + tiling
+knobs apply.
+
+Naming: the VAE architecture is officially "Oobleck" (per Stability
+AI's stable-audio-tools) — the surrounding model family is "Stable
+Audio Open 1.0". This config is named after the architecture
+(`OobleckVAEConfig`) since the same VAE is shared across Stable Audio
+checkpoints; downstream pipelines reference it by its arch name, not
+by a host-pipeline name.
+"""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+
+from fastvideo.configs.models.vaes.base import VAEArchConfig, VAEConfig
+
+
+@dataclass
+class OobleckVAEArchConfig(VAEArchConfig):
+    """Stable Audio Open 1.0 VAE architecture constants."""
+
+    architectures: list[str] = field(default_factory=lambda: ["AutoencoderOobleck"])
+
+    # From stabilityai/stable-audio-open-1.0/vae/config.json.
+    encoder_hidden_size: int = 128
+    downsampling_ratios: list[int] = field(default_factory=lambda: [2, 4, 4, 8, 8])
+    channel_multiples: list[int] = field(default_factory=lambda: [1, 2, 4, 8, 16])
+    decoder_channels: int = 128
+    decoder_input_channels: int = 64
+    audio_channels: int = 2  # stereo
+    sampling_rate: int = 44100
+
+
+@dataclass
+class OobleckVAEConfig(VAEConfig):
+    """FastVideo VAE config wrapping the Oobleck arch.
+
+    Audio VAEs don't use the temporal/spatial tiling defaults that the
+    base VAEConfig is shaped for (those exist for video VAEs); they are
+    retained but irrelevant for audio.
+    """
+
+    arch_config: VAEArchConfig = field(default_factory=OobleckVAEArchConfig)
+
+    # Audio is 1-D, so the video-VAE tiling defaults are inert. Disable
+    # them so callers don't accidentally trip on tile-stride math built
+    # for spatial tensors.
+    use_tiling: bool = False
+    use_temporal_tiling: bool = False
+    use_parallel_tiling: bool = False
+
+    # Where the FastVideo loader / pipeline-glue wrapper should fetch
+    # weights from when no local path is supplied. Gated repo — caller's
+    # HF token must have accepted terms on
+    # https://huggingface.co/stabilityai/stable-audio-open-1.0.
+    pretrained_path: str = "stabilityai/stable-audio-open-1.0"
+    pretrained_subfolder: str = "vae"
+    # Match official `stable_audio_tools`: VAE runs in fp16 (the
+    # `pretransform.model_half` path in
+    # `stable_audio_tools/models/pretransforms.py`).
+    pretrained_dtype: str = "float16"

fastvideo/configs/pipelines/stable_audio.py

@@ -0,0 +1,67 @@
+# SPDX-License-Identifier: Apache-2.0
+"""`PipelineConfig` for Stable Audio Open 1.0."""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+
+from fastvideo.configs.models import DiTConfig, VAEConfig
+from fastvideo.configs.models.dits import StableAudioConfig
+from fastvideo.configs.models.vaes import OobleckVAEConfig
+from fastvideo.configs.pipelines.base import PipelineConfig
+
+
+@dataclass
+class StableAudioT2AConfig(PipelineConfig):
+    """Stable Audio Open 1.0 pipeline config."""
+
+    dit_config: DiTConfig = field(default_factory=StableAudioConfig)
+    # Standard `TransformerLoader` reads `dit_precision`; default in
+    # `PipelineConfig` is bf16, but we want fp16 to match official.
+    dit_precision: str = "fp16"
+
+    vae_config: VAEConfig = field(default_factory=OobleckVAEConfig)
+    vae_tiling: bool = False
+    vae_sp: bool = False
+
+    # `StableAudioMultiConditioner` owns its own T5; zero out the
+    # parent's text-encoder slots so the length-equality validator passes.
+    text_encoder_configs: tuple = field(default_factory=tuple)
+    preprocess_text_funcs: tuple = field(default_factory=tuple)
+    postprocess_text_funcs: tuple = field(default_factory=tuple)
+
+    num_inference_steps: int = 100
+    guidance_scale: float = 7.0
+    audio_end_in_s: float = 10.0  # short-clip default
+    audio_start_in_s: float = 0.0
+    sampling_rate: int = 44100
+    audio_channels: int = 2
+    # Stable Audio Open 1.0 was trained at a fixed 2,097,152-sample
+    # window (= 2097152 / 44100 ≈ 47.55s). Anything past this is
+    # silently truncated by the post-decode slice — validate up-front.
+    sample_size: int = 2097152
+    max_audio_duration_s: float = 2097152 / 44100
+
+    # Match the official `stable_audio_tools` defaults (`model_half=True`
+    # in `run_gradio.py`), which loads the DiT, VAE, and T5 in fp16 and
+    # wraps T5 forward in `autocast(fp16)`. fp16 is also a hard
+    # requirement for FlashAttention-2 / FA-3.
+    precision: str = "fp16"
+    vae_precision: str = "fp16"
+    text_encoder_precisions: tuple[str, ...] = field(default_factory=tuple)
+
+    def __post_init__(self) -> None:
+        # A2A needs encode; load both halves for either path.
+        self.vae_config.load_encoder = True
+        self.vae_config.load_decoder = True
+
+
+@dataclass
+class StableAudioOpenSmallConfig(StableAudioT2AConfig):
+    """`stable-audio-open-small` overrides: shorter training window
+    (524288 samples ≈ 11.89s @ 44.1 kHz) and a faster default sampler
+    config carried by the small preset.
+    """
+
+    sample_size: int = 524288
+    max_audio_duration_s: float = 524288 / 44100
+    audio_end_in_s: float = 6.0  # short-clip default suitable for the small window

fastvideo/entrypoints/cli/eval.py

@@ -1,216 +0,0 @@
-"""``fastvideo eval`` CLI: list registered eval metrics and run them
-against a set of videos.
-
-This is a thin wrapper around :mod:`fastvideo.eval`. Heavy lifting
-(metric loading, GPU handling, batching) lives in
-:func:`fastvideo.eval.create_evaluator`.
-
-Examples::
-
-    fastvideo eval list
-    fastvideo eval list --group vbench
-    fastvideo eval run --videos path/to/videos/*.mp4 \\
-        --metrics common.ssim --reference path/to/refs/
-    fastvideo eval run --videos clip.mp4 --metrics vbench.aesthetic_quality \\
-        --output scores.json
-"""
-
-from __future__ import annotations
-
-import argparse
-import glob
-import json
-from pathlib import Path
-from typing import cast
-
-from fastvideo.entrypoints.cli.cli_types import CLISubcommand
-from fastvideo.logger import init_logger
-from fastvideo.utils import FlexibleArgumentParser
-
-logger = init_logger(__name__)
-
-
-class EvalSubcommand(CLISubcommand):
-    """The ``eval`` subcommand — entry point for the eval suite."""
-
-    def __init__(self) -> None:
-        self.name = "eval"
-        super().__init__()
-
-    def cmd(self, args: argparse.Namespace) -> None:
-        action = getattr(args, "eval_action", None)
-        if action == "list":
-            _cmd_list(args)
-        elif action == "run":
-            _cmd_run(args)
-        else:
-            # Re-print help if no action was given.
-            self._parser.print_help()  # type: ignore[attr-defined]
-
-    def validate(self, args: argparse.Namespace) -> None:
-        action = getattr(args, "eval_action", None)
-        if action == "run" and not args.videos:
-            raise SystemExit("`fastvideo eval run` requires --videos")
-
-    def subparser_init(self, subparsers: argparse._SubParsersAction) -> FlexibleArgumentParser:
-        eval_parser = subparsers.add_parser(
-            "eval",
-            help="Run video-gen evaluation metrics",
-            usage="fastvideo eval {list,run} [...]",
-        )
-        sub = eval_parser.add_subparsers(dest="eval_action", required=False)
-
-        # `eval list`
-        list_p = sub.add_parser("list", help="List registered metrics")
-        list_p.add_argument("--group", type=str, default=None, help="Filter to a metric group (e.g. 'vbench').")
-
-        # `eval run`
-        run_p = sub.add_parser("run", help="Evaluate videos against one or more metrics")
-        run_p.add_argument("--videos",
-                           type=str,
-                           nargs="+",
-                           required=False,
-                           help="Path, glob, or directory of generated videos.")
-        run_p.add_argument("--reference",
-                           type=str,
-                           default=None,
-                           help="Path / glob / dir of reference videos (for paired metrics).")
-        run_p.add_argument("--metrics", type=str, default="all", help="Comma-separated metric names, or 'all'.")
-        run_p.add_argument("--device", type=str, default="cuda", help="Torch device (e.g. 'cuda', 'cuda:0', 'cpu').")
-        run_p.add_argument("--text-prompt",
-                           type=str,
-                           nargs="*",
-                           default=None,
-                           help="Prompt(s) for text-conditioned metrics. One per video.")
-        run_p.add_argument("--fps", type=float, default=None, help="Frame-rate annotation passed to fps-aware metrics.")
-        run_p.add_argument("--output",
-                           type=str,
-                           default=None,
-                           help="Write results as JSON to this path (default: stdout).")
-
-        # Stash the parser so cmd() can re-print help on no-action.
-        self._parser = eval_parser  # type: ignore[attr-defined]
-        return cast(FlexibleArgumentParser, eval_parser)
-
-
-def _cmd_list(args: argparse.Namespace) -> None:
-    from fastvideo.eval import list_metrics
-    names = list_metrics()
-    if args.group:
-        prefix = args.group.rstrip(".") + "."
-        names = [n for n in names if n == args.group or n.startswith(prefix)]
-    if not names:
-        print(f"(no metrics matched group {args.group!r})")
-        return
-    for name in names:
-        print(name)
-
-
-def _cmd_run(args: argparse.Namespace) -> None:
-    from fastvideo.eval import create_evaluator
-    from fastvideo.eval.io import load_video
-
-    video_paths = _expand_paths(args.videos)
-    if not video_paths:
-        raise SystemExit(f"No videos matched: {args.videos}")
-    ref_paths = _expand_paths([args.reference]) if args.reference else None
-
-    metrics_arg: list[str] | str = ("all" if args.metrics == "all" else
-                                    [m.strip() for m in args.metrics.split(",") if m.strip()])
-
-    evaluator = create_evaluator(metrics=metrics_arg, device=args.device)
-
-    all_results: list[dict] = []
-    for i, vp in enumerate(video_paths):
-        logger.info("Evaluating %s (%d/%d)", vp, i + 1, len(video_paths))
-        kwargs: dict = {"video": load_video(vp)}
-        if ref_paths is not None:
-            ref = ref_paths[i] if i < len(ref_paths) else ref_paths[0]
-            kwargs["reference"] = load_video(ref)
-        if args.text_prompt is not None:
-            prompt = (args.text_prompt[i] if i < len(args.text_prompt) else args.text_prompt[0])
-            kwargs["text_prompt"] = [prompt]
-        if args.fps is not None:
-            kwargs["fps"] = args.fps
-
-        results = evaluator.evaluate(**kwargs)
-        all_results.append({
-            "video": str(vp),
-            "scores": _serialize_results(results),
-        })
-
-    payload = json.dumps(all_results, indent=2, default=_jsonable)
-    if args.output:
-        Path(args.output).write_text(payload)
-        logger.info("Wrote results to %s", args.output)
-    else:
-        print(payload)
-
-
-def _expand_paths(patterns: list[str]) -> list[str]:
-    out: list[str] = []
-    for pat in patterns:
-        p = Path(pat)
-        if p.is_dir():
-            for ext in (".mp4", ".avi", ".mov", ".mkv", ".gif"):
-                out.extend(sorted(str(f) for f in p.iterdir() if f.suffix.lower() == ext))
-        elif any(c in pat for c in "*?["):
-            out.extend(sorted(glob.glob(pat)))
-        else:
-            out.append(pat)
-    # de-dup, preserve order
-    seen: set[str] = set()
-    deduped: list[str] = []
-    for x in out:
-        if x not in seen:
-            seen.add(x)
-            deduped.append(x)
-    return deduped
-
-
-def _serialize_results(results) -> dict | list:
-    """Turn evaluator output (dict or list-of-dicts) into JSON-friendly form."""
-    if isinstance(results, list):
-        return [_serialize_results(r) for r in results]
-    if isinstance(results, dict):
-        return {k: _serialize_metric_result(v) for k, v in results.items()}
-    return _serialize_metric_result(results)
-
-
-def _serialize_metric_result(mr) -> dict:
-    return {
-        "name": getattr(mr, "name", None),
-        "score": getattr(mr, "score", None),
-        "details": getattr(mr, "details", None),
-    }
-
-
-def _jsonable(obj):
-    """``json.dumps(default=...)`` coercer for metric outputs.
-
-    Metrics frequently land numpy scalars / arrays, torch tensors, and
-    pathlib paths inside ``MetricResult.details`` (e.g. ``optical_flow``
-    populates ``per_frame_metrics`` with numpy floats). The stdlib JSON
-    encoder rejects all of those by default — this callback walks the
-    leaves and coerces them to native Python types.
-    """
-    import numpy as np
-    import torch
-
-    if isinstance(obj, np.integer):
-        return int(obj)
-    if isinstance(obj, np.floating):
-        return float(obj)
-    if isinstance(obj, np.bool_):
-        return bool(obj)
-    if isinstance(obj, np.ndarray):
-        return obj.tolist()
-    if isinstance(obj, torch.Tensor):
-        return obj.detach().cpu().tolist()
-    if isinstance(obj, Path):
-        return str(obj)
-    raise TypeError(f"Object of type {type(obj).__name__} is not JSON serializable")
-
-
-def cmd_init() -> list[CLISubcommand]:
-    return [EvalSubcommand()]

fastvideo/entrypoints/cli/main.py

@@ -3,9 +3,10 @@
 from fastvideo.entrypoints.cli.cli_types import CLISubcommand
 from fastvideo.entrypoints.cli.generate import cmd_init as generate_cmd_init
 from fastvideo.utils import FlexibleArgumentParser
+from fastvideo.entrypoints.cli.router_serve import (
+    cmd_init as router_serve_cmd_init, )
 from fastvideo.entrypoints.cli.serve import cmd_init as serve_cmd_init
 from fastvideo.entrypoints.cli.bench import cmd_init as bench_cmd_init
-from fastvideo.entrypoints.cli.eval import cmd_init as eval_cmd_init
 
 
 def cmd_init() -> list[CLISubcommand]:
@@ -13,8 +14,8 @@ def cmd_init() -> list[CLISubcommand]:
     commands = []
     commands.extend(generate_cmd_init())
     commands.extend(serve_cmd_init())
+    commands.extend(router_serve_cmd_init())
     commands.extend(bench_cmd_init())
-    commands.extend(eval_cmd_init())
     return commands
 
