[docs] cosmos3: record real-weights E2E acceptance + scheduler fix (I003)

tests/local_tests/cosmos3/PORT_STATUS.md

@@ -13,10 +13,10 @@
 
 ## Current Phase
 
-- phase: `PR1 (video core) COMPLETE: DiT / VAE / scheduler / strict-load / sequence-packing / pipeline all framework-parity verified (suite 83 passed, 0 skipped). Remaining acceptance: real-weights (33 GB, GPU) end-to-end generation vs framework. Then PR2 audio / PR3 action / PR4 reasoning.`
+- phase: `PR1 (video core) COMPLETE + real-weights E2E generation VERIFIED: DiT / VAE / scheduler / strict-load / sequence-packing / pipeline all framework-parity verified (suite 95 passed, 0 skipped). Real-weights T2V on B200 (1280x704, 29f, 35 steps) produces coherent video matching the prompt. Next: PR2 audio / PR3 action / PR4 reasoning.`
 - status: `in_progress`
 - owner: `orchestrator`
-- last_updated: `2026-06-06`
+- last_updated: `2026-06-07`
 - env: `fv-cosmos3` (conda clone of fv-main; `fastvideo` editable repointed to this worktree). Run tests from the worktree cwd with this env's python.
 - branch: rebased onto `origin/main` @ `1c627a3f9` (was 33 behind, merge-base 2026-05-22); now 6 commits ahead; `fastvideo` imports clean; Tier-A `13 passed, 2 skipped`.
 
@@ -26,7 +26,7 @@
 |---|---|---|---|---|---|---|---|---|---|
 | transformer | dit | port | `diffusers_cosmos3/transformer.py:Cosmos3OmniTransformer` (model_type `qwen3_vl_text`, MoT + MRoPE) | `model_index.json: transformer`; `cosmos_framework/model/vfm/mot/cosmos3_vfm_network.py`, `omni_mot_model.py` | `fastvideo/models/dits/cosmos3.py` (branch: `Cosmos3VFMTransformer`+`Cosmos3LanguageModel` — reconcile to `Cosmos3OmniTransformer`) | skeleton | not_started | scaffold_skip | I001 |
 | vae | vae | reuse | diffusers `AutoencoderKLWan` | `model_index.json: vae` | reuse Wan VAE (`fastvideo/models/vaes/`, cf. `cosmos25wanvae.py`) | not_started | passthrough? | not_started | Q001 |
-| scheduler | generic | reuse | diffusers `UniPCMultistepScheduler` | `model_index.json: scheduler` | reuse | not_started | n/a | scaffold_skip (`test_cosmos3_scheduler_default_parity`) | - |
+| scheduler | generic | reuse (flow-coerced) | framework `FlowUniPCMultistepScheduler` (`cosmos_framework/.../fm_solvers_unipc.py`; checkpoint ships diffusers-style config) | `model_index.json: scheduler`; `cosmos_framework/.../samplers/unipc.py:UniPCSampler` | FastVideo-native `UniPCMultistepScheduler` (flow config), coerced in `initialize_pipeline` | done | n/a | framework-parity DONE (`test_cosmos3_scheduler_parity`: timesteps bit-exact, sigmas ~1e-8, trajectory <~1e-6) | I003 (resolved) |
 | text_tokenizer | tokenizer | reuse | transformers `Qwen2TokenizerFast` | `model_index.json: text_tokenizer` | reuse (tokenizer = allowed third-party) | not_started | passthrough | scaffold_skip (`test_cosmos3_tokenizer_chat_template`) | - |
 | vision_encoder | encoder | port | transformers `Qwen3VLVisionModel` | `model_index.json: vision_encoder` | new encoder bucket OR documented lazy-wrapper | not_started | not_started | not_started | Q002 |
 | sound_tokenizer | generic/vae | port | `Cosmos3AVAEAudioTokenizer` (model_type `autoencoder_v2`) | `model_index.json: sound_tokenizer` | new audio component | not_started | not_started | not_started | Q003 |
@@ -63,6 +63,7 @@
 |---|---|---|---|---|---|---|---|---|
 | I001 | port | transformer | high | Branch DiT (`Cosmos3VFMTransformer`+`Cosmos3LanguageModel`) built vs vllm-omni #3454; official checkpoint loads `Cosmos3OmniTransformer` (diffusers shim). Class/structure reconciliation required. | `model_index.json`; `diffusers_cosmos3/transformer.py`; branch commit `52bb65f49` | orchestrator | resolved | DiT rewritten to checkpoint layout (single `layers` dual-pathway, BaseDiT-conformant); bit-identical framework parity (3d_rope + unified_3d_mrope), commits 59a4a571c/7c4633295 |
 | I002 | all | tests | medium | Tier-A conftest+tests mirror vllm-omni line-by-line (stubs, `vllm_omni...guardrails`). Must be repointed to `diffusers_cosmos3` / official structures. | `tests/local_tests/cosmos3/conftest.py` | orchestrator | open | |
+| I003 | inference | scheduler | high | First real-weights T2V was all-black: checkpoint `scheduler_config.json` sets `use_karras_sigmas=true`; vendored UniPC checks karras before `use_flow_sigmas` -> diffusion (beta) sigmas -> `scheduler.step` -> NaN latents. DiT/CFG velocity was clean. The scheduler had never been parity-tested vs the framework (`test_cosmos3_denoise_cfg_parity` used diffusers UniPC on both sides). | `result_latent` NaN at denoise step 0 (v_pred clean); ffprobe 3 KB black mp4 | orchestrator | resolved | Coerce loaded config to flow setup in `initialize_pipeline`; switch pipeline+tests to native UniPC (no diffusers at runtime); add `test_cosmos3_scheduler_parity` vs framework `FlowUniPCMultistepScheduler`; repoint denoise_cfg oracle to the framework scheduler. Commit 255311cf2 |
 
