nodes/hao-ai-lab-FastVideo
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[evidence]: serving identity gate pass — served latents bitwise vs offline, live per-step WS

Browse Source
This commit is contained in:
SolitaryThinker
2026-07-23 11:23:59 -07:00
parent b376207026
commit 8cc6abc107
2 changed files with 14 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastvideo2/evidence/ledger.jsonl
+1
View File
@@ -164,3 +164,4 @@
{"ts": "2026-07-23T13:02:54Z", "gate": "anchor.train-qat-main", "status": "pass", "model_id": "wan2.1-t2v-1.3b", "card_digest": "7171d8729ae72ae4", "metrics": {"noisy.step0": 0.0, "pred.step0": 0.0, "loss.step0": 0.0, "gnorm.step0": 0.0003916621208190918, "loss.step1": 2.3126602172851562e-05, "gnorm.step1": 2.3497252464294434, "loss.step2": 0.00043427199125289917, "gnorm.step2": 0.09579342603683472, "loss.step3": 0.012151837348937988, "gnorm.step3": 2.280292510986328, "loss.step4": 0.05067098140716553, "gnorm.step4": 1.1354649066925049, "params.w0": 0.0, "params.w5": 0.004772356349228613}, "tolerances": {"step0": 0.0, "later": "main self-noise band (see manifest)"}, "env": {"python": "3.12.13", "machine": "aarch64", "commit": "088b9b4", "torch": "2.12.0+cu130", "cuda": "13.0", "gpu": "NVIDIA GB200", "diffusers": "0.39.0", "transformers": "5.13.0", "attention": {"policy": "AUTO", "available": ["FLASH_ATTN", "SDPA"]}}, "detail": "goldens c459a1897"}
{"ts": "2026-07-23T13:08:02Z", "gate": "anchor.train-qad-main", "status": "pass", "model_id": "wan2.1-t2v-1.3b", "card_digest": "7171d8729ae72ae4", "metrics": {"x0.step0": 0.0, "fake.step0": 7.806811481714249e-07, "x0.step1": 0.0, "gen.step1": 0.00018775463104248047, "fake.step1": 1.8295831978321075e-05, "x0.step2": 1.0, "fake.step2": 0.0001232912763953209, "x0.step3": 1.0, "gen.step3": 0.0005949586629867554, "fake.step3": 0.00010955892503261566, "x0.step4": 1.0, "fake.step4": 1.1123716831207275e-05, "params.w5_student": 4.994589533982423e-06, "params.w5_critic": 7.754974838099682e-06}, "tolerances": {"x0.step0": 0.0, "later": "band"}, "env": {"python": "3.12.13", "machine": "aarch64", "commit": "a8205ea", "torch": "2.12.0+cu130", "cuda": "13.0", "gpu": "NVIDIA GB200", "diffusers": "0.39.0", "transformers": "5.13.0", "attention": {"policy": "AUTO", "available": ["FLASH_ATTN", "SDPA"]}}, "detail": "goldens c459a1897"}
{"ts": "2026-07-23T17:51:55Z", "gate": "anchor.train-self_forcing-main", "status": "pass", "model_id": "sfwan-t2v-1.3b", "card_digest": "9c9b4cc573606028", "metrics": {"x0.step0": 0.0, "fake.step0": 0.00029414892196655273, "x0.step1": 0.0, "fake.step1": 0.0005563050508499146, "x0.step2": 0.0, "fake.step2": 0.00035434961318969727, "x0.step3": 0.0, "fake.step3": 0.0002406388521194458, "x0.step4": 0.0, "fake.step4": 0.00022238492965698242, "fwd.matches": 0.0}, "tolerances": {"x0-pre-update": 0.0, "later": "band"}, "env": {"python": "3.12.13", "machine": "aarch64", "commit": "93153f8", "torch": "2.12.0+cu130", "cuda": "13.0", "gpu": "NVIDIA GB200", "diffusers": "0.39.0", "transformers": "5.13.0", "attention": {"policy": "AUTO", "available": ["FLASH_ATTN", "SDPA"]}}, "detail": "goldens c459a1897"}
{"ts": "2026-07-23T18:23:19Z", "gate": "anchor.serve-identity", "status": "pass", "model_id": "fastwan-qad-fp8-1.3b", "card_digest": "1c8e6f7d1380552d", "metrics": {"identity": 0.0, "ws_steps": 3.0}, "tolerances": {"identity": 0.0}, "env": {"python": "3.12.13", "machine": "aarch64", "commit": "b376207", "torch": "2.12.0+cu130", "cuda": "13.0", "gpu": "NVIDIA GB200", "diffusers": "0.39.0", "transformers": "5.13.0", "attention": {"policy": "AUTO", "available": ["FLASH_ATTN", "SDPA"]}}, "detail": "served 41c4817d3d0f666b offline 41c4817d3d0f666b"}
fastvideo2/evidence/train_numerics_report.md
+13
View File
@@ -157,3 +157,16 @@ forward pass. A checkpointing-aware generator-step gate is deferred; the DMD
loss math itself is already parity-proven by the dmd2/qad gates. Config:
240x416 (resolution-agnostic math), 21 latent frames (pins the block-count
draw and avoids the >21-frame VAE re-encode branch).
## Serving identity — added 2026-07-23
Gate `anchor.serve-identity`: PASS. `fastvideo2.serve` (async-job REST +
WebSocket) over the engine: a served fastwan-qad request produced latents
BITWISE identical to the offline SDK for the same request/seed
(sha 41c4817d3d0f666b both sides), and the WebSocket delivered one LIVE
progress event per denoise step (3/3) via the engine's new `on_step` hook —
true per-step streaming, which main's servers list as a TODO.
Environment note (in code): this venv's FastAPI websocket dependency
injection closes connections without invoking endpoints; the stream route
is registered at the starlette level.