 

fastvideo/entrypoints/cli/router_serve.py

@@ -0,0 +1,115 @@
+# SPDX-License-Identifier: Apache-2.0
+"""``fastvideo router-serve`` CLI subcommand.
+
+Launches the streaming router from a YAML config. Separate from
+``fastvideo serve`` because the router is an orthogonal process: it
+fronts one or more running servers rather than hosting a generator
+itself.
+"""
+from __future__ import annotations
+
+import argparse
+import os
+from typing import cast
+
+from fastvideo.api.parser import load_raw_config
+from fastvideo.entrypoints.cli.cli_types import CLISubcommand
+from fastvideo.entrypoints.streaming.router.config import (
+    ReplicaEndpoint,
+    RouterConfig,
+)
+from fastvideo.logger import init_logger
+from fastvideo.utils import FlexibleArgumentParser
+
+logger = init_logger(__name__)
+
+
+class RouterServeSubcommand(CLISubcommand):
+    """Start the multi-replica WebSocket router."""
+
+    def __init__(self) -> None:
+        self.name = "router-serve"
+        super().__init__()
+
+    def cmd(self, args: argparse.Namespace) -> None:
+        config = _load_router_config(args.config)
+        logger.info(
+            "router listening on %s:%d (%d replicas, %d primary)",
+            config.host,
+            config.port,
+            len(config.replicas),
+            sum(1 for r in config.replicas if r.primary),
+        )
+        from fastvideo.entrypoints.streaming.router.main import run_router
+
+        run_router(config)
+
+    def validate(self, args: argparse.Namespace) -> None:
+        if not args.config:
+            raise ValueError("fastvideo router-serve requires --config PATH")
+        if not os.path.exists(args.config):
+            raise ValueError(f"Router config file not found: {args.config}")
+
+    def subparser_init(
+        self,
+        subparsers: argparse._SubParsersAction,
+    ) -> FlexibleArgumentParser:
+        parser = subparsers.add_parser(
+            "router-serve",
+            help="Start the streaming router (multi-replica load balancer)",
+            usage="fastvideo router-serve --config ROUTER_CONFIG",
+        )
+        parser.add_argument(
+            "--config",
+            type=str,
+            default="",
+            required=False,
+            help="Path to a YAML/JSON router config. Required.",
+        )
+        return cast(FlexibleArgumentParser, parser)
+
+
+def _load_router_config(path: str) -> RouterConfig:
+    raw = load_raw_config(path)
+    router_raw = raw.get("router") if isinstance(raw, dict) else None
+    if not isinstance(router_raw, dict):
+        raise ValueError(f"Router config {path!r} must have a top-level `router:` block")
+
+    replicas_raw = router_raw.get("replicas", [])
+    if not isinstance(replicas_raw, list):
+        raise ValueError(f"router.replicas must be a list, got {type(replicas_raw).__name__}")
+    replicas = []
+    for i, r in enumerate(replicas_raw):
+        if not isinstance(r, dict):
+            raise ValueError(f"router.replicas[{i}] must be a mapping, got {type(r).__name__}")
+        url = r.get("url")
+        if not url:
+            raise ValueError(f"router.replicas[{i}] is missing required key 'url'")
+        replicas.append(
+            ReplicaEndpoint(
+                url=url,
+                name=r.get("name"),
+                primary=bool(r.get("primary", False)),
+                weight=float(r.get("weight", 1.0)),
+            ))
+    if not replicas:
+        raise ValueError("Router config must list at least one replica under `router.replicas`")
+
+    health_check = router_raw.get("health_check") or {}
+    return RouterConfig(
+        host=str(router_raw.get("host", "0.0.0.0")),
+        port=int(router_raw.get("port", 9000)),
+        replicas=replicas,
+        health_check_path=str(health_check.get("path", "/health")),
+        health_check_interval_seconds=float(health_check.get("interval_seconds", 5.0)),
+        health_check_timeout_seconds=float(health_check.get("timeout_seconds", 2.0)),
+        failure_threshold=int(health_check.get("failure_threshold", 3)),
+        recovery_threshold=int(health_check.get("recovery_threshold", 2)),
+    )
+
+
+def cmd_init() -> list[CLISubcommand]:
+    return [RouterServeSubcommand()]
+
+
+__all__ = ["RouterServeSubcommand", "cmd_init"]

fastvideo/entrypoints/streaming/__init__.py

@@ -11,6 +11,28 @@ from fastvideo.entrypoints.streaming.session_store import (
     InMemorySessionStore,
     SessionStore,
 )
+from fastvideo.entrypoints.streaming.gpu_pool import (
+    GpuPool,
+    InProcessGpuPool,
+    PoolAcquireTimeout,
+    SubprocessGpuPool,
+)
+from fastvideo.entrypoints.streaming.mock_server import (
+    MockGenerator,
+    build_mock_app,
+)
+from fastvideo.entrypoints.streaming.prompt import (
+    LLMProvider,
+    PromptEnhancer,
+)
+from fastvideo.entrypoints.streaming.prompt.safety import (
+    PromptSafetyFilter,
+    SafetyDecision,
+)
+from fastvideo.entrypoints.streaming.session_logger import (
+    SessionLogEvent,
+    SessionLogger,
+)
 from fastvideo.entrypoints.streaming.stream import (
     FragmentedMP4Chunk,
     FragmentedMP4Encoder,
@@ -20,12 +42,24 @@ __all__ = [
     "BlobStore",
     "FragmentedMP4Chunk",
     "FragmentedMP4Encoder",
+    "GpuPool",
     "InMemoryBlobStore",
     "InMemorySessionStore",
+    "InProcessGpuPool",
+    "LLMProvider",
+    "MockGenerator",
+    "PoolAcquireTimeout",
+    "PromptEnhancer",
+    "PromptSafetyFilter",
+    "SafetyDecision",
+    "SessionLogEvent",
+    "SessionLogger",
+    "build_mock_app",
     "Session",
     "SessionManager",
     "SessionState",
     "SessionStore",
+    "SubprocessGpuPool",
     "build_app",
     "run_server",
 ]

fastvideo/entrypoints/streaming/gpu_pool.py

@@ -0,0 +1,542 @@
+# SPDX-License-Identifier: Apache-2.0
+"""GPU pool manager for the streaming server.
+
+Replaces the single-generator path in PR 7.5 with a typed pool
+abstraction. Three implementations ship here:
+
+* :class:`InProcessGpuPool` — one in-process ``VideoGenerator``; used
+  by tests and single-GPU dev deployments.
+* :class:`SubprocessGpuPool` — one ``multiprocessing.Process`` per
+  GPU, each running :func:`worker_main` against a ``GeneratorConfig``.
+  Jobs are dispatched via ``multiprocessing.Queue``.
+* :class:`GpuPool` (abstract) — the interface both use.
+
+Session-to-GPU binding lives in the pool so continuation state stays
+on the GPU that generated the previous segment (matching the internal
+``gpu_pool.py``'s per-GPU cache behavior). Cross-GPU handoff is
+supported via :class:`SessionStore` snapshot + hydrate, which
+serializes the state before the migration and rehydrates it on the
+new worker.
+
+Typed config: workers start from a :class:`GeneratorConfig` (no flat
+LTX-2 kwargs), satisfying the PR 6 + PR 7 contracts that the public
+surface doesn't reintroduce the legacy kwarg bag.
+"""
+from __future__ import annotations
+
+import asyncio
+import multiprocessing as mp
+import queue
+import threading
+import time
+import uuid
+from abc import ABC, abstractmethod
+from concurrent.futures import Future
+from dataclasses import dataclass, field
+from typing import Any, Protocol
+
+from fastvideo.api.schema import (
+    GeneratorConfig,
+    GenerationRequest,
+    GpuPoolConfig,
+    WarmupConfig,
+)
+from fastvideo.entrypoints.streaming.session_store import (
+    InMemorySessionStore,
+    SessionStore,
+)
+from fastvideo.entrypoints.streaming.worker import worker_main
+from fastvideo.logger import init_logger
+
+logger = init_logger(__name__)
+
+# ---------------------------------------------------------------------------
+# Public interface
+# ---------------------------------------------------------------------------
+
+
+class _GeneratorLike(Protocol):
+    """Subset the pool calls on a worker-side generator."""
+
+    def generate(self, request: GenerationRequest) -> Any:
+        ...
+
+
+@dataclass
+class PoolAssignment:
+    """The worker a session is currently bound to."""
+
+    gpu_id: int
+    worker_id: str
+    pinned_at: float = field(default_factory=time.monotonic)
+
+
+class GpuPool(ABC):
+    """Abstract GPU pool.
+
+    ``acquire`` binds a session to a worker and holds that binding
+    across segments so continuation state can stay hot. ``run`` submits
+    a single ``GenerationRequest`` for a bound session.
+
+    Acquire / release are independent of run — a session can run many
+    segments on one acquired worker, and must release on disconnect.
+    """
+
+    @abstractmethod
+    async def acquire(
+        self,
+        session_id: str,
+        *,
+        timeout: float | None = None,
+    ) -> PoolAssignment:
+        ...
+
+    @abstractmethod
+    async def run(
+        self,
+        session_id: str,
+        request: GenerationRequest,
+    ) -> Any:
+        ...
+
+    @abstractmethod
+    async def release(self, session_id: str) -> None:
+        ...
+
+    @abstractmethod
+    async def shutdown(self) -> None:
+        ...
+
+    @abstractmethod
+    def health(self) -> PoolHealth:
+        ...
+
+
+@dataclass
+class PoolHealth:
+    total_workers: int
+    available_workers: int
+    active_sessions: int
+    queued_sessions: int = 0
+
+
+class PoolAcquireTimeout(RuntimeError):
+    """Raised when ``acquire`` times out waiting for a free worker."""
+
+
+# ---------------------------------------------------------------------------
+# In-process implementation (single-worker, test / dev)
+# ---------------------------------------------------------------------------
+
+
+class InProcessGpuPool(GpuPool):
+    """Single-process pool backed by one :class:`_GeneratorLike`.
+
+    This is what PR 7.5's server uses by default; PR 7.6 adds the real
+    ``SubprocessGpuPool`` alternative but keeps this one for tests and
+    small deployments.
+    """
+
+    def __init__(
+        self,
+        generator: _GeneratorLike,
+        *,
+        gpu_id: int = 0,
+        session_store: SessionStore | None = None,
+    ) -> None:
+        self._generator = generator
+        self._gpu_id = gpu_id
+        self._worker_id = f"inproc-{uuid.uuid4().hex[:6]}"
+        self._session_store = session_store or InMemorySessionStore()
+        self._active: dict[str, PoolAssignment] = {}
+        self._lock = asyncio.Lock()
+        self._gen_lock = asyncio.Lock()
+
+    async def acquire(
+        self,
+        session_id: str,
+        *,
+        timeout: float | None = None,
+    ) -> PoolAssignment:
+        async with self._lock:
+            existing = self._active.get(session_id)
+            if existing is not None:
+                return existing
+            assignment = PoolAssignment(gpu_id=self._gpu_id, worker_id=self._worker_id)
+            self._active[session_id] = assignment
+            return assignment
+
+    async def run(
+        self,
+        session_id: str,
+        request: GenerationRequest,
+    ) -> Any:
+        if session_id not in self._active:
+            raise RuntimeError(f"session {session_id!r} is not acquired on this pool")
+        # Serialize generator access so one GPU runs one request at a
+        # time, matching the internal gpu_pool's per-GPU lock.
+        async with self._gen_lock:
+            loop = asyncio.get_running_loop()
+            return await loop.run_in_executor(None, self._generator.generate, request)
+
+    async def release(self, session_id: str) -> None:
+        async with self._lock:
+            self._active.pop(session_id, None)
+
+    async def shutdown(self) -> None:
+        self._active.clear()
+
+    def health(self) -> PoolHealth:
+        return PoolHealth(
+            total_workers=1,
+            available_workers=1 if not self._active else 0,
+            active_sessions=len(self._active),
+        )
+
+
+# ---------------------------------------------------------------------------
+# Subprocess implementation (multi-worker, real deployment)
+# ---------------------------------------------------------------------------
+
+
+@dataclass
+class _WorkerHandle:
+    process: Any  # mp.Process or compatible handle with is_alive / join / kill
+    job_queue: mp.Queue
+    result_queue: mp.Queue
+    gpu_id: int
+    worker_id: str
+    ready: threading.Event
+    # ``ready`` flips on either successful boot or boot failure so the
+    # parent stops waiting; ``boot_ok`` is set only on a real ready
+    # acknowledgement and is what gates pool admission.
+    boot_ok: threading.Event
+    shutdown_event: Any  # mp.Event is a factory, not a type — Any keeps mypy sane
+
+
+@dataclass
+class _PendingJob:
+    job_id: str
+    future: Future
+    session_id: str
+    worker_id: str
+
+
+class SubprocessGpuPool(GpuPool):
+    """One ``multiprocessing.Process`` per GPU.
+
+    Each worker boots :class:`fastvideo.VideoGenerator` from a typed
+    :class:`GeneratorConfig` inside the child process (post-
+    ``CUDA_VISIBLE_DEVICES`` setup) and consumes jobs from an mp Queue.
+
+    This is the production shape: the parent process stays CPU-only, and
+    GPU state never crosses process boundaries. Continuation state is
+    serialized through :class:`SessionStore` for cross-GPU handoff.
+
+    PR 7.6 ships this as an opt-in; PR 7.5's in-process pool remains the
+    default until nightly runs validate the subprocess path.
+    """
+
+    def __init__(
+        self,
+        generator_config: GeneratorConfig,
+        *,
+        pool_config: GpuPoolConfig,
+        warmup_config: WarmupConfig | None = None,
+        session_store: SessionStore | None = None,
+        worker_factory: WorkerFactory | None = None,
+    ) -> None:
+        self._generator_config = generator_config
+        self._pool_config = pool_config
+        self._warmup_config = warmup_config or WarmupConfig()
+        self._session_store = session_store or InMemorySessionStore()
+        self._worker_factory = worker_factory or _default_worker_factory
+        self._workers: list[_WorkerHandle] = []
+        self._available: asyncio.Queue[int] = asyncio.Queue()
+        self._assignments: dict[str, PoolAssignment] = {}
+        self._worker_by_id: dict[str, _WorkerHandle] = {}
+        self._pending: dict[str, _PendingJob] = {}
+        self._lock = asyncio.Lock()
+        self._result_reader_tasks: list[asyncio.Task] = []
+
+    async def start(self) -> None:
+        """Spawn worker processes and wait for each to report ready."""
+        num_workers = self._pool_config.num_workers or 1
+        for gpu_id in range(num_workers):
+            handle = self._worker_factory(
+                gpu_id=gpu_id,
+                generator_config=self._generator_config,
+                warmup_config=self._warmup_config,
+            )
+            self._workers.append(handle)
+            self._worker_by_id[handle.worker_id] = handle
+
+        # Wait for each worker's ready event in a thread to avoid
+        # blocking the event loop.
+        loop = asyncio.get_running_loop()
+        await asyncio.gather(*[
+            loop.run_in_executor(None, handle.ready.wait, self._warmup_config.timeout_seconds)
+            for handle in self._workers
+        ])
+
+        # Start background result readers — one task per worker
+        # drains its result queue and resolves futures in _pending.
+        for handle in self._workers:
+            task = asyncio.create_task(self._drain_results(handle))
+            self._result_reader_tasks.append(task)
+
+        # Only admit workers that successfully booted. Anything that
+        # failed boot (timeout, crash, error sentinel) stays out of the
+        # available queue so we never assign a session to it.
+        for idx, handle in enumerate(self._workers):
+            if handle.boot_ok.is_set():
+                await self._available.put(idx)
+            else:
+                logger.error(
+                    "pool: worker %s failed to boot; skipping",
+                    handle.worker_id,
+                )
+
+    async def acquire(
+        self,
+        session_id: str,
+        *,
+        timeout: float | None = None,
+    ) -> PoolAssignment:
+        async with self._lock:
+            existing = self._assignments.get(session_id)
+            if existing is not None:
+                return existing
+        try:
+            idx = await asyncio.wait_for(self._available.get(), timeout=timeout)
+        except asyncio.TimeoutError as exc:
+            raise PoolAcquireTimeout(f"no worker available after {timeout}s") from exc
+        handle = self._workers[idx]
+        assignment = PoolAssignment(gpu_id=handle.gpu_id, worker_id=handle.worker_id)
+        async with self._lock:
+            self._assignments[session_id] = assignment
+        return assignment
+
+    async def run(
+        self,
+        session_id: str,
+        request: GenerationRequest,
+    ) -> Any:
+        assignment = self._assignments.get(session_id)
+        if assignment is None:
+            raise RuntimeError(f"session {session_id!r} not acquired on this pool")
+        handle = self._worker_by_id[assignment.worker_id]
+        job_id = uuid.uuid4().hex
+        future: Future = Future()
+        self._pending[job_id] = _PendingJob(
+            job_id=job_id,
+            future=future,
+            session_id=session_id,
+            worker_id=handle.worker_id,
+        )
+        # mp.Queue.put can block if the underlying pipe buffer is full;
+        # offload to a thread so the event loop keeps serving other
+        # sessions. If the put itself fails, drop the pending entry so
+        # _drain_results doesn't dangle a future forever.
+        loop = asyncio.get_running_loop()
+        try:
+            await loop.run_in_executor(
+                None,
+                handle.job_queue.put,
+                {
+                    "job_id": job_id,
+                    "request": request
+                },
+            )
+        except Exception:
+            self._pending.pop(job_id, None)
+            raise
+        return await asyncio.wrap_future(future)
+
+    async def release(self, session_id: str) -> None:
+        async with self._lock:
+            assignment = self._assignments.pop(session_id, None)
+        if assignment is None:
+            return
+        idx = next((i for i, h in enumerate(self._workers) if h.worker_id == assignment.worker_id), None)
+        if idx is None:
+            return
+        # Don't return a dead worker to the pool; otherwise the next
+        # acquire will hand a session to a process that can't run jobs.
+        if not self._workers[idx].process.is_alive():
+            logger.warning(
+                "pool: worker %s died; not returning to available queue",
+                self._workers[idx].worker_id,
+            )
+            return
+        await self._available.put(idx)
+
+    async def shutdown(self) -> None:
+        loop = asyncio.get_running_loop()
+
+        # Signal all workers in parallel; .put may block on a full pipe,
+        # so off-load it the same way run() does.
+        async def _signal(handle: _WorkerHandle) -> None:
+            try:
+                handle.shutdown_event.set()
+                await loop.run_in_executor(None, handle.job_queue.put, None)
+            except Exception:  # pragma: no cover - best-effort cleanup
+                pass
+
+        await asyncio.gather(*(_signal(h) for h in self._workers))
+        # Join in parallel so total shutdown is bounded by the slowest
+        # worker, not the sum of all timeouts.
+        await asyncio.gather(*(loop.run_in_executor(None, handle.process.join, 5.0) for handle in self._workers))
+        for handle in self._workers:
+            if handle.process.is_alive():
+                handle.process.kill()
+        for task in self._result_reader_tasks:
+            task.cancel()
+        self._result_reader_tasks.clear()
+        self._workers.clear()
+        self._worker_by_id.clear()
+
+    def health(self) -> PoolHealth:
+        return PoolHealth(
+            total_workers=len(self._workers),
+            available_workers=self._available.qsize(),
+            active_sessions=len(self._assignments),
+        )
+
+    async def _drain_results(self, handle: _WorkerHandle) -> None:
+        loop = asyncio.get_running_loop()
+        try:
+            while not handle.shutdown_event.is_set():
+                try:
+                    msg = await loop.run_in_executor(None, _safe_queue_get, handle.result_queue, 0.5)
+                except Exception:
+                    logger.exception("pool: worker %s result reader failed", handle.worker_id)
+                    return
+                if msg is None:
+                    continue
+                job_id = msg.get("job_id")
+                if job_id is None:
+                    continue
+                pending = self._pending.pop(job_id, None)
+                if pending is None:
+                    continue
+                if msg.get("kind") == "error":
+                    pending.future.set_exception(RuntimeError(msg["error"]))
+                else:
+                    pending.future.set_result(msg.get("result"))
+        finally:
+            # If we exit for any reason — shutdown, exception, cancel —
+            # surface that to any in-flight jobs on this worker so their
+            # await never hangs on a future no one will resolve.
+            for jid in [jid for jid, job in self._pending.items() if job.worker_id == handle.worker_id]:
+                pending = self._pending.pop(jid, None)
+                if pending is not None and not pending.future.done():
+                    pending.future.set_exception(
+                        RuntimeError(f"worker {handle.worker_id} result reader exited "
+                                     "with pending jobs"))
+
+
+def _safe_queue_get(q: mp.Queue, timeout: float) -> Any | None:
+    try:
+        return q.get(timeout=timeout)
+    except queue.Empty:
+        return None
+
+
+# ---------------------------------------------------------------------------
+# Worker process
+# ---------------------------------------------------------------------------
+
+
+class WorkerFactory(Protocol):
+
+    def __call__(
+        self,
+        *,
+        gpu_id: int,
+        generator_config: GeneratorConfig,
+        warmup_config: WarmupConfig,
+    ) -> _WorkerHandle:
+        ...
+
+
+def _default_worker_factory(
+    *,
+    gpu_id: int,
+    generator_config: GeneratorConfig,
+    warmup_config: WarmupConfig,
+) -> _WorkerHandle:
+    """Spawn a real multiprocessing worker.
+
+    The child process calls :func:`worker_main` which constructs a
+    :class:`VideoGenerator` from ``generator_config`` and runs a
+    blocking job loop. The ``ready`` event flips after the warmup
+    request completes.
+    """
+    ctx = mp.get_context("spawn")
+    job_queue: mp.Queue = ctx.Queue()
+    result_queue: mp.Queue = ctx.Queue()
+    ready = threading.Event()
+    boot_ok = threading.Event()
+    shutdown_event = ctx.Event()
+    worker_id = f"gpu{gpu_id}-{uuid.uuid4().hex[:6]}"
+    process = ctx.Process(
+        target=worker_main,
+        kwargs={
+            "gpu_id": gpu_id,
+            "worker_id": worker_id,
+            "generator_config": generator_config,
+            "warmup_config": warmup_config,
+            "job_queue": job_queue,
+            "result_queue": result_queue,
+            "shutdown_event": shutdown_event,
+        },
+        daemon=False,
+    )
+    process.start()
+
+    # Block the parent-side ``ready`` flag until the worker posts a
+    # ready acknowledgement on the result queue. We drain that single
+    # sentinel here; subsequent results belong to jobs. ``boot_ok``
+    # only flips on a real ready; on error we set ``ready`` to unblock
+    # the parent's wait but leave ``boot_ok`` clear so the pool keeps
+    # the worker out of the available queue.
+    def _await_ready() -> None:
+        while not shutdown_event.is_set():
+            try:
+                msg = result_queue.get(timeout=1.0)
+            except queue.Empty:
+                continue
+            if isinstance(msg, dict) and msg.get("kind") == "ready":
+                boot_ok.set()
+                ready.set()
+                return
+            if isinstance(msg, dict) and msg.get("kind") == "error":
+                logger.error("pool: worker %s failed to boot: %s", worker_id, msg.get("error"))
+                ready.set()
+                return
+
+    threading.Thread(target=_await_ready, daemon=True).start()
+
+    return _WorkerHandle(
+        process=process,
+        job_queue=job_queue,
+        result_queue=result_queue,
+        gpu_id=gpu_id,
+        worker_id=worker_id,
+        ready=ready,
+        boot_ok=boot_ok,
+        shutdown_event=shutdown_event,
+    )
+
+
+__all__ = [
+    "GpuPool",
+    "InProcessGpuPool",
+    "PoolAcquireTimeout",
+    "PoolAssignment",
+    "PoolHealth",
+    "SubprocessGpuPool",
+    "WorkerFactory",
+    "worker_main",
+]

