wlsaidhi commited on
Commit
cf2917a
·
verified ·
1 Parent(s): b790390

[docs]: align FastH3 Preview v1 naming

Browse files

Update the model card and package self-references after the server-side repository rename.

README.md CHANGED
@@ -16,16 +16,18 @@ tags:
16
  - minimax-h3
17
  - fastvideo
18
  - fasth3
 
19
  ---
20
 
21
- # FastVideo-FastH3-4-step-v1
22
 
23
- Private FastVideo evaluation snapshot of the v10.5 data-free DMD2 MiniMax-H3
24
- student at training step 1300. It generates synchronized video and audio with
25
- four transformer forwards.
 
26
 
27
- This repository is intentionally private. It is an internal checkpoint for
28
- evaluation and UI integration, not a public release recommendation.
29
 
30
  ## Checkpoint identity
31
 
@@ -58,7 +60,7 @@ four-point scheduler grid.
58
  export FASTVIDEO_DMD_DENOISING_STEPS=999,749,500,250
59
 
60
  python examples/inference/basic/basic_fasth3.py \
61
- --model-path FastVideo/FastVideo-FastH3-4-step-v1 \
62
  --prompt '<MiniMax-H3 T2AV prompt>' \
63
  --steps 5 \
64
  --num-gpus 4 \
@@ -69,7 +71,7 @@ python examples/inference/basic/basic_fasth3.py \
69
 
70
  The five scheduler grid points execute the four trained jumps. Use guidance
71
  scale 1.0, VSA-H3 at 90% sparsity with 64-token tiles, and SP=4/FSDP across one
72
- four-GPU GB200 tray for the validated route. The exact v10.5 execution commit
73
  runs the sparse DiT eagerly; regional compile is disabled for this snapshot.
74
 
75
  ## Training summary
@@ -82,12 +84,12 @@ runs the sparse DiT eagerly; regional compile is disabled for this snapshot.
82
  - Global batch size: 64.
83
  - Student and critic learning rate: `2e-6`.
84
  - Precision: fp32 training state and bf16 inference export.
85
- - Training topology: 32 GB200 GPUs, SP=4, full-world FSDP sharding.
86
 
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.
 
16
  - minimax-h3
17
  - fastvideo
18
  - fasth3
19
+ - preview
20
  ---
21
 
22
+ # FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree
23
 
24
+ FastH3 4-Step Preview v1 checkpoint from FastVideo. This variant is the v10.5
25
+ data-free DMD2 MiniMax-H3 student at training step 1300. It generates
26
+ synchronized video and audio with four transformer forwards and uses VSA-H3 at
27
+ 90% sparsity.
28
 
29
+ See the complete [FastH3 4-Step Preview v1
30
+ collection](https://huggingface.co/collections/FastVideo/fastvideo-fasth3).
31
 
32
  ## Checkpoint identity
33
 
 
60
  export FASTVIDEO_DMD_DENOISING_STEPS=999,749,500,250
61
 
62
  python examples/inference/basic/basic_fasth3.py \
63
+ --model-path FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree \
64
  --prompt '<MiniMax-H3 T2AV prompt>' \
65
  --steps 5 \
66
  --num-gpus 4 \
 
71
 
72
  The five scheduler grid points execute the four trained jumps. Use guidance
73
  scale 1.0, VSA-H3 at 90% sparsity with 64-token tiles, and SP=4/FSDP across one
74
+ four-GPU B200 tray for the validated route. The exact v10.5 execution commit
75
  runs the sparse DiT eagerly; regional compile is disabled for this snapshot.
76
 
77
  ## Training summary
 
84
  - Global batch size: 64.
85
  - Student and critic learning rate: `2e-6`.
86
  - Precision: fp32 training state and bf16 inference export.
87
+ - Training topology: 32 B200 GPUs, SP=4, full-world FSDP sharding.
88
 
89
  ## Limitations
90
 
91
  - Intermediate checkpoint at step 1300 of a planned 4000-step run.
92
+ - This historical checkpoint predates the later continuous score-clock
93
  and FastGen-alignment corrections. Use it to reproduce v10.5 results, not as
94
  the reference implementation of the corrected training recipe.
95
  - T2AV only; the reference-conditioned transformer was not distilled here.
fastvideo_inference.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "schema_version": "fasth3-inference-contract-v1",
3
- "model_id": "FastVideo/FastVideo-FastH3-4-step-v1",
4
  "training_parent_run_id": "20260823T132231Z-v10p5-datafree-sp4",
5
  "checkpoint_step": 1300,
6
  "checkpoint_content_sha256": "b36987515e4c75fa4c7aaa632a7842c829ea141b235358a54d782b51230497b3",
 
