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

[test]: seed Cosmos3 I2V fingerprint

Browse Source
This commit is contained in:
SolitaryThinker
2026-07-21 16:50:23 -07:00
parent 2e64d70389
commit 5129a450b0
3 changed files with 14 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastvideo/tests/transformers/test_cosmos3_block_fingerprint.py
+2 -2
View File
@@ -62,11 +62,11 @@ INPUT_SEED = 20260721
# Golden values seeded on the exact Modal L40S image asserted below.
EXPECTED_ENVIRONMENT_SHA256 = "ad3d22373224999f0a0520c96eb406335ad5b58bdf4fd2f3e4ec29de2f6e1dda"
EXPECTED_CONTRACT_SHA256 = "5043ad8a281ddfd2595317c45943c2c06d71cf8012525db7efed1db7af7e31e3"
EXPECTED_INPUT_SHA256 = "TO_BE_SEEDED"
EXPECTED_INPUT_SHA256 = "60ed6cb71913c98bcdfe4b7e921097592d1b3715f411adce57f93c75ab913aeb"
EXPECTED_WEIGHTS_SHA256 = "480b5a497bb0c9ef71edabeee62c057b2c54fce753362e8938d073d6bf7cfa62"
EXPECTED_OUTPUT_SHA256 = {
"action2world": "4cf042b91e1a0c4131efbb454e23d679b73542abda9b36af18cd81bb694b1b34",
"i2v": "TO_BE_SEEDED",
"i2v": "4b7c8bec2e3a3a710b7c022ff74ba83579c02881e29d5db3bad5fb78eff2ba40",
"reason": "112b9ecd765b0e1afc7d21d169b4f3e941901d975c27f59df6062dedef60bc35",
"t2v": "2b15c03d936bace07acea7aba55fe850dd3cadc344b361dce8da68acd94f9cb7",
"t2vs": "1bffb714b7f316665e0176ddfe2b930df70d6dae1f75b3f761e92b5b9d8c26dd",
tests/local_tests/cosmos3/PORT_STATUS.md
+8 -5
View File
@@ -40,8 +40,8 @@ generated-media/audio regression remain follow-ups.
|---|---|---|
| Local framework component suite | `150 passed, 0 skipped` | Recorded on 2026-06-07 before the final aggregate rebase; this is not a fresh run against the fingerprint's newer official pin |
| Real-weight B200 inference | T2V, I2V, T2I, T2VS, text reasoning | Recorded on 2026-06-07; outputs were inspected and described as coherent/prompt-matching |
| Transformer fingerprint seed | `2 passed` | Modal L40S, 2026-07-21; T2V/T2VS/action decoder-layer outputs matched the pinned official framework layer bit-exact; config contract passed |
| Transformer fingerprint normal mode | `2 passed` | Modal L40S, 2026-07-21; T2V/T2VS/action/reasoning committed hashes passed without loading the official framework |
| Transformer fingerprint seed | `2 passed` | Modal L40S, 2026-07-21; T2V/I2V/T2VS/action decoder-layer outputs matched the pinned official framework layer bit-exact; config contract passed |
| Transformer fingerprint normal mode | `2 passed` | Modal L40S, 2026-07-21; T2V/I2V/T2VS/action/reasoning committed hashes passed without loading the official framework |
| Modal checkout unit tests | `12 passed` | Exact immutable Buildkite commit checkout and bounded retries |
Run the normal transformer fingerprint with:
@@ -69,11 +69,13 @@ for `nvidia/Cosmos3-Nano`.
- Config/structural contract SHA-256:
`5043ad8a281ddfd2595317c45943c2c06d71cf8012525db7efed1db7af7e31e3`
- Input SHA-256:
`4c2f1879c1847f65372fbb08df981bbb00e2e4b7ad6176c22db5515e69b2846f`
`60ed6cb71913c98bcdfe4b7e921097592d1b3715f411adce57f93c75ab913aeb`
- Selected weights SHA-256:
`480b5a497bb0c9ef71edabeee62c057b2c54fce753362e8938d073d6bf7cfa62`
- T2V outputs SHA-256:
`2b15c03d936bace07acea7aba55fe850dd3cadc344b361dce8da68acd94f9cb7`
- I2V outputs SHA-256:
`4b7c8bec2e3a3a710b7c022ff74ba83579c02881e29d5db3bad5fb78eff2ba40`
- T2VS outputs SHA-256:
`1bffb714b7f316665e0176ddfe2b930df70d6dae1f75b3f761e92b5b9d8c26dd`
- Action2world outputs SHA-256:
@@ -81,8 +83,9 @@ for `nvidia/Cosmos3-Nano`.
- Reasoning outputs SHA-256:
`112b9ecd765b0e1afc7d21d169b4f3e941901d975c27f59df6062dedef60bc35`
Normal CI runs separate depth-one T2V, T2VS, action2world, and deepstack-reason
forwards twice and checks exact output identity plus the committed hashes.
Normal CI runs separate depth-one T2V, I2V, T2VS, action2world, and
deepstack-reason forwards twice and checks exact output identity plus the
committed hashes. The I2V case keeps frame 0 clean and predicts only frame 1.
Seed/reseed mode additionally imports the pinned NVIDIA framework and compares
the captured generation decoder layer bit-for-bit. No extracted fixture or
separate HF upload is needed.
tests/local_tests/cosmos3/README.md
+4 -4
View File
@@ -65,15 +65,15 @@ a separate metric because SSIM cannot cover it.
## Transformer CI fingerprint
The routine transformer gate runs separate depth-one FastVideo T2V, T2VS,
action2world, and deepstack-reasoning forwards with production hidden, head,
MLP, latent, sound, and action dimensions. It reads 44 tensors or slices
The routine transformer gate runs separate depth-one FastVideo T2V, I2V,
T2VS, action2world, and deepstack-reasoning forwards with production hidden,
head, MLP, latent, sound, and action dimensions. It reads 44 tensors or slices
(408,486,912 parameters, about 779.13 MiB in BF16) from the pinned
`nvidia/Cosmos3-Nano` checkpoint cache. It hashes the production/fingerprint
config contract, fixed inputs, selected weights, and every named output.
The golden was seeded and verified on Modal L40S using the pinned CI image. Seed
mode also proved the captured FastVideo decoder-layer outputs for all three
mode also proved the captured FastVideo decoder-layer outputs for all four
generation cases bit-exact against the pinned NVIDIA framework layer. Normal
mode reran every case twice from the committed hashes and passed. No extracted
fixture or separate weights repository is required; SSIM remains the