fastvideo/entrypoints/streaming/mock_server.py

@@ -0,0 +1,122 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Mock streaming server — a frontend dev aid.
+
+Boots the same FastAPI app the real streaming server uses, but backs
+it with :class:`InProcessGpuPool` wrapping a synthetic generator that
+emits pre-baked RGB frames. No GPU or model weights required.
+
+Use cases:
+
+* Frontend development without a real model loaded.
+* Integration tests that exercise the WS protocol end-to-end.
+* Reproducing protocol bugs locally.
+
+Launch: ``python -m fastvideo.entrypoints.streaming.mock_server``.
+"""
+from __future__ import annotations
+
+import argparse
+import time
+from dataclasses import dataclass
+from typing import Any
+
+import numpy as np
+
+from fastvideo.api.schema import (
+    ContinuationState,
+    GenerationRequest,
+    GeneratorConfig,
+    SamplingConfig,
+    ServeConfig,
+    StreamingConfig,
+)
+from fastvideo.entrypoints.streaming.server import build_app
+
+
+@dataclass
+class MockGenerator:
+    """Generator stand-in that returns synthetic gradient frames.
+
+    Each call produces one segment worth of frames whose pixels vary by
+    a constant derived from the request seed and segment index. Latency
+    is configurable via ``sleep_ms`` so the caller can exercise slow-
+    generate scenarios without spinning a GPU.
+    """
+
+    sleep_ms: float = 0.0
+
+    def generate(self, request: GenerationRequest) -> dict[str, Any]:
+        if self.sleep_ms:
+            time.sleep(self.sleep_ms / 1000.0)
+        width = max(16, request.sampling.width)
+        height = max(16, request.sampling.height)
+        num_frames = max(1, request.sampling.num_frames)
+        frames = [_gradient_frame(height, width, idx, seed=request.sampling.seed) for idx in range(num_frames)]
+        state = ContinuationState(
+            kind="ltx2.v1",
+            payload={
+                "schema_version": 1,
+                "segment_index": 0,
+                "source_prompt": request.prompt,
+            },
+        )
+        return {
+            "frames": frames,
+            "audio_sample_rate": 24000,
+            "state": state,
+        }
+
+
+def _gradient_frame(height: int, width: int, idx: int, *, seed: int) -> np.ndarray:
+    base = (idx * 17 + seed * 3) % 256
+    row = np.linspace(base, (base + 64) % 256, width, dtype=np.uint8)
+    frame = np.tile(row, (height, 1))
+    stacked = np.stack([frame, np.roll(frame, 8, axis=1), np.roll(frame, 16, axis=1)], axis=-1)
+    return stacked.astype(np.uint8)
+
+
+def build_mock_app(*, sleep_ms: float = 0.0):
+    """Build a FastAPI app backed by :class:`MockGenerator`."""
+    serve_config = ServeConfig(
+        generator=GeneratorConfig(model_path="/models/mock"),
+        streaming=StreamingConfig(
+            session_timeout_seconds=120,
+            generation_segment_cap=6,
+        ),
+    )
+    serve_config.default_request.sampling = SamplingConfig(
+        num_frames=24,
+        height=256,
+        width=256,
+        fps=24,
+        num_inference_steps=1,
+    )
+    return build_app(serve_config, MockGenerator(sleep_ms=sleep_ms))
+
+
+def main() -> None:  # pragma: no cover - CLI entry
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--host", default="127.0.0.1")
+    parser.add_argument("--port", type=int, default=8000)
+    parser.add_argument(
+        "--sleep-ms",
+        type=float,
+        default=0.0,
+        help="Per-segment artificial latency for testing slow paths",
+    )
+    args = parser.parse_args()
+
+    import uvicorn
+
+    app = build_mock_app(sleep_ms=args.sleep_ms)
+    uvicorn.run(app, host=args.host, port=args.port)
+
+
+__all__ = [
+    "MockGenerator",
+    "build_mock_app",
+    "main",
+]
+
+if __name__ == "__main__":  # pragma: no cover - CLI entry
+    main()

fastvideo/entrypoints/streaming/prompt/__init__.py

@@ -0,0 +1,36 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Prompt pipeline for the streaming server.
+
+* :mod:`providers` — LLM backend abstraction + built-in adapters
+* :mod:`enhancer` — provider-agnostic enhance / auto-extend / rewrite
+  operations on top of the provider layer
+
+All of this is optional; the streaming server runs fine without it
+(PR 7.5's skeleton never invokes the enhancer). When the operator
+enables ``ServeConfig.streaming.prompt.enabled``, the server routes
+each ``session_init_v2`` curated prompt through ``enhance`` before the
+first segment.
+"""
+from fastvideo.entrypoints.streaming.prompt.enhancer import (
+    PromptEnhancer,
+    PromptOperation,
+)
+from fastvideo.entrypoints.streaming.prompt.providers.base import (
+    LLMMessage,
+    LLMProvider,
+    LLMProviderError,
+    LLMRequest,
+    LLMResponse,
+    LLMTimeoutError,
+)
+
+__all__ = [
+    "LLMMessage",
+    "LLMProvider",
+    "LLMProviderError",
+    "LLMRequest",
+    "LLMResponse",
+    "LLMTimeoutError",
+    "PromptEnhancer",
+    "PromptOperation",
+]