 ## Escape Hatches
 
@@ -97,3 +98,10 @@
 - BaseDiT signature is `__init__(self, config: DiTConfig, hf_config: dict)`; the branch `Cosmos3VFMTransformer` uses `fastvideo_args`/SimpleNamespace and does NOT conform — rewrite to conform + match the checkpoint key surface (single `layers` dual-pathway, not split language_model/gen_layers).
 - Native layers (per cosmos2_5): `ReplicatedLinear`/`MLP`/`RMSNorm` (fastvideo.layers.*), `LocalAttention`/`DistributedAttention` (fastvideo.attention), `apply_rotary_emb` (use_real_unbind_dim=-2 for Cosmos). EntryClass at module bottom; class attrs bound from config; 3D-MRoPE has no reusable util — adapt Cosmos25RotaryPosEmbed.
 - NEXT: write native `fastvideo/models/dits/cosmos3.py` + fastvideo-vs-framework forward parity test (copy framework weights into the FastVideo DiT, compare outputs), then conversion script (diffusers checkpoint -> FastVideo) + strict-load, then video pipeline/packing.
+
+### PR1 (video core) acceptance — real-weights E2E — 2026-06-07
+- First real-weights T2V (`examples/inference/basic/basic_cosmos3_new_api.py`, `COSMOS3_MODEL_PATH=official_weights/cosmos3`) ran mechanically but produced an all-black 3 KB mp4. Instrumenting the denoise loop showed `v_pred` clean at step 0 but `scheduler.step` -> NaN. Root cause I003: checkpoint `scheduler_config.json` is diffusers-style (`use_karras_sigmas=true`), and the vendored UniPC checks karras before `use_flow_sigmas` -> diffusion (beta) sigmas instead of flow sigmas -> NaN. The framework actually samples with `FlowUniPCMultistepScheduler` (pure flow: `shift` + `num_train_timesteps`).
+- Fix (commit `255311cf2`): coerce the loaded scheduler to the flow setup in `Cosmos3OmniDiffusersPipeline.initialize_pipeline`; use FastVideo's native UniPC (not diffusers) in pipeline + tests. Added `test_cosmos3_scheduler_parity.py` (native UniPC flow-config vs framework `FlowUniPCMultistepScheduler`: timesteps bit-exact, sigmas ~1e-8, full trajectory <~1e-6 over shift in {10,3}, steps in {4,10,35}). Repointed `test_cosmos3_denoise_cfg_parity` oracle to the framework scheduler (it previously compared diffusers-vs-diffusers, so the scheduler was never checked against the framework).
+- Also wired the remaining integration glue (registry alias `Cosmos3OmniTransformer`->`Cosmos3VFMTransformer`; `text_tokenizer`->TokenizerLoader; scheduler config param-filtering; DiT `materialize_non_persistent_buffers` + compute-dtype casts; packing device-move in `to_dit_kwargs`; empty text-preprocess).
+- Verified: 1280x704, 29 frames, 35 steps on a single B200 -> coherent golden-retriever-in-meadow video matching the prompt (no NaNs; per-frame pixel std ~58; visible temporal motion). Full cosmos3 suite: 95 passed, 0 skipped.
+- NEXT: PR2 audio (`sound_tokenizer` AVAE) / PR3 action / PR4 reasoning. Optional: I2V/T2I real-weights spot-checks; force-push branch (needs explicit OK).