wlsaidhi commited on
Commit
b790390
·
verified ·
1 Parent(s): 8b09666

[model]: document v10.5 checkpoint provenance

Browse files
Files changed (2) hide show
  1. README.md +4 -0
  2. provenance.json +19 -0
README.md CHANGED
@@ -30,6 +30,7 @@ evaluation and UI integration, not a public release recommendation.
30
  ## Checkpoint identity
31
 
32
  - Training run: `20260823T132231Z-v10p5-datafree-sp4`
 
33
  - Step: 1300
34
  - Transformer content SHA-256: `b36987515e4c75fa4c7aaa632a7842c829ea141b235358a54d782b51230497b3`
35
  - Checkpoint metadata SHA-256: `dcad0fbee2a7c7e75e53435f4fd98fccf3138844883874edf057962ab48fa428`
@@ -86,6 +87,9 @@ runs the sparse DiT eagerly; regional compile is disabled for this snapshot.
86
  ## Limitations
87
 
88
  - Intermediate checkpoint at step 1300 of a planned 4000-step run.
 
 
 
89
  - T2AV only; the reference-conditioned transformer was not distilled here.
90
  - Quality remains below base MiniMax-H3 on difficult motion, fine structure,
91
  and some audio details.
 
30
  ## Checkpoint identity
31
 
32
  - Training run: `20260823T132231Z-v10p5-datafree-sp4`
33
+ - W&B: [a1etree2](https://wandb.ai/wlsaidhi/h3-dmd2-vsa/runs/a1etree2)
34
  - Step: 1300
35
  - Transformer content SHA-256: `b36987515e4c75fa4c7aaa632a7842c829ea141b235358a54d782b51230497b3`
36
  - Checkpoint metadata SHA-256: `dcad0fbee2a7c7e75e53435f4fd98fccf3138844883874edf057962ab48fa428`
 
87
  ## Limitations
88
 
89
  - Intermediate checkpoint at step 1300 of a planned 4000-step run.
90
+ - This historical checkpoint predates the later strict continuous score-clock
91
+ and FastGen-alignment corrections. Use it to reproduce v10.5 results, not as
92
+ the reference implementation of the corrected training recipe.
93
  - T2AV only; the reference-conditioned transformer was not distilled here.
94
  - Quality remains below base MiniMax-H3 on difficult motion, fine structure,
95
  and some audio details.
provenance.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": "fasth3-checkpoint-provenance-v1",
3
+ "model_id": "FastVideo/FastVideo-FastH3-4-step-v1",
4
+ "base_model": "hf://MiniMaxAI/MiniMax-H3@9bfb6693f2cf6de171db46d1aa586f67d773a1da",
5
+ "training_run_id": "20260823T132231Z-v10p5-datafree-sp4",
6
+ "wandb_url": "https://wandb.ai/wlsaidhi/h3-dmd2-vsa/runs/a1etree2",
7
+ "checkpoint_step": 1300,
8
+ "checkpoint_content_sha256": "b36987515e4c75fa4c7aaa632a7842c829ea141b235358a54d782b51230497b3",
9
+ "checkpoint_metadata_sha256": "dcad0fbee2a7c7e75e53435f4fd98fccf3138844883874edf057962ab48fa428",
10
+ "fastvideo_commit": "48a047c05ff4138f20cfa33351499c6ec5945f5d",
11
+ "student_tensor_count": 688,
12
+ "student_tensor_bytes": 70099502592,
13
+ "student_dtype": "bfloat16",
14
+ "dmd_denoising_steps": [999, 749, 500, 250],
15
+ "attention_backend": "VIDEO_SPARSE_ATTN_H3",
16
+ "vsa_tile_size": 64,
17
+ "vsa_sparsity": 0.9,
18
+ "known_recipe_deviation": "Predates strict continuous score-clock and later FastGen-alignment corrections."
19
+ }