fastvideo/entrypoints/streaming/prompt/enhancer.py

@@ -0,0 +1,197 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Provider-agnostic prompt orchestration for the streaming server.
+
+Three operations the streaming server needs:
+
+* ``enhance`` — polish a user prompt (add cinematic detail, fix syntax)
+* ``auto_extend`` — generate a follow-on prompt for loop generation
+* ``rewrite`` — rewrite a seed prompt for a user-directed rewrite flow
+
+All three share the same orchestration: pick a provider in priority
+order, submit an ``LLMRequest``, fall back to the next provider on
+retryable errors, and surface a structured :class:`LLMResponse` back
+to the caller.
+
+System prompts are loaded from ``system_prompt_dir`` on construction
+and can be hot-reloaded via :meth:`PromptEnhancer.reload_system_prompts`.
+The streaming server's management endpoint calls that method in
+response to a ``rewrite_seed_prompts_started`` frame.
+"""
+from __future__ import annotations
+
+import enum
+import os
+from collections.abc import Sequence
+from dataclasses import dataclass, replace
+
+from fastvideo.entrypoints.streaming.prompt.providers.base import (
+    LLMMessage,
+    LLMProvider,
+    LLMProviderError,
+    LLMRequest,
+    LLMResponse,
+)
+from fastvideo.logger import init_logger
+
+logger = init_logger(__name__)
+
+
+class PromptOperation(enum.Enum):
+    ENHANCE = "enhance"
+    AUTO_EXTEND = "auto_extend"
+    REWRITE = "rewrite"
+
+
+@dataclass
+class _SystemPrompts:
+    enhance: str
+    auto_extend: str
+    rewrite: str
+
+
+_DEFAULT_SYSTEM_PROMPTS = _SystemPrompts(
+    enhance=("You are a prompt enhancer for cinematic video generation. Given "
+             "a user prompt, produce an enhanced prompt that is more vivid, "
+             "specific, and concrete. Keep the subject intact; add lighting, "
+             "camera, and motion detail. Reply with just the enhanced prompt."),
+    auto_extend=("You are a video continuation assistant. Given the current "
+                 "sequence of prompts, produce one new prompt that naturally "
+                 "continues the sequence. Reply with just the next prompt."),
+    rewrite=("You are a creative prompt rewriter. Given a seed prompt, produce "
+             "a set of alternative prompts that explore different angles, "
+             "styles, and moods. Reply with one prompt per line."),
+)
+
+
+class PromptEnhancer:
+    """Orchestrates prompt operations across a priority-ordered provider
+    list with structured fallback + hot-reloadable system prompts.
+
+    Usage::
+
+        enhancer = PromptEnhancer(
+            providers=[CerebrasProvider(), GroqProvider()],
+            model="gpt-oss-120b",
+            system_prompt_dir="/etc/fastvideo/prompts",
+        )
+        response = await enhancer.enhance("a fox running through snow")
+    """
+
+    def __init__(
+        self,
+        *,
+        providers: Sequence[LLMProvider],
+        model: str,
+        timeout_ms: int = 20000,
+        temperature: float = 0.7,
+        max_tokens: int | None = 256,
+        system_prompt_dir: str | None = None,
+    ) -> None:
+        if not providers:
+            raise ValueError("PromptEnhancer requires at least one LLMProvider")
+        self._providers = list(providers)
+        self._model = model
+        self._timeout_ms = timeout_ms
+        self._temperature = temperature
+        self._max_tokens = max_tokens
+        self._system_prompt_dir = system_prompt_dir
+        self._system_prompts = self._load_system_prompts()
+
+    @property
+    def providers(self) -> list[LLMProvider]:
+        return list(self._providers)
+
+    def register_provider(self, provider: LLMProvider, *, priority: int = -1) -> None:
+        """Insert an additional provider. ``priority=0`` makes it primary;
+        ``priority=-1`` (default) appends as a fallback."""
+        if priority < 0:
+            self._providers.append(provider)
+        else:
+            self._providers.insert(priority, provider)
+
+    def reload_system_prompts(self) -> None:
+        """Re-read the system prompt files from ``system_prompt_dir``.
+
+        The streaming server exposes this via a management endpoint so
+        operators can iterate on prompt templates without restarting
+        workers.
+        """
+        self._system_prompts = self._load_system_prompts()
+        logger.info("prompt enhancer: reloaded system prompts from %s", self._system_prompt_dir or "defaults")
+
+    async def enhance(self, prompt: str) -> LLMResponse:
+        return await self._run(
+            PromptOperation.ENHANCE,
+            system=self._system_prompts.enhance,
+            user=prompt,
+        )
+
+    async def auto_extend(self, prior_prompts: Sequence[str]) -> LLMResponse:
+        user = "\n".join(prior_prompts)
+        return await self._run(
+            PromptOperation.AUTO_EXTEND,
+            system=self._system_prompts.auto_extend,
+            user=user,
+        )
+
+    async def rewrite(self, seed_prompt: str) -> LLMResponse:
+        return await self._run(
+            PromptOperation.REWRITE,
+            system=self._system_prompts.rewrite,
+            user=seed_prompt,
+        )
+
+    async def _run(
+        self,
+        operation: PromptOperation,
+        *,
+        system: str,
+        user: str,
+    ) -> LLMResponse:
+        request = LLMRequest(
+            messages=[
+                LLMMessage(role="system", content=system),
+                LLMMessage(role="user", content=user),
+            ],
+            model=self._model,
+            max_tokens=self._max_tokens,
+            temperature=self._temperature,
+            timeout_ms=self._timeout_ms,
+        )
+        last_error: LLMProviderError | None = None
+        for idx, provider in enumerate(self._providers):
+            try:
+                response = await provider.complete(request)
+                if idx > 0:
+                    # Mark the fallback flag without losing any other
+                    # response fields the provider populated.
+                    response = replace(response, fallback_used=True)
+                return response
+            except LLMProviderError as exc:
+                logger.warning("prompt %s: provider %s failed: %s; trying next", operation.value, provider.name, exc)
+                last_error = exc
+                if not exc.retryable:
+                    break
+        assert last_error is not None
+        raise last_error
+
+    def _load_system_prompts(self) -> _SystemPrompts:
+        if not self._system_prompt_dir:
+            return _DEFAULT_SYSTEM_PROMPTS
+        return _SystemPrompts(
+            enhance=_read_prompt(self._system_prompt_dir, "enhance.txt", _DEFAULT_SYSTEM_PROMPTS.enhance),
+            auto_extend=_read_prompt(self._system_prompt_dir, "auto_extend.txt", _DEFAULT_SYSTEM_PROMPTS.auto_extend),
+            rewrite=_read_prompt(self._system_prompt_dir, "rewrite.txt", _DEFAULT_SYSTEM_PROMPTS.rewrite),
+        )
+
+
+def _read_prompt(dirname: str, filename: str, default: str) -> str:
+    path = os.path.join(dirname, filename)
+    if not os.path.exists(path):
+        return default
+    with open(path, encoding="utf-8") as f:
+        content = f.read().strip()
+    return content or default
+
+
+__all__ = ["PromptEnhancer", "PromptOperation"]

fastvideo/entrypoints/streaming/prompt/providers/__init__.py

@@ -0,0 +1,24 @@
+# SPDX-License-Identifier: Apache-2.0
+"""LLM provider implementations used by the prompt enhancer."""
+from fastvideo.entrypoints.streaming.prompt.providers.base import (
+    LLMMessage,
+    LLMProvider,
+    LLMProviderError,
+    LLMRequest,
+    LLMResponse,
+    LLMTimeoutError,
+)
+from fastvideo.entrypoints.streaming.prompt.providers.cerebras import (
+    CerebrasProvider, )
+from fastvideo.entrypoints.streaming.prompt.providers.groq import GroqProvider
+
+__all__ = [
+    "CerebrasProvider",
+    "GroqProvider",
+    "LLMMessage",
+    "LLMProvider",
+    "LLMProviderError",
+    "LLMRequest",
+    "LLMResponse",
+    "LLMTimeoutError",
+]

fastvideo/entrypoints/streaming/prompt/providers/_openai_compat.py

@@ -0,0 +1,101 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Shared HTTP path for OpenAI-compatible ``/chat/completions`` providers.
+
+Cerebras and Groq both expose the OpenAI chat-completions schema, so
+the request shape, error mapping, and response decoding are identical
+between them. This module centralizes that logic; the per-provider
+modules stay thin (just defaults + env var wiring).
+"""
+from __future__ import annotations
+
+import time
+
+from fastvideo.entrypoints.streaming.prompt.providers.base import (
+    LLMProviderError,
+    LLMRequest,
+    LLMResponse,
+    LLMTimeoutError,
+)
+
+
+async def complete_openai_compatible(
+    *,
+    api_key: str | None,
+    api_key_hint: str,
+    base_url: str,
+    provider_name: str,
+    request: LLMRequest,
+) -> LLMResponse:
+    """Issue a chat-completions call and decode the OpenAI response."""
+    if not api_key:
+        raise LLMProviderError(
+            f"{provider_name} provider requires {api_key_hint} "
+            "(or explicit api_key=...)",
+            retryable=False,
+        )
+    try:
+        import httpx
+    except ImportError as exc:  # pragma: no cover - optional dep
+        raise LLMProviderError(
+            f"{provider_name} provider requires httpx; install httpx",
+            retryable=False,
+        ) from exc
+
+    timeout_s = (request.timeout_ms or 20000) / 1000.0
+    t0 = time.perf_counter()
+    try:
+        async with httpx.AsyncClient(timeout=timeout_s) as client:
+            response = await client.post(
+                f"{base_url}/chat/completions",
+                headers={
+                    "Authorization": f"Bearer {api_key}",
+                    "Content-Type": "application/json",
+                },
+                json={
+                    "model": request.model,
+                    "messages": [{
+                        "role": m.role,
+                        "content": m.content
+                    } for m in request.messages],
+                    "max_tokens": request.max_tokens,
+                    "temperature": request.temperature,
+                },
+            )
+    except httpx.TimeoutException as exc:
+        raise LLMTimeoutError(f"{provider_name} timed out after {timeout_s}s") from exc
+    except httpx.HTTPError as exc:
+        raise LLMProviderError(f"{provider_name} HTTP error: {exc}") from exc
+
+    if response.status_code >= 400:
+        # 5xx and 429 (rate-limit) are retryable: another provider may
+        # succeed. 4xx (auth, bad-request, etc.) are client errors —
+        # the enhancer should stop fallback traversal.
+        retryable = (response.status_code >= 500 or response.status_code == 429)
+        raise LLMProviderError(
+            f"{provider_name} returned {response.status_code}: "
+            f"{response.text[:200]}",
+            retryable=retryable,
+        )
+
+    try:
+        data = response.json()
+    except Exception as exc:
+        # Non-JSON body usually means a proxy / load-balancer error
+        # page; leave it retryable so a fallback provider can try.
+        raise LLMProviderError(f"{provider_name} returned non-JSON body: {exc}") from exc
+
+    choices = data.get("choices") or []
+    if not choices:
+        raise LLMProviderError(f"{provider_name} returned no choices")
+    content = choices[0].get("message", {}).get("content") or ""
+
+    latency_ms = (time.perf_counter() - t0) * 1000.0
+    return LLMResponse(
+        content=content.strip(),
+        provider=provider_name,
+        model=request.model,
+        latency_ms=latency_ms,
+    )
+
+
+__all__ = ["complete_openai_compatible"]

fastvideo/entrypoints/streaming/prompt/providers/base.py

@@ -0,0 +1,85 @@
+# SPDX-License-Identifier: Apache-2.0
+"""LLM provider protocol + DTOs used by the prompt enhancer.
+
+Third-party users add a new provider by implementing
+:class:`LLMProvider` and registering it with a prompt enhancer
+instance. The shipped providers live in sibling modules
+(``cerebras.py``, ``groq.py``) and each is ~100-200 LOC — the
+provider layer is intentionally thin so the enhancer stays
+provider-agnostic.
+"""
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Literal, Protocol, runtime_checkable
+
+
+@dataclass
+class LLMMessage:
+    role: Literal["system", "user", "assistant"]
+    content: str
+
+
+@dataclass
+class LLMRequest:
+    messages: list[LLMMessage]
+    model: str
+    max_tokens: int | None = None
+    temperature: float | None = None
+    timeout_ms: int | None = None
+
+
+@dataclass
+class LLMResponse:
+    content: str
+    provider: str
+    model: str
+    latency_ms: float
+    fallback_used: bool = False
+
+
+class LLMProviderError(RuntimeError):
+    """Raised when an LLM provider fails a request.
+
+    ``retryable`` controls whether the enhancer falls back to the next
+    provider. It is settable per-instance so the same exception type
+    can describe retryable transport errors (5xx, 429) and
+    non-retryable client errors (4xx auth/bad-request) without forcing
+    a separate subclass for every status family.
+    """
+
+    def __init__(self, message: str, *, retryable: bool = True) -> None:
+        super().__init__(message)
+        self.retryable = retryable
+
+
+class LLMTimeoutError(LLMProviderError):
+    """Raised when an LLM provider times out — always retryable."""
+
+    def __init__(self, message: str) -> None:
+        super().__init__(message, retryable=True)
+
+
+@runtime_checkable
+class LLMProvider(Protocol):
+    """Provider interface every LLM adapter implements.
+
+    Providers are async-first because every built-in implementation
+    talks to an HTTP API. Synchronous providers can wrap their call in
+    ``asyncio.to_thread`` internally.
+    """
+
+    name: str
+
+    async def complete(self, request: LLMRequest) -> LLMResponse:
+        ...
+
+
+__all__ = [
+    "LLMMessage",
+    "LLMProvider",
+    "LLMProviderError",
+    "LLMRequest",
+    "LLMResponse",
+    "LLMTimeoutError",
+]

fastvideo/entrypoints/streaming/prompt/providers/cerebras.py

@@ -0,0 +1,44 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Cerebras LLM provider (OpenAI-compatible chat endpoint)."""
+from __future__ import annotations
+
+import os
+from dataclasses import dataclass
+
+from fastvideo.entrypoints.streaming.prompt.providers._openai_compat import (
+    complete_openai_compatible, )
+from fastvideo.entrypoints.streaming.prompt.providers.base import (
+    LLMRequest,
+    LLMResponse,
+)
+
+_DEFAULT_BASE_URL = "https://api.cerebras.ai/v1"
+_API_KEY_ENV = "CEREBRAS_API_KEY"
+
+
+@dataclass
+class CerebrasProvider:
+    """Cerebras inference adapter.
+
+    ``api_key`` falls back to ``CEREBRAS_API_KEY`` when unset.
+    """
+
+    api_key: str | None = None
+    base_url: str = _DEFAULT_BASE_URL
+    name: str = "cerebras"
+
+    def __post_init__(self) -> None:
+        if self.api_key is None:
+            self.api_key = os.environ.get(_API_KEY_ENV)
+
+    async def complete(self, request: LLMRequest) -> LLMResponse:
+        return await complete_openai_compatible(
+            api_key=self.api_key,
+            api_key_hint=_API_KEY_ENV,
+            base_url=self.base_url,
+            provider_name=self.name,
+            request=request,
+        )
+
+
+__all__ = ["CerebrasProvider"]

fastvideo/entrypoints/streaming/prompt/providers/groq.py

@@ -0,0 +1,46 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Groq LLM provider (OpenAI-compatible chat endpoint)."""
+from __future__ import annotations
+
+import os
+from dataclasses import dataclass
+
+from fastvideo.entrypoints.streaming.prompt.providers._openai_compat import (
+    complete_openai_compatible, )
+from fastvideo.entrypoints.streaming.prompt.providers.base import (
+    LLMRequest,
+    LLMResponse,
+)
+
+_DEFAULT_BASE_URL = "https://api.groq.com/openai/v1"
+_API_KEY_ENV = "GROQ_API_KEY"
+
+
+@dataclass
+class GroqProvider:
+    """Groq inference adapter.
+
+    Identical wire format to :class:`CerebrasProvider`; both go through
+    :func:`complete_openai_compatible`. The two providers differ only
+    in base URL, env var, and model id conventions.
+    """
+
+    api_key: str | None = None
+    base_url: str = _DEFAULT_BASE_URL
+    name: str = "groq"
+
+    def __post_init__(self) -> None:
+        if self.api_key is None:
+            self.api_key = os.environ.get(_API_KEY_ENV)
+
+    async def complete(self, request: LLMRequest) -> LLMResponse:
+        return await complete_openai_compatible(
+            api_key=self.api_key,
+            api_key_hint=_API_KEY_ENV,
+            base_url=self.base_url,
+            provider_name=self.name,
+            request=request,
+        )
+
+
+__all__ = ["GroqProvider"]

fastvideo/entrypoints/streaming/prompt/rewrite.py

@@ -0,0 +1,82 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Rewrite payload builder.
+
+The UI's "rewrite seed prompts" flow asks the enhancer to produce a
+batch of alternative prompts given one seed. This module packages the
+seed + options into the payload the enhancer expects and unpacks the
+response back into a typed :class:`RewriteResult`.
+
+Separating this from :mod:`enhancer` keeps the enhancer provider-
+agnostic; anything UI-specific (how many alternatives to request, how
+to split the response, temperature) lives here.
+"""
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass
+
+from fastvideo.entrypoints.streaming.prompt.enhancer import PromptEnhancer
+
+_LEADING_MARKER_RE = re.compile(r"^(?:[-*•]\s*|\d+\s*[.)]\s*)+")
+
+
+@dataclass
+class RewriteOptions:
+    count: int = 3
+    """Number of alternative prompts to request."""
+    temperature: float | None = None
+
+
+@dataclass
+class RewriteResult:
+    seed_prompt: str
+    alternatives: list[str]
+    provider: str
+    model: str
+    latency_ms: float
+    fallback_used: bool = False
+
+
+async def build_rewrite(
+    enhancer: PromptEnhancer,
+    seed_prompt: str,
+    *,
+    options: RewriteOptions | None = None,
+) -> RewriteResult:
+    """Run a rewrite op through the enhancer and return a typed result."""
+    if not seed_prompt.strip():
+        raise ValueError("rewrite seed prompt must be non-empty")
+    options = options or RewriteOptions()
+    response = await enhancer.rewrite(seed_prompt)
+    alternatives = _split_response(response.content, limit=options.count)
+    return RewriteResult(
+        seed_prompt=seed_prompt,
+        alternatives=alternatives,
+        provider=response.provider,
+        model=response.model,
+        latency_ms=response.latency_ms,
+        fallback_used=response.fallback_used,
+    )
+
+
+def _split_response(content: str, *, limit: int) -> list[str]:
+    """Split the LLM response into discrete prompt candidates.
+
+    The shipped system prompt instructs the model to emit one prompt
+    per line; this function is forgiving about numbered lists or
+    leading bullets so user-supplied system prompts don't break it.
+    """
+    lines = [line.strip() for line in content.splitlines() if line.strip()]
+    cleaned: list[str] = []
+    for line in lines:
+        stripped = _LEADING_MARKER_RE.sub("", line).strip()
+        if stripped:
+            cleaned.append(stripped)
+    return cleaned[:max(1, limit)]
+
+
+__all__ = [
+    "RewriteOptions",
+    "RewriteResult",
+    "build_rewrite",
+]