1
  {
2
  "schema_version": "fasth3-inference-contract-v1",
3
+ "model_id": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree",
4
  "training_parent_run_id": "20260823T132231Z-v10p5-datafree-sp4",
5
  "checkpoint_step": 1300,
6
  "checkpoint_content_sha256": "b36987515e4c75fa4c7aaa632a7842c829ea141b235358a54d782b51230497b3",
modular_model_index.json CHANGED
@@ -10,7 +10,7 @@
10
  "transformers",
11
  "Qwen3VLForConditionalGeneration"
12
  ],
13
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-v1",
14
  "subfolder": "text_encoder",
15
  "variant": null,
16
  "revision": null
@@ -24,7 +24,7 @@
24
  "transformers",
25
  "Qwen2TokenizerFast"
26
  ],
27
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-v1",
28
  "subfolder": "tokenizer",
29
  "variant": null,
30
  "revision": null
@@ -38,7 +38,7 @@
38
  "transformers",
39
  "Qwen3VLProcessor"
40
  ],
41
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-v1",
42
  "subfolder": "processor",
43
  "variant": null,
44
  "revision": null
@@ -52,7 +52,7 @@
52
  "diffusers",
53
  "AutoencoderKLMiniMaxH3"
54
  ],
55
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-v1",
56
  "subfolder": "vae",
57
  "variant": null,
58
  "revision": null
@@ -66,7 +66,7 @@
66
  "diffusers",
67
  "AutoencoderKLMiniMaxH3Audio"
68
  ],
69
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-v1",
70
  "subfolder": "audio_vae",
71
  "variant": null,
72
  "revision": null
@@ -80,7 +80,7 @@
80
  "diffusers",
81
  "MiniMaxH3Transformer3DModel"
82
  ],
83
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-v1",
84
  "subfolder": "transformer",
85
  "variant": null,
86
  "revision": null
@@ -108,7 +108,7 @@
108
  "diffusers",
109
  "MiniMaxH3Scheduler"
110
  ],
111
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-v1",
112
  "subfolder": "scheduler",
113
  "variant": null,
114
  "revision": null
@@ -122,7 +122,7 @@
122
  "diffusers",
123
  "MiniMaxH3Scheduler"
124
  ],
125
- "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-v1",
126
  "subfolder": "audio_scheduler",
127
  "variant": null,
128
  "revision": null
 
10
  "transformers",
11
  "Qwen3VLForConditionalGeneration"
12
  ],
13
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree",
14
  "subfolder": "text_encoder",
15
  "variant": null,
16
  "revision": null
 
24
  "transformers",
25
  "Qwen2TokenizerFast"
26
  ],
27
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree",
28
  "subfolder": "tokenizer",
29
  "variant": null,
30
  "revision": null
 
38
  "transformers",
39
  "Qwen3VLProcessor"
40
  ],
41
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree",
42
  "subfolder": "processor",
43
  "variant": null,
44
  "revision": null
 
52
  "diffusers",
53
  "AutoencoderKLMiniMaxH3"
54
  ],
55
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree",
56
  "subfolder": "vae",
57
  "variant": null,
58
  "revision": null
 
66
  "diffusers",
67
  "AutoencoderKLMiniMaxH3Audio"
68
  ],
69
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree",
70
  "subfolder": "audio_vae",
71
  "variant": null,
72
  "revision": null
 
80
  "diffusers",
81
  "MiniMaxH3Transformer3DModel"
82
  ],
83
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree",
84
  "subfolder": "transformer",
85
  "variant": null,
86
  "revision": null
 
108
  "diffusers",
109
  "MiniMaxH3Scheduler"
110
  ],
111
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree",
112
  "subfolder": "scheduler",
113
  "variant": null,
114
  "revision": null
 
122
  "diffusers",
123
  "MiniMaxH3Scheduler"
124
  ],
125
+ "pretrained_model_name_or_path": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree",
126
  "subfolder": "audio_scheduler",
127
  "variant": null,
128
  "revision": null
provenance.json CHANGED
@@ -1,6 +1,6 @@
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",
@@ -15,5 +15,5 @@
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
  }
 
1
  {
2
  "schema_version": "fasth3-checkpoint-provenance-v1",
3
+ "model_id": "FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree",
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",
 
15
  "attention_backend": "VIDEO_SPARSE_ATTN_H3",
16
  "vsa_tile_size": 64,
17
  "vsa_sparsity": 0.9,
18
+ "known_recipe_deviation": "Predates later continuous score-clock and FastGen-alignment corrections."
19
  }