fastvideo/entrypoints/streaming/prompt/safety.py

@@ -0,0 +1,146 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Optional prompt safety filter.
+
+Uses a fastText classifier to score prompts against a banned-content
+rubric. Only loaded when ``ServeConfig.streaming.safety.enabled`` is
+True and fastText is installed — users who don't need it see no
+runtime cost.
+
+Install: ``pip install fastvideo[prompt-safety]`` (ships fasttext as an
+optional extra) or install fasttext directly.
+"""
+from __future__ import annotations
+
+import enum
+import threading
+from dataclasses import dataclass
+from typing import Any
+
+from fastvideo.logger import init_logger
+
+logger = init_logger(__name__)
+
+
+class SafetyDecision(enum.Enum):
+    ALLOW = "allow"
+    BLOCK = "block"
+    UNAVAILABLE = "unavailable"
+    """Returned when the classifier can't run (not configured, fastText
+    missing). Safety is opt-in; the server treats ``UNAVAILABLE`` as
+    ``ALLOW`` but logs it so operators know the filter is off."""
+
+
+@dataclass
+class SafetyResult:
+    prompt: str
+    decision: SafetyDecision
+    score: float = 0.0
+    label: str | None = None
+    reason: str | None = None
+
+
+class PromptSafetyFilter:
+    """Minimal fastText-backed prompt safety filter.
+
+    Loads the classifier lazily on first use so the streaming server
+    can construct the filter eagerly at startup without paying the
+    model-load cost when safety is disabled.
+    """
+
+    def __init__(
+        self,
+        *,
+        classifier_path: str | None,
+        enabled: bool = True,
+        block_threshold: float = 0.5,
+    ) -> None:
+        self._classifier_path = classifier_path
+        self._enabled = enabled
+        self._block_threshold = block_threshold
+        self._model: Any | None = None
+        self._load_attempted = False
+        self._load_lock = threading.Lock()
+
+    @property
+    def enabled(self) -> bool:
+        return self._enabled and self._classifier_path is not None
+
+    def classify(self, prompt: str) -> SafetyResult:
+        if not self.enabled:
+            return SafetyResult(
+                prompt=prompt,
+                decision=SafetyDecision.UNAVAILABLE,
+                reason="safety filter not enabled",
+            )
+        model = self._ensure_loaded()
+        if model is None:
+            return SafetyResult(
+                prompt=prompt,
+                decision=SafetyDecision.UNAVAILABLE,
+                reason="fastText model unavailable",
+            )
+        try:
+            labels, probs = model.predict(prompt.replace("\n", " "), k=1)
+        except Exception as exc:  # pragma: no cover - defensive
+            logger.warning("safety: classifier failed: %s", exc)
+            return SafetyResult(
+                prompt=prompt,
+                decision=SafetyDecision.UNAVAILABLE,
+                reason=f"classifier error: {exc}",
+            )
+        label = labels[0].removeprefix("__label__") if labels else None
+        score = float(probs[0]) if len(probs) else 0.0
+        decision = (SafetyDecision.BLOCK if
+                    (label == "unsafe" and score >= self._block_threshold) else SafetyDecision.ALLOW)
+        return SafetyResult(
+            prompt=prompt,
+            decision=decision,
+            score=score,
+            label=label,
+        )
+
+    def _ensure_loaded(self) -> Any | None:
+        if self._model is not None:
+            return self._model
+        if self._load_attempted:
+            return None
+        with self._load_lock:
+            if self._model is not None:
+                return self._model
+            if self._load_attempted:
+                return None
+            self._load_attempted = True
+            if self._classifier_path is None:
+                return None
+            try:
+                import fasttext  # type: ignore[import-not-found]
+            except ImportError:
+                logger.warning("safety: fasttext not installed; safety filter disabled. "
+                               "Install fastvideo[prompt-safety] to enable.")
+                return None
+            try:
+                self._model = fasttext.load_model(self._classifier_path)
+            except Exception as exc:  # pragma: no cover - requires real model
+                logger.warning("safety: failed to load %s: %s", self._classifier_path, exc)
+                return None
+            return self._model
+
+
+def first_blocked(
+    filter_: PromptSafetyFilter,
+    prompts: list[str],
+) -> SafetyResult | None:
+    """Return the first prompt the filter blocks, or ``None``."""
+    for prompt in prompts:
+        result = filter_.classify(prompt)
+        if result.decision is SafetyDecision.BLOCK:
+            return result
+    return None
+
+
+__all__ = [
+    "PromptSafetyFilter",
+    "SafetyDecision",
+    "SafetyResult",
+    "first_blocked",
+]

fastvideo/entrypoints/streaming/router/__init__.py

@@ -0,0 +1,27 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Multi-replica load balancer + WebSocket proxy for the streaming server.
+
+Sits in front of one-or-more streaming-server replicas and forwards
+WebSocket sessions to a healthy primary, with failover to secondaries.
+Kept in-repo under ``fastvideo/entrypoints/streaming/router/`` per the
+PR plan's default; the alternative (separate package) is an open
+question deferred to review.
+"""
+from fastvideo.entrypoints.streaming.router.registry import (
+    Replica,
+    ReplicaHealth,
+    ReplicaRegistry,
+    ReplicaStatus,
+)
+from fastvideo.entrypoints.streaming.router.config import RouterConfig
+from fastvideo.entrypoints.streaming.router.main import build_router_app, run_router
+
+__all__ = [
+    "Replica",
+    "ReplicaHealth",
+    "ReplicaRegistry",
+    "ReplicaStatus",
+    "RouterConfig",
+    "build_router_app",
+    "run_router",
+]

fastvideo/entrypoints/streaming/router/config.py

@@ -0,0 +1,88 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Typed router configuration."""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from urllib.parse import urlparse
+
+
+@dataclass
+class ReplicaEndpoint:
+    """One backend replica the router can route to."""
+
+    url: str
+    """HTTP base URL, e.g. ``http://host:8000``. WebSocket URL is
+    derived automatically by replacing the scheme."""
+    name: str | None = None
+    primary: bool = False
+    """``True`` = prefer this replica over others in steady state."""
+    weight: float = 1.0
+
+
+@dataclass
+class RouterConfig:
+    """Typed router config loaded from a YAML file.
+
+    Example::
+
+        router:
+          host: 0.0.0.0
+          port: 9000
+          replicas:
+            - url: http://streamer-a:8000
+              primary: true
+            - url: http://streamer-b:8000
+          health_check:
+            path: /health
+            interval_seconds: 5
+            failure_threshold: 3
+
+    Validation runs in ``__post_init__``: empty replicas, non-positive
+    intervals/timeouts, thresholds < 1, non-http(s) URLs, and more than
+    one primary all raise ``ValueError`` so misconfigurations surface at
+    load time rather than as confusing runtime failures.
+    """
+
+    host: str = "0.0.0.0"
+    port: int = 9000
+    replicas: list[ReplicaEndpoint] = field(default_factory=list)
+    health_check_path: str = "/health"
+    health_check_interval_seconds: float = 5.0
+    health_check_timeout_seconds: float = 2.0
+    failure_threshold: int = 3
+    recovery_threshold: int = 2
+
+    def __post_init__(self) -> None:
+        if not self.replicas:
+            raise ValueError("RouterConfig.replicas must list at least one replica")
+        if self.health_check_interval_seconds <= 0:
+            raise ValueError(f"health_check_interval_seconds must be > 0, got {self.health_check_interval_seconds}")
+        if self.health_check_timeout_seconds <= 0:
+            raise ValueError(f"health_check_timeout_seconds must be > 0, got {self.health_check_timeout_seconds}")
+        if self.failure_threshold < 1:
+            raise ValueError(f"failure_threshold must be >= 1, got {self.failure_threshold}")
+        if self.recovery_threshold < 1:
+            raise ValueError(f"recovery_threshold must be >= 1, got {self.recovery_threshold}")
+        seen_urls: set[str] = set()
+        for replica in self.replicas:
+            if not replica.url.startswith(("http://", "https://")):
+                raise ValueError(f"ReplicaEndpoint.url must start with http:// or https://, got {replica.url!r}")
+            parsed = urlparse(replica.url)
+            if parsed.path not in ("", "/"):
+                raise ValueError(f"ReplicaEndpoint.url must be a base host[:port] URL without a path; "
+                                 f"got {replica.url!r} with path {parsed.path!r}. The router appends "
+                                 "`/health` and `/v1/stream` itself.")
+            if parsed.query or parsed.fragment:
+                raise ValueError(f"ReplicaEndpoint.url must not include query/fragment; got {replica.url!r}")
+            if replica.url in seen_urls:
+                raise ValueError(f"Duplicate ReplicaEndpoint.url {replica.url!r}; "
+                                 "router selection keys by URL so duplicates would silently collapse")
+            seen_urls.add(replica.url)
+        primaries = sum(1 for r in self.replicas if r.primary)
+        if primaries > 1:
+            raise ValueError(f"RouterConfig allows at most one primary replica; got {primaries}. "
+                             "Multi-primary load distribution is deferred — promote one replica to "
+                             "primary and treat the rest as secondaries.")
+
+
+__all__ = ["ReplicaEndpoint", "RouterConfig"]

fastvideo/entrypoints/streaming/router/main.py

@@ -0,0 +1,218 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Router FastAPI entry point.
+
+Exposes the same ``/v1/stream`` WebSocket path the backend servers do,
+accepts a client, picks a healthy replica from the registry, and
+proxies frames bidirectionally.
+
+PR 7.9 ships the minimum-viable shape: explicit replica list, single
+primary, JSON + binary passthrough in both directions, and a
+``/status`` endpoint for operators. Sticky-session routing (so a
+reconnect lands on the same backend) is left for a follow-up.
+"""
+from __future__ import annotations
+
+import asyncio
+import contextlib
+from dataclasses import dataclass
+
+from fastapi import FastAPI, WebSocket, WebSocketDisconnect
+from fastapi.responses import JSONResponse
+
+from fastvideo.entrypoints.streaming.router.config import RouterConfig
+from fastvideo.entrypoints.streaming.router.registry import (
+    ReplicaRegistry,
+    run_health_check_loop,
+)
+from fastvideo.logger import init_logger
+
+logger = init_logger(__name__)
+
+
+@dataclass
+class _RouterState:
+    config: RouterConfig
+    registry: ReplicaRegistry
+    stop_event: asyncio.Event
+    health_task: asyncio.Task | None = None
+
+
+def build_router_app(
+    config: RouterConfig,
+    *,
+    registry: ReplicaRegistry | None = None,
+) -> FastAPI:
+    """Build the router FastAPI app.
+
+    ``registry`` can be injected for tests; defaults to one built from
+    ``config.replicas``.
+    """
+    registry = registry or ReplicaRegistry(config.replicas)
+    state = _RouterState(
+        config=config,
+        registry=registry,
+        stop_event=asyncio.Event(),
+    )
+
+    @contextlib.asynccontextmanager
+    async def _lifespan(_app: FastAPI):
+        state.health_task = asyncio.create_task(
+            run_health_check_loop(
+                registry=state.registry,
+                config=state.config,
+                stop_event=state.stop_event,
+            ))
+        try:
+            yield
+        finally:
+            state.stop_event.set()
+            if state.health_task is not None:
+                with contextlib.suppress(asyncio.CancelledError):
+                    await state.health_task
+
+    app = FastAPI(title="FastVideo Streaming Router", lifespan=_lifespan)
+
+    @app.get("/status")
+    async def _status() -> JSONResponse:
+        return JSONResponse({
+            "replicas": [{
+                "url": r.url,
+                "primary": r.primary,
+                "status": r.health.status.value,
+                "last_ok_at": r.health.last_ok_at,
+                "last_latency_ms": r.health.last_latency_ms,
+                "consecutive_failures": r.health.consecutive_failures,
+            } for r in state.registry.all()],
+        })
+
+    @app.websocket("/v1/stream")
+    async def _proxy(websocket: WebSocket) -> None:
+        await websocket.accept()
+        replica = state.registry.select()
+        if replica is None:
+            await websocket.send_json({
+                "type": "error",
+                "code": "gpu_unavailable",
+                "message": "router: no healthy replica available",
+                "retryable": True,
+            })
+            await websocket.close(code=1013, reason="no_healthy_replica")
+            return
+
+        ws_url = _websocket_url_for(replica.url)
+        try:
+            await _bridge_session(websocket, ws_url)
+        except WebSocketDisconnect:
+            logger.info("router: client disconnected")
+        except Exception as exc:
+            logger.exception("router: bridge failed: %s", exc)
+            with contextlib.suppress(RuntimeError):
+                await websocket.send_json({
+                    "type": "error",
+                    "code": "worker_failed",
+                    "message": f"router bridge failed: {exc}",
+                    "retryable": True,
+                })
+            with contextlib.suppress(RuntimeError):
+                await websocket.close(code=1011)
+
+    app.state.router_state = state
+    return app
+
+
+def run_router(config: RouterConfig) -> None:  # pragma: no cover - CLI
+    import uvicorn
+
+    app = build_router_app(config)
+    uvicorn.run(app, host=config.host, port=config.port)
+
+
+async def _bridge_session(
+    client_ws: WebSocket,
+    backend_ws_url: str,
+) -> None:
+    """Connect to backend and shuttle messages in both directions.
+
+    Uses ``websockets`` for the backend side; imported lazily to keep
+    the router's import graph small for users who only want the server.
+
+    Cancellation: when either direction completes (client disconnect,
+    backend close, exception), the other is cancelled explicitly and
+    both are drained before returning. Unexpected exceptions from the
+    direction that completed first are re-raised; normal disconnect
+    paths (``WebSocketDisconnect``, ``ConnectionClosed``,
+    ``CancelledError``) are swallowed.
+    """
+    try:
+        import websockets
+    except ImportError as exc:  # pragma: no cover - optional extra
+        raise RuntimeError("router requires the `websockets` package for backend proxying") from exc
+
+    async with websockets.connect(backend_ws_url + "/v1/stream") as backend_ws:
+        c2b = asyncio.create_task(_forward_client_to_backend(client_ws, backend_ws))
+        b2c = asyncio.create_task(_forward_backend_to_client(backend_ws, client_ws))
+        try:
+            done, _pending = await asyncio.wait(
+                {c2b, b2c},
+                return_when=asyncio.FIRST_COMPLETED,
+            )
+        finally:
+            for task in (c2b, b2c):
+                if not task.done():
+                    task.cancel()
+            await asyncio.gather(c2b, b2c, return_exceptions=True)
+        for task in done:
+            task_exc = task.exception()
+            if task_exc is not None and not _is_normal_disconnect(task_exc):
+                raise task_exc
+
+
+def _is_normal_disconnect(exc: BaseException) -> bool:
+    """Whether ``exc`` is a routine WebSocket teardown vs a real bridge fault."""
+    if isinstance(exc, asyncio.CancelledError | WebSocketDisconnect):
+        return True
+    name = type(exc).__name__
+    # websockets.exceptions.ConnectionClosed{,OK,Error} all subclass
+    # WebSocketException; check by name to avoid the lazy-import dance.
+    return name.startswith("ConnectionClosed")
+
+
+async def _forward_client_to_backend(client_ws: WebSocket, backend_ws) -> None:
+    try:
+        while True:
+            msg = await client_ws.receive()
+            if msg.get("type") == "websocket.disconnect":
+                break
+            if "text" in msg and msg["text"] is not None:
+                await backend_ws.send(msg["text"])
+            elif "bytes" in msg and msg["bytes"] is not None:
+                await backend_ws.send(msg["bytes"])
+    finally:
+        with contextlib.suppress(Exception):
+            await backend_ws.close()
+
+
+async def _forward_backend_to_client(backend_ws, client_ws: WebSocket) -> None:
+    try:
+        async for frame in backend_ws:
+            if isinstance(frame, bytes):
+                await client_ws.send_bytes(frame)
+            else:
+                await client_ws.send_text(frame)
+    finally:
+        with contextlib.suppress(Exception):
+            await client_ws.close()
+
+
+def _websocket_url_for(http_url: str) -> str:
+    if http_url.startswith("https://"):
+        return "wss://" + http_url[len("https://"):]
+    if http_url.startswith("http://"):
+        return "ws://" + http_url[len("http://"):]
+    return http_url
+
+
+__all__ = [
+    "build_router_app",
+    "run_router",
+]

fastvideo/entrypoints/streaming/router/registry.py

@@ -0,0 +1,268 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Replica registry + health-check loop.
+
+The registry tracks the set of known backend replicas and their live
+health. The router consults it for "pick a backend for this session"
+decisions and a background task updates it from periodic HTTP probes.
+
+State machine per replica::
+
+    HEALTHY ──(N consecutive failures)──▶ UNHEALTHY
+       ▲                                     │
+       └──────(M consecutive successes)──────┘
+
+Where N = :attr:`RouterConfig.failure_threshold` and
+M = :attr:`RouterConfig.recovery_threshold`.
+"""
+from __future__ import annotations
+
+import asyncio
+import contextlib
+import enum
+import time
+from collections.abc import AsyncIterator, Awaitable, Callable
+from dataclasses import dataclass, field
+from typing import Any
+
+from fastvideo.entrypoints.streaming.router.config import (
+    ReplicaEndpoint,
+    RouterConfig,
+)
+from fastvideo.logger import init_logger
+
+HttpProbe = Any
+"""Structural alias for health-probe callables. Concrete signature is
+``async def __call__(url: str, *, timeout: float) -> tuple[float,
+str | None]``; typing.Callable cannot express keyword-only parameters,
+so duck-typing is the pragmatic compromise."""
+
+logger = init_logger(__name__)
+
+
+class ReplicaStatus(enum.Enum):
+    UNKNOWN = "unknown"
+    HEALTHY = "healthy"
+    UNHEALTHY = "unhealthy"
+
+
+@dataclass
+class ReplicaHealth:
+    status: ReplicaStatus = ReplicaStatus.UNKNOWN
+    last_ok_at: float | None = None
+    last_failure_at: float | None = None
+    consecutive_failures: int = 0
+    consecutive_successes: int = 0
+    last_latency_ms: float | None = None
+
+
+@dataclass
+class Replica:
+    endpoint: ReplicaEndpoint
+    health: ReplicaHealth = field(default_factory=ReplicaHealth)
+
+    @property
+    def url(self) -> str:
+        return self.endpoint.url
+
+    @property
+    def primary(self) -> bool:
+        return self.endpoint.primary
+
+    @property
+    def is_healthy(self) -> bool:
+        return self.health.status is ReplicaStatus.HEALTHY
+
+
+class ReplicaRegistry:
+    """Stateful map of replica URL → :class:`Replica`.
+
+    Selection favors primary replicas when healthy; otherwise the first
+    healthy non-primary is returned. When none are healthy, the
+    registry returns ``None`` so the router can reject incoming
+    sessions with ``gpu_unavailable``.
+    """
+
+    def __init__(self, replicas: list[ReplicaEndpoint]) -> None:
+        if not replicas:
+            raise ValueError("ReplicaRegistry requires at least one replica")
+        self._replicas: dict[str, Replica] = {endpoint.url: Replica(endpoint=endpoint) for endpoint in replicas}
+        self._lock = asyncio.Lock()
+
+    def all(self) -> list[Replica]:
+        return list(self._replicas.values())
+
+    def get(self, url: str) -> Replica | None:
+        return self._replicas.get(url)
+
+    def primaries(self) -> list[Replica]:
+        return [r for r in self._replicas.values() if r.primary]
+
+    def select(self) -> Replica | None:
+        """Pick the best healthy replica.
+
+        Priority order:
+
+        1. The first healthy primary (insertion order).
+        2. The first healthy non-primary (insertion order).
+        3. ``None`` when nothing is healthy.
+
+        This MVP picks the first match within each tier; it does NOT
+        load-balance across multiple healthy replicas of the same tier.
+        Round-robin and weighted distribution are deferred until a real
+        N-way active deployment exists.
+        """
+        healthy_primaries = [r for r in self._replicas.values() if r.primary and r.is_healthy]
+        if healthy_primaries:
+            return healthy_primaries[0]
+        healthy = [r for r in self._replicas.values() if r.is_healthy]
+        if healthy:
+            return healthy[0]
+        return None
+
+    async def record_success(
+        self,
+        replica: Replica,
+        *,
+        recovery_threshold: int,
+        latency_ms: float,
+    ) -> None:
+        async with self._lock:
+            h = replica.health
+            h.last_ok_at = time.time()
+            h.last_latency_ms = latency_ms
+            h.consecutive_failures = 0
+            h.consecutive_successes += 1
+            # State machine: UNKNOWN -> HEALTHY is immediate; only the
+            # UNHEALTHY -> HEALTHY transition is gated by recovery_threshold.
+            if h.status is ReplicaStatus.UNKNOWN:
+                logger.info("router: replica %s initial probe ok, marking HEALTHY", replica.url)
+                h.status = ReplicaStatus.HEALTHY
+                h.consecutive_successes = 0
+            elif (h.status is ReplicaStatus.UNHEALTHY and h.consecutive_successes >= recovery_threshold):
+                logger.info("router: replica %s recovered to HEALTHY after %d successes", replica.url,
+                            h.consecutive_successes)
+                h.status = ReplicaStatus.HEALTHY
+                h.consecutive_successes = 0
+
+    async def record_failure(
+        self,
+        replica: Replica,
+        *,
+        failure_threshold: int,
+        reason: str,
+    ) -> None:
+        async with self._lock:
+            h = replica.health
+            h.last_failure_at = time.time()
+            h.consecutive_successes = 0
+            h.consecutive_failures += 1
+            if (h.status is not ReplicaStatus.UNHEALTHY and h.consecutive_failures >= failure_threshold):
+                logger.warning("router: replica %s marked UNHEALTHY after %d failures: %s", replica.url,
+                               h.consecutive_failures, reason)
+                h.status = ReplicaStatus.UNHEALTHY
+
+
+async def run_health_check_loop(
+    registry: ReplicaRegistry,
+    config: RouterConfig,
+    *,
+    stop_event: asyncio.Event,
+    http_get: HttpProbe | None = None,
+) -> None:
+    """Poll all replicas' health endpoints in parallel on a fixed interval.
+
+    ``http_get`` is pluggable so unit tests can inject a deterministic
+    probe without hitting the network. The default builds a single
+    ``httpx.AsyncClient`` shared across the loop's lifetime so the
+    common case (steady polling against a stable replica set) reuses
+    TCP/TLS connections instead of paying handshake cost per probe.
+
+    Probes within one polling cycle run concurrently via ``asyncio.gather``
+    so a slow replica doesn't push the cycle past
+    ``health_check_interval_seconds``.
+    """
+    if http_get is not None:
+        await _run_loop(registry, config, stop_event, http_get)
+        return
+    async with _build_default_probe(config) as probe:
+        await _run_loop(registry, config, stop_event, probe)
+
+
+async def _run_loop(
+    registry: ReplicaRegistry,
+    config: RouterConfig,
+    stop_event: asyncio.Event,
+    http_get: Callable[..., Awaitable[tuple[float, str | None]]],
+) -> None:
+    while not stop_event.is_set():
+        replicas = registry.all()
+        results = await asyncio.gather(
+            *[
+                http_get(replica.url + config.health_check_path, timeout=config.health_check_timeout_seconds)
+                for replica in replicas
+            ],
+            return_exceptions=True,
+        )
+        for replica, result in zip(replicas, results, strict=True):
+            if isinstance(result, BaseException):
+                await registry.record_failure(
+                    replica,
+                    failure_threshold=config.failure_threshold,
+                    reason=f"{type(result).__name__}: {result}",
+                )
+                continue
+            status_ms, error = result
+            if error is None:
+                await registry.record_success(
+                    replica,
+                    recovery_threshold=config.recovery_threshold,
+                    latency_ms=status_ms,
+                )
+            else:
+                await registry.record_failure(
+                    replica,
+                    failure_threshold=config.failure_threshold,
+                    reason=error,
+                )
+        try:
+            await asyncio.wait_for(
+                stop_event.wait(),
+                timeout=config.health_check_interval_seconds,
+            )
+        except asyncio.TimeoutError:
+            continue
+
+
+@contextlib.asynccontextmanager
+async def _build_default_probe(
+    config: RouterConfig, ) -> AsyncIterator[Callable[..., Awaitable[tuple[float, str | None]]]]:
+    try:
+        import httpx
+    except ImportError as exc:  # pragma: no cover - optional extra
+        raise RuntimeError("router health checks require httpx; install with "
+                           "`pip install fastvideo[streaming]` or `pip install httpx`") from exc
+
+    async with httpx.AsyncClient(timeout=config.health_check_timeout_seconds) as client:
+
+        async def probe(url: str, *, timeout: float) -> tuple[float, str | None]:
+            start = time.perf_counter()
+            try:
+                response = await client.get(url, timeout=timeout)
+            except Exception as exc:
+                return 0.0, f"{type(exc).__name__}: {exc}"
+            latency_ms = (time.perf_counter() - start) * 1000.0
+            if response.status_code >= 400:
+                return latency_ms, f"HTTP {response.status_code}"
+            return latency_ms, None
+
+        yield probe
+
+
+__all__ = [
+    "HttpProbe",
+    "Replica",
+    "ReplicaHealth",
+    "ReplicaRegistry",
+    "ReplicaStatus",
+    "run_health_check_loop",
+]

fastvideo/entrypoints/streaming/server.py

@@ -51,6 +51,11 @@ from fastvideo.entrypoints.streaming.session import (
 )
 from fastvideo.entrypoints.streaming.session_init_image import (
     persist_session_init_image, )
+from fastvideo.entrypoints.streaming.gpu_pool import (
+    GpuPool,
+    InProcessGpuPool,
+    PoolAcquireTimeout,
+)
 from fastvideo.entrypoints.streaming.session_store import (
     InMemorySessionStore,
     SessionStore,
@@ -75,25 +80,37 @@ class _GeneratorProto(Protocol):
 @dataclass
 class ServerState:
     serve_config: ServeConfig
-    generator: _GeneratorProto
+    pool: GpuPool
     sessions: SessionManager
     session_store: SessionStore
 
 
 def build_app(
     serve_config: ServeConfig,
-    generator: _GeneratorProto,
+    generator: _GeneratorProto | None = None,
     *,
+    pool: GpuPool | None = None,
     session_store: SessionStore | None = None,
 ) -> FastAPI:
     """Build the FastAPI app used by :func:`run_server`.
 
     Exposed so tests can drive the WebSocket endpoint in-process via
     ``starlette.testclient.TestClient(app).websocket_connect(...)``.
+
+    Exactly one of ``generator`` (backed by :class:`InProcessGpuPool`)
+    or ``pool`` (for the subprocess-backed production shape) must be
+    given.
     """
     if serve_config.streaming is None:
         raise ValueError("ServeConfig.streaming must be set to launch the streaming "
                          "server; got None. Add a `streaming:` block to your serve config.")
+    if (generator is None) == (pool is None):
+        raise ValueError("build_app requires exactly one of `generator` or `pool`")
+
+    store = session_store or InMemorySessionStore()
+    if pool is None:
+        assert generator is not None
+        pool = InProcessGpuPool(generator, session_store=store)
 
     sessions = SessionManager(
         segment_cap=serve_config.streaming.generation_segment_cap,
@@ -101,9 +118,9 @@ def build_app(
     )
     state = ServerState(
         serve_config=serve_config,
-        generator=generator,
+        pool=pool,
         sessions=sessions,
-        session_store=session_store or InMemorySessionStore(),
+        session_store=store,
     )
 
     app = FastAPI(title="FastVideo Streaming")
@@ -135,6 +152,8 @@ def build_app(
             with contextlib.suppress(InvalidSessionTransition):
                 session.transition(SessionState.ERROR)
         finally:
+            with contextlib.suppress(Exception):
+                await state.pool.release(session.id)
             _cleanup_session(session, state)
 
     app.state.server_state = state
@@ -178,10 +197,22 @@ async def _handle_session(
     await _apply_session_init(session, init, state)
     await _send_json(websocket, QueueStatus(position=0, queue_depth=0))
     session.transition(SessionState.GPU_BINDING)
-    await _send_json(websocket, GpuAssigned(
-        gpu_id=0,
-        session_timeout=state.sessions.session_timeout_seconds,
-    ))
+    try:
+        assignment = await state.pool.acquire(
+            session.id,
+            timeout=float(state.sessions.session_timeout_seconds),
+        )
+    except PoolAcquireTimeout as exc:
+        await _send_error(websocket, "gpu_unavailable", str(exc), retryable=True)
+        with contextlib.suppress(InvalidSessionTransition):
+            session.transition(SessionState.TIMEOUT)
+        return
+    session.gpu_id = assignment.gpu_id
+    await _send_json(websocket,
+                     GpuAssigned(
+                         gpu_id=assignment.gpu_id,
+                         session_timeout=state.sessions.session_timeout_seconds,
+                     ))
     session.transition(SessionState.ACTIVE)
     await _send_json(websocket, _build_stream_start(session, state))
 
@@ -327,15 +358,13 @@ async def _run_segment(
         ))
 
     start = time.perf_counter()
-    loop = asyncio.get_running_loop()
-    # TODO: executor-wrapped generate() cannot be cancelled, so a
-    # client disconnect mid-segment leaves the GPU work running to
-    # completion. Real cancellation needs the generate_async API.
+    # TODO: pool.run() runs to completion even if the client disconnects
+    # mid-segment. Real cancellation needs the generate_async API.
     try:
-        result = await loop.run_in_executor(None, state.generator.generate, request)
+        result = await state.pool.run(session.id, request)
     except Exception as exc:
-        logger.exception("session %s: generator failed", session.id[:8])
-        await _send_error(websocket, "worker_failed", f"generator.generate failed: {exc}", retryable=True)
+        logger.exception("session %s: pool.run failed", session.id[:8])
+        await _send_error(websocket, "worker_failed", f"pool.run failed: {exc}", retryable=True)
         with contextlib.suppress(InvalidSessionTransition):
             session.transition(SessionState.ERROR)
         return

fastvideo/entrypoints/streaming/session_logger.py

@@ -0,0 +1,113 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Per-session JSONL event logger.
+
+Each session gets its own JSONL file under the configured log root so
+post-hoc analytics (enhancer latency, GPU assignment, segment timings)
+can be recovered without a tracing backend. The internal UI uses this
+format; keeping the same shape makes log tooling portable.
+"""
+from __future__ import annotations
+
+import contextlib
+import json
+import os
+import re
+import threading
+import time
+from dataclasses import dataclass, field
+from typing import Any, TextIO
+
+_FILENAME_SANITIZE_RE = re.compile(r"[^A-Za-z0-9._-]")
+
+
+@dataclass
+class SessionLogEvent:
+    """One line in the session JSONL file."""
+
+    session_id: str
+    event: str
+    payload: dict[str, Any] = field(default_factory=dict)
+    ts: float = field(default_factory=time.time)
+
+
+class SessionLogger:
+    """Append-only JSONL logger keyed by session id.
+
+    Thread-safe; the server may be writing from multiple asyncio tasks
+    (fMP4 encoder thread + control-frame handler) for the same session.
+    """
+
+    def __init__(self, log_dir: str | None) -> None:
+        self._log_dir = log_dir
+        self._files: dict[str, TextIO] = {}
+        self._locks: dict[str, threading.Lock] = {}
+        self._registry_lock = threading.Lock()
+        self._ensure_dir()
+
+    def log(self, event: SessionLogEvent) -> None:
+        if self._log_dir is None:
+            return
+        opened = self._get_file(event.session_id)
+        if opened is None:
+            return
+        handle, lock = opened
+        line = json.dumps({
+            "session_id": event.session_id,
+            "event": event.event,
+            "ts": event.ts,
+            "payload": event.payload,
+        })
+        with lock, contextlib.suppress(ValueError):
+            handle.write(line + "\n")
+            handle.flush()
+
+    def close(self, session_id: str) -> None:
+        with self._registry_lock:
+            handle = self._files.pop(session_id, None)
+            lock = self._locks.pop(session_id, None)
+        if handle is None or lock is None:
+            return
+        with lock, contextlib.suppress(Exception):
+            handle.close()
+
+    def close_all(self) -> None:
+        with self._registry_lock:
+            sids = list(self._files)
+        for sid in sids:
+            self.close(sid)
+
+    def _ensure_dir(self) -> None:
+        if self._log_dir is None:
+            return
+        os.makedirs(self._log_dir, exist_ok=True)
+
+    def _get_file(self, session_id: str) -> tuple[TextIO, threading.Lock] | None:
+        if self._log_dir is None:
+            return None
+        with self._registry_lock:
+            handle = self._files.get(session_id)
+            lock = self._locks.get(session_id)
+            if handle is not None and lock is not None:
+                return handle, lock
+            # Defense-in-depth: session_id is server-generated UUID today,
+            # but sanitize against path traversal in case future code paths
+            # allow client-supplied ids.
+            safe_id = _FILENAME_SANITIZE_RE.sub("_", session_id) or "unknown"
+            path = os.path.join(
+                self._log_dir,
+                f"session-{safe_id}.jsonl",
+            )
+            try:
+                handle = open(path, "a", encoding="utf-8")  # noqa: SIM115
+            except OSError:
+                return None
+            lock = threading.Lock()
+            self._files[session_id] = handle
+            self._locks[session_id] = lock
+            return handle, lock
+
+
+__all__ = [
+    "SessionLogEvent",
+    "SessionLogger",
+]

fastvideo/entrypoints/streaming/worker.py

@@ -0,0 +1,133 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Per-GPU worker subprocess entry for :class:`SubprocessGpuPool`.
+
+The pool manages binding, lifecycle, and message dispatch in the parent
+process. The worker constructs its :class:`VideoGenerator` from a typed
+:class:`GeneratorConfig`, runs the two-segment warmup so both
+initial-segment and continuation-branch compile graphs are hot, and
+then loops on the job queue.
+"""
+from __future__ import annotations
+
+import multiprocessing as mp
+import queue
+from typing import Any
+
+from fastvideo.api.schema import (
+    GeneratorConfig,
+    GenerationRequest,
+    InputConfig,
+    OutputConfig,
+    SamplingConfig,
+    WarmupConfig,
+)
+from fastvideo.logger import init_logger
+
+logger = init_logger(__name__)
+
+# Synthetic warmup dimensions: small enough to keep boot fast, big enough
+# to exercise the real shape-dependent compile paths. Keep in sync with
+# WarmupConfig if these become user-tunable.
+_WARMUP_NUM_FRAMES = 8
+_WARMUP_HEIGHT = 256
+_WARMUP_WIDTH = 256
+_WARMUP_NUM_INFERENCE_STEPS = 1
+
+
+def worker_main(
+    *,
+    gpu_id: int,
+    worker_id: str,
+    generator_config: GeneratorConfig,
+    warmup_config: WarmupConfig,
+    job_queue: mp.Queue,
+    result_queue: mp.Queue,
+    shutdown_event: Any,
+) -> None:  # pragma: no cover - exercised via integration only
+    """Per-worker subprocess entry.
+
+    Runs inside the child spawned by ``SubprocessGpuPool``. Blocking
+    ``VideoGenerator`` construction + generation happens here, not in
+    the parent's event loop.
+    """
+    import os
+
+    os.environ["CUDA_VISIBLE_DEVICES"] = str(gpu_id)
+    try:
+        from fastvideo import VideoGenerator
+
+        generator = VideoGenerator.from_pretrained(config=generator_config)
+        if warmup_config.enabled:
+            _warmup_worker(generator, warmup_config)
+        result_queue.put({"kind": "ready", "worker_id": worker_id})
+    except Exception as exc:
+        result_queue.put({"kind": "error", "error": repr(exc)})
+        return
+
+    while not shutdown_event.is_set():
+        try:
+            item = job_queue.get(timeout=0.5)
+        except queue.Empty:
+            continue
+        if item is None:
+            break
+        job_id = item["job_id"]
+        request = item["request"]
+        try:
+            result = generator.generate(request)
+            result_queue.put({
+                "kind": "result",
+                "job_id": job_id,
+                "result": result,
+            })
+        except Exception as exc:
+            result_queue.put({
+                "kind": "error",
+                "job_id": job_id,
+                "error": repr(exc),
+            })
+
+
+def _warmup_worker(
+    generator: Any,
+    warmup_config: WarmupConfig,
+) -> None:
+    """Run two synthetic generations so both compile branches are primed.
+
+    Segment 1 is a fresh start (no continuation state) and exercises
+    the initial-segment graph. Segment 2 feeds segment 1's continuation
+    state back in so the conditioning branch is also compiled before
+    the first user request lands.
+    """
+    sampling = SamplingConfig(
+        num_frames=_WARMUP_NUM_FRAMES,
+        height=_WARMUP_HEIGHT,
+        width=_WARMUP_WIDTH,
+        num_inference_steps=_WARMUP_NUM_INFERENCE_STEPS,
+    )
+    seg1 = GenerationRequest(
+        prompt=warmup_config.prompt,
+        sampling=sampling,
+        inputs=InputConfig(),
+        output=OutputConfig(save_video=False, return_frames=False, return_state=True),
+    )
+    seg1_result = generator.generate(seg1)
+
+    seg2 = GenerationRequest(
+        prompt=warmup_config.prompt,
+        sampling=sampling,
+        inputs=InputConfig(),
+        output=OutputConfig(save_video=False, return_frames=False),
+        state=_extract_continuation_state(seg1_result),
+    )
+    generator.generate(seg2)
+
+
+def _extract_continuation_state(result: Any) -> Any:
+    state = getattr(result, "state", None)
+    if state is None and isinstance(result, dict):
+        state = result.get("state")
+    return state
+
+
+__all__ = ["worker_main"]

fastvideo/entrypoints/video_generator.py

@@ -65,6 +65,7 @@ _FROM_PRETRAINED_CONVENIENCE_KWARGS = frozenset({
     "pin_cpu_memory",
     "enable_torch_compile",
     "torch_compile_kwargs",
+    "output_type",
 })
 
 
@@ -601,10 +602,20 @@ class VideoGenerator:
         thread = threading.Thread(target=execute_forward_thread)
         thread.start()
         latent_batch_size = _infer_latent_batch_size(batch)
-        samples = torch.empty(
-            (latent_batch_size, 3, sampling_param.num_frames, sampling_param.height, sampling_param.width),
-            device='cpu',
-            pin_memory=fastvideo_args.pin_cpu_memory)
+        # When ``output_type == "latent"`` the forward output has latent
+        # shape (e.g. ``[B, C_latent, T_latent, H_latent, W_latent]``)
+        # rather than the pre-allocation's pixel shape. Skip the pinned
+        # ~50 MB buffer entirely; we always fall through to the
+        # ``samples = output_batch.output.cpu()`` branch below in that
+        # mode. ``skip_pixel_prealloc`` also gates the slow-path warning.
+        skip_pixel_prealloc = fastvideo_args.output_type == "latent"
+        if skip_pixel_prealloc:
+            samples = torch.empty(0, device='cpu')
+        else:
+            samples = torch.empty(
+                (latent_batch_size, 3, sampling_param.num_frames, sampling_param.height, sampling_param.width),
+                device='cpu',
+                pin_memory=fastvideo_args.pin_cpu_memory)
         thread.join()
 
         if thread_error["error"] is not None:
@@ -619,30 +630,61 @@ class VideoGenerator:
         if output_batch.output.shape == samples.shape:
             samples.copy_(output_batch.output)
         else:
-            logger.warning("Output shape %s does not match expected shape %s; use slow path", output_batch.output.shape,
-                           samples.shape)
+            if not skip_pixel_prealloc:
+                logger.warning("Output shape %s does not match expected shape %s; use slow path",
+                               output_batch.output.shape, samples.shape)
             samples = output_batch.output.cpu()
         logging_info = output_batch.logging_info
 
         gen_time = time.perf_counter() - start_time
         logger.info("Generated successfully in %.2f seconds", gen_time)
 
-        # Process outputs
-        videos = rearrange(samples, "b c t h w -> t b c h w")
-        frames = []
-        for x in videos:
-            x = torchvision.utils.make_grid(x, nrow=6)
-            x = x.permute(1, 2, 0).squeeze(-1)
-            x = (x * 255).to(torch.uint8)
-            frames.append(x.cpu().numpy())
+        # Three mutually-exclusive output modes determine whether (a) we
+        # build an RGB frame buffer and (b) what file we write to disk:
+        #
+        #   1. `output_type == "latent"` — VAE is bypassed in DecodingStage
+        #      and `samples` holds raw latents (arbitrary channel count).
+        #      The RGB grid / uint8 / mp4 / png pipeline below cannot
+        #      consume those, so we skip it entirely and let callers work
+        #      with the latent tensor directly via `result["samples"]`.
+        #   2. Audio-only workload — `samples` is a 1×3×1×8×8 placeholder
+        #      no caller will use; skip the grid loop and save a `.wav`.
+        #   3. Pixel video / image — the historical happy path.
+        is_latent_output = fastvideo_args.output_type == "latent"
+        audio_only = bool(output_batch.extra.get("audio_only"))
 
-        # Save output if requested
-        if batch.save_video:
-            if self._is_image_workload():
+        frames: list[np.ndarray] | None
+        if is_latent_output or audio_only:
+            frames = None if is_latent_output else []
+        else:
+            videos = rearrange(samples, "b c t h w -> t b c h w")
+            frames = []
+            for x in videos:
+                x = torchvision.utils.make_grid(x, nrow=6)
+                x = x.permute(1, 2, 0).squeeze(-1)
+                x = (x * 255).to(torch.uint8)
+                frames.append(x.cpu().numpy())
+
+        save_to_disk = batch.save_video and not is_latent_output
+        if save_to_disk:
+            if audio_only:
+                # Audio-only workload: write a standalone .wav rather than
+                # muxing the audio into a placeholder mp4 (which forces
+                # ffmpeg to round 8x8 placeholder frames up to 16x16).
+                output_path = self._rewrite_extension(output_path, ".wav")
+                self._write_pcm_wav(
+                    output_path,
+                    output_batch.extra["audio"],
+                    int(output_batch.extra["audio_sample_rate"]),
+                )
+                logger.info("Saved audio to %s", output_path)
+            elif self._is_image_workload():
                 # Image workloads (t2i, i2i, …): save the first frame as PNG.
+                assert frames is not None  # implied by save_to_disk and not audio_only
                 imageio.imwrite(output_path, frames[0])
                 logger.info("Saved image to %s", output_path)
             else:
+                assert frames is not None  # implied by save_to_disk and not audio_only
                 imageio.mimsave(output_path, frames, fps=batch.fps, format="mp4")
                 logger.info("Saved video to %s", output_path)
                 audio = output_batch.extra.get("audio")
@@ -655,14 +697,18 @@ class VideoGenerator:
             "prompts": prompt,
             "samples": samples if batch.return_frames else None,
             "frames": frames if batch.return_frames else None,
-            "audio": output_batch.extra.get("audio") if batch.return_frames else None,
+            # Audio is the primary output for audio workloads — return it
+            # whenever the pipeline produced one, regardless of
+            # `return_frames` (which gates the video-shaped buffers).
+            "audio": output_batch.extra.get("audio"),
+            "audio_sample_rate": output_batch.extra.get("audio_sample_rate"),
             "size": (target_height, target_width, batch.num_frames),
             "generation_time": gen_time,
             "logging_info": logging_info,
             "trajectory": output_batch.trajectory_latents,
             "trajectory_timesteps": output_batch.trajectory_timesteps,
             "trajectory_decoded": output_batch.trajectory_decoded,
-            "video_path": output_path if batch.save_video else None,
+            "video_path": output_path if save_to_disk else None,
             "peak_memory_mb": output_batch.extra.get("peak_memory_mb"),
         }
 
@@ -683,7 +729,55 @@ class VideoGenerator:
         return result.to_legacy_dict()
 
     @staticmethod
+    def _rewrite_extension(path: str, new_ext: str) -> str:
+        root, old_ext = os.path.splitext(path)
+        new_path = root + new_ext
+        if old_ext and old_ext.lower() != new_ext.lower():
+            logger.info("Rewriting output extension %s -> %s.", old_ext, new_ext)
+        return new_path
+
+    @staticmethod
+    def _audio_to_int16(audio: torch.Tensor | np.ndarray, ) -> tuple[np.ndarray, int]:
+        """Normalize `[samples]` / `[samples, channels]` / `[channels,
+        samples]` audio in roughly [-1, 1] to a `(int16 [samples,
+        channels], num_channels)` pair. Raises `ValueError` for shapes
+        we can't classify.
+        """
+        if torch.is_tensor(audio):
+            audio_np = audio.detach().cpu().float().numpy()
+        else:
+            audio_np = np.asarray(audio, dtype=np.float32)
+        if audio_np.ndim == 1:
+            audio_np = audio_np[:, None]
+        elif audio_np.ndim == 2:
+            if audio_np.shape[0] <= 8 and audio_np.shape[1] > audio_np.shape[0]:
+                audio_np = audio_np.T
+        else:
+            raise ValueError(f"Unexpected audio shape {audio_np.shape}.")
+        audio_np = np.clip(audio_np, -1.0, 1.0)
+        audio_int16 = (audio_np * 32767.0).astype(np.int16)
+        return audio_int16, audio_int16.shape[1]
+
+    @classmethod
+    def _write_pcm_wav(
+        cls,
+        wav_path: str,
+        audio: torch.Tensor | np.ndarray,
+        sample_rate: int,
+    ) -> int:
+        """Write 16-bit PCM WAV; returns the channel count."""
+        import wave
+        audio_int16, num_channels = cls._audio_to_int16(audio)
+        with wave.open(wav_path, "wb") as f:
+            f.setnchannels(num_channels)
+            f.setsampwidth(2)
+            f.setframerate(sample_rate)
+            f.writeframes(audio_int16.tobytes())
+        return num_channels
+
+    @classmethod
     def _mux_audio(
+        cls,
         video_path: str,
         audio: torch.Tensor | np.ndarray,
         sample_rate: int,
@@ -693,40 +787,16 @@ class VideoGenerator:
             import av
         except ImportError:
             logger.warning("PyAV not installed; cannot mux audio. "
-                           "Install with: pip install av")
+                           "Install with: uv pip install av")
             return False
 
-        if torch.is_tensor(audio):
-            audio_np = audio.detach().cpu().float().numpy()
-        else:
-            audio_np = np.asarray(audio, dtype=np.float32)
-
-        if audio_np.ndim == 1:
-            audio_np = audio_np[:, None]
-        elif audio_np.ndim == 2:
-            if audio_np.shape[0] <= 8 and audio_np.shape[1] > audio_np.shape[0]:
-                audio_np = audio_np.T
-        else:
-            logger.warning("Unexpected audio shape %s; skipping mux.", audio_np.shape)
-            return False
-
-        audio_np = np.clip(audio_np, -1.0, 1.0)
-        audio_int16 = (audio_np * 32767.0).astype(np.int16)
-        num_channels = audio_int16.shape[1]
-        layout = "stereo" if num_channels == 2 else "mono"
-
         try:
-            import wave
             with tempfile.TemporaryDirectory() as tmpdir:
                 out_path = os.path.join(tmpdir, "muxed.mp4")
                 wav_path = os.path.join(tmpdir, "audio.wav")
 
-                # Write audio to WAV file
-                with wave.open(wav_path, "wb") as wav_file:
-                    wav_file.setnchannels(num_channels)
-                    wav_file.setsampwidth(2)
-                    wav_file.setframerate(sample_rate)
-                    wav_file.writeframes(audio_int16.tobytes())
+                num_channels = cls._write_pcm_wav(wav_path, audio, sample_rate)
+                layout = "stereo" if num_channels == 2 else "mono"
 
                 # Open input video and audio
                 input_video = av.open(video_path)

fastvideo/eval/README.md

@@ -1,214 +0,0 @@
-# `fastvideo.eval`
-
-In-process evaluation suite for video generations. Includes pixel
-metrics (SSIM, PSNR, LPIPS), optical-flow comparisons, the full VBench
-suite, Physics-IQ, and a VLM scorer (VideoScore-2) behind a single
-registry-driven API.
-
-## Install
-
-| Use case | Install |
-|---|---|
-| Default (common, optical_flow, vbench-light, physics_iq, videoscore2) | `uv pip install -e .[eval]` |
-| Just VBench (12 of 16 sub-metrics) | `uv pip install -e .[eval-vbench]` |
-| Just Physics-IQ (covered by `[eval]`) | `uv pip install -e .[eval-physics-iq]` |
-| Plus `vbench.scene` (AVoCaDO) | `uv pip install -e .[eval-full]` |
-| Plus `vbench.{color, multiple_objects, object_class, spatial_relationship}` (GRiT) | `uv pip install -e .[eval-vbench]` then `uv pip install --no-build-isolation 'git+https://github.com/facebookresearch/detectron2.git'` |
-
-To use VBench, also pull the upstream submodule:
-
-```bash
-git submodule update --init --recursive  # fetches vbench + kernel deps
-```
-
-The submodule is a clean upstream pin. Compat with current
-transformers/numpy/timm versions is applied at import time in
-`fastvideo/eval/metrics/vbench/__init__.py` via attribute-level
-monkey-patches; the submodule files are unchanged.
-
-## Public API
-
-```python
-from fastvideo.eval import (
-    create_evaluator,        # build a reusable Evaluator
-    evaluate,                # one-shot helper
-    Evaluator,               # the class itself
-    BaseMetric, MetricResult,
-    register, list_metrics, get_metric,
-    ensure_checkpoint, get_cache_dir,
-)
-
-ev = create_evaluator(metrics=["common.ssim", "vbench.aesthetic_quality"],
-                      device="cuda")
-scores = ev.evaluate(video=tensor, reference=ref, fps=8.0)
-```
-
-`evaluate` accepts either a pre-loaded `(T, C, H, W)` tensor or a path
-string for `video` and `reference`. Paths are decoded inside the worker
-that picks up the sample, so peak memory stays bounded by `num_gpus`
-when scoring large batches.
-
-### CLI
-
-```bash
-fastvideo eval list                              # list registered metrics
-fastvideo eval list --group vbench               # filter by group
-fastvideo eval run --videos clip.mp4 \
-    --metrics vbench.aesthetic_quality \
-    --output scores.json
-fastvideo eval run --videos generated/*.mp4 \
-    --reference reference/ \
-    --metrics common.ssim,common.lpips
-```
-
-### Generate-then-score example
-
-`examples/inference/eval/eval_ltx2_vbench.py` runs an LTX2 prompt
-through `VideoGenerator` and scores the resulting mp4 with
-`vbench.aesthetic_quality` and `vbench.subject_consistency`. Use it as
-a template for end-to-end "generate then score" pipelines.
-
-## Layout
-
-```
-fastvideo/
-├── eval/
-│   ├── api.py, evaluator.py, registry.py, models.py, ...
-│   ├── io/                        # video loading helpers
-│   ├── datasets/                  # prompt corpora (vbench, physics_iq)
-│   └── metrics/
-│       ├── base.py                # BaseMetric + @register contract
-│       ├── common/                # SSIM, PSNR, LPIPS
-│       ├── optical_flow/          # gt_optical_flow, synthetic_optical_flow
-│       ├── videoscore2/           # VideoScore-2 (Qwen2.5-VL)
-│       ├── physics_iq/            # PhysicsIQ + sub-metrics
-│       └── vbench/                # adapter: sys.path bootstrap + shims
-│           ├── __init__.py
-│           └── <16 sub-metric pkgs>
-└── third_party/
-    └── eval/
-        └── vbench/                # git submodule (Vchitect/VBench)
-```
-
-### Prompt datasets
-
-```python
-from fastvideo.eval.datasets import get_dataset, list_datasets
-
-list_datasets()                    # ['physics_iq', 'vbench']
-
-ds = get_dataset("physics_iq", limit=4)   # auto-fetches assets on first miss
-for row in ds:
-    # row contains 'prompt', 'reference', 'reference_take2', and
-    # metric-specific aux fields. Drop straight into Evaluator.evaluate(**row).
-    ...
-```
-
-The Physics-IQ manifest CSV is vendored at
-`fastvideo/eval/metrics/physics_iq/_vendored/descriptions.csv`.
-Per-scenario videos, masks, and switch-frames auto-fetch on first use
-into `${FASTVIDEO_EVAL_CACHE}/datasets/physics_iq/`. For air-gapped
-runs, pass `auto_download=False` or `dataset_root=` a pre-downloaded
-copy. Set `FASTVIDEO_PHYSICS_IQ_BUCKET_URL` to redirect the fetch to
-an internal mirror.
-
-## Adding a new metric
-
-The full porting guide is at
-[`docs/contributing/eval-metrics.md`](../../docs/contributing/eval-metrics.md).
-Summary below.
-
-### Native metric (no submodule)
-
-```python
-# fastvideo/eval/metrics/common/<your_metric>/metric.py
-from fastvideo.eval.metrics.base import BaseMetric
-from fastvideo.eval.registry import register
-from fastvideo.eval.types import MetricResult
-
-
-@register("common.your_metric")
-class YourMetric(BaseMetric):
-    name = "common.your_metric"
-    requires_reference = True
-    needs_gpu = False
-    dependencies: list[str] = []  # e.g. ["pyiqa"] if relevant
-
-    def compute(self, sample) -> list[MetricResult]:
-        ...
-```
-
-The metric is auto-discovered by `fastvideo/eval/metrics/__init__.py`,
-which walks all non-underscore subdirectories and imports their
-`metric` module.
-
-### Wrapping upstream code via a submodule
-
-See `fastvideo/eval/metrics/vbench/` for a worked example. The
-contract is:
-
-1. Upstream lives as a git submodule under
-   `fastvideo/third_party/eval/<bench>/`, pinned to a SHA in repo-root
-   `.gitmodules`.
-2. The metric package's `__init__.py`
-   (`fastvideo/eval/metrics/<bench>/__init__.py`) inserts that
-   submodule path on `sys.path` and installs any compat shims for
-   modern torch/transformers/numpy. Do not modify upstream files on
-   disk.
-3. Per-sub-metric `metric.py` files use `@register("<bench>.<name>")`.
-
-Patches live as Python in the metric's `__init__.py` so they are
-grep-able and reviewable.
-
-## Caches
-
-Eval cache root: `${FASTVIDEO_CACHE_ROOT}/eval/`, default
-`~/.cache/fastvideo/eval/`. Override with `FASTVIDEO_EVAL_CACHE`.
-
-```
-${FASTVIDEO_CACHE_ROOT}/eval/
-├── models/      # URL-fetched checkpoints (LAION head, AMT, GRiT)
-├── torch/       # redirected TORCH_HOME (DINO via torch.hub, lpips)
-├── clip/        # passed as download_root= to clip.load callsites
-└── datasets/    # auto-fetched dataset assets, one subdir per benchmark
-                 # (e.g. datasets/physics_iq/{split-videos,switch-frames,...})
-```
-
-HF-hosted models stay in HF's default cache
-(`~/.cache/huggingface/hub/`) so they dedupe with other ML projects on
-the same host.
-
-### Convention for new metrics
-
-If your metric wraps a third-party loader that has its own cache
-directory, route it through `get_cache_dir()` so users get one knob
-to redirect everything.
-
-```python
-# CLIP: pass download_root explicitly
-import clip
-from fastvideo.eval.models import get_cache_dir
-model, _ = clip.load("ViT-B/32", device=device,
-                     download_root=str(get_cache_dir() / "clip"))
-
-# torch.hub is already redirected by fastvideo.eval.__init__ via
-# TORCH_HOME; no per-callsite work needed.
-
-# transformers / huggingface_hub: leave alone. HF's default cache is
-# shared with other tools.
-```
-
-For libraries that do not honour any env var or kwarg (pyiqa, funasr),
-their cache lands in the library's own dir. Document the exception in
-the metric's docstring if it matters.
-
-## Out of scope (follow-up PRs)
-
-- **MIND** metrics. Depend on a separate `vipe` upstream submodule.
-- **VBench-2.0**. Sibling vbench2 package; needs its own port.
-- **FVD as a registered metric**. Currently still at `benchmarks/fvd/`.
-  FVD is a set-vs-set distribution distance and does not fit the
-  per-sample `BaseMetric.compute` API without a stateful accumulator;
-  conversion is a designed follow-up.
-- **Training-time eval callback** (`EvalCallback`) and the
-  `RolloutEvaluator` helper.

fastvideo/eval/__init__.py

@@ -1,45 +0,0 @@
-from fastvideo.eval.models import ensure_checkpoint, get_cache_dir
-
-
-def _redirect_third_party_caches() -> None:
-    """Point libraries that respect env vars at the eval cache root.
-
-    Run before metric modules import torch.hub (and friends) so the
-    redirect actually takes effect. We intentionally leave ``HF_HOME``
-    alone — HF's default cache (``~/.cache/huggingface/hub``) is widely
-    shared with other ML projects, and isolating it for eval would force
-    users to re-download already-cached transformers weights.
-
-    Other libraries with non-standard caches (CLIP, pyiqa) don't honour
-    env vars at all; their callsites in metric.py files pass
-    ``download_root=str(get_cache_dir() / "<library>")`` directly.
-    """
-    import os
-    root = get_cache_dir()
-    os.environ.setdefault("TORCH_HOME", str(root / "torch"))
-
-
-_redirect_third_party_caches()
-
-from fastvideo.eval.types import MetricResult, Video  # noqa: E402
-from fastvideo.eval.metrics.base import BaseMetric  # noqa: E402
-from fastvideo.eval.registry import register, list_metrics, get_metric  # noqa: E402
-from fastvideo.eval.api import evaluate  # noqa: E402
-from fastvideo.eval.evaluator import Evaluator, create_evaluator  # noqa: E402
-
-# Trigger metric auto-discovery
-import fastvideo.eval.metrics  # noqa: F401, E402
-
-__all__ = [
-    "evaluate",
-    "Evaluator",
-    "create_evaluator",
-    "MetricResult",
-    "Video",
-    "BaseMetric",
-    "register",
-    "list_metrics",
-    "get_metric",
-    "ensure_checkpoint",
-    "get_cache_dir",
-]

fastvideo/eval/api.py

@@ -1,36 +0,0 @@
-from __future__ import annotations
-
-from pathlib import Path
-
-import torch
-
-from fastvideo.eval.evaluator import create_evaluator
-from fastvideo.eval.types import MetricResult
-
-
-def evaluate(
-    generated: torch.Tensor | str | Path,
-    reference: torch.Tensor | str | Path | None = None,
-    metrics: list[str] | str = "all",
-    device: str = "cuda",
-    **kwargs,
-) -> dict[str, MetricResult] | list[dict[str, MetricResult]]:
-    """One-shot evaluation. For repeated use, prefer :func:`create_evaluator`.
-
-    Parameters
-    ----------
-    generated : Tensor | str | Path
-        Generated video. Either a pre-loaded ``(T, C, H, W)`` tensor or a
-        path to an mp4/avi/etc. — paths are decoded by the worker.
-    reference : Tensor | str | Path | None
-        Reference video (same accepted shapes as *generated*).
-    metrics : list[str] | str
-        Metric names, or ``"all"``.
-    device : str
-        PyTorch device string.
-    """
-    ev = create_evaluator(metrics=metrics, device=device)
-    kw: dict = {"video": generated, **kwargs}
-    if reference is not None:
-        kw["reference"] = reference
-    return ev.evaluate(**kw)

fastvideo/eval/datasets/__init__.py

@@ -1,51 +0,0 @@
-"""Prompt-corpus datasets for end-to-end benchmark evaluation.
-
-Public API mirrors :mod:`fastvideo.eval` (metrics side):
-
-    from fastvideo.eval.datasets import (
-        PromptDataset, Sample,
-        register_dataset, get_dataset, list_datasets,
-    )
-
-A dataset is an iterable of plain dicts (one per sample). Built-in
-datasets self-register at import time. To add one, drop a module into
-this package that subclasses :class:`PromptDataset` and decorates with
-``@register_dataset("name")`` — auto-discovery picks it up.
-"""
-from fastvideo.eval.datasets.base import (BasePromptDataset, PromptDataset, Sample)
-from fastvideo.eval.datasets.registry import (get_dataset, list_datasets, register_dataset)
-
-
-def _autodiscover() -> None:
-    """Import every non-underscore .py module / subpackage in this package
-    so the ``@register_dataset`` decorators fire."""
-    import importlib
-    import os
-
-    for entry in os.listdir(os.path.dirname(__file__)):
-        if entry.startswith("_") or entry.startswith("."):
-            continue
-        if entry in {"base.py", "registry.py"}:
-            continue
-        if entry.endswith(".py"):
-            importlib.import_module(f"{__name__}.{entry[:-3]}")
-        elif os.path.isdir(os.path.join(os.path.dirname(__file__), entry)) \
-                and os.path.exists(os.path.join(
-                    os.path.dirname(__file__), entry, "__init__.py")):
-            importlib.import_module(f"{__name__}.{entry}")
-
-
-_autodiscover()
-
-# Re-export the canonical class for typed imports.
-from fastvideo.eval.datasets.vbench import VBenchPromptDataset  # noqa: E402
-
-__all__ = [
-    "PromptDataset",
-    "BasePromptDataset",
-    "Sample",
-    "register_dataset",
-    "get_dataset",
-    "list_datasets",
-    "VBenchPromptDataset",
-]