tarekziade HF Staff
Publish MIC usage, slim fixture and reviewed Transformers expectations
cb65d7e verified | { | |
| "mic": "0.3", | |
| "name": "MiniMax-M3-MXFP8 VLM", | |
| "usage": "AutoProcessor + MiniMaxM3SparseForConditionalGeneration, text inference and projector training step", | |
| "architecture": "MiniMaxM3SparseForConditionalGeneration", | |
| "checkpoints": { | |
| "MiniMaxAI/MiniMax-M3-MXFP8": { | |
| "fixture": "huggingface-mic/MiniMax-M3-MXFP8-mic" | |
| } | |
| }, | |
| "fixtures": { | |
| "huggingface-mic/MiniMax-M3-MXFP8-mic": { | |
| "built_from": "MiniMaxAI/MiniMax-M3-MXFP8", | |
| "covers": [ | |
| "MiniMaxAI/MiniMax-M3-MXFP8" | |
| ], | |
| "build": { | |
| "seed": 17, | |
| "config_overrides": { | |
| "fields": { | |
| "text_config.index_block_size": 8, | |
| "text_config.index_topk_blocks": 4, | |
| "text_config.index_local_blocks": 1, | |
| "text_config.index_n_heads": 2 | |
| }, | |
| "reason": "Shrink the Lightning indexer like the Transformers unit-test tester (block 8, top-k 4) so a short image prompt is far longer than top-k x block and selection is genuinely sparse; the real checkpoint only goes sparse above 2,048 tokens." | |
| } | |
| } | |
| } | |
| }, | |
| "integrations": { | |
| "transformers": { | |
| "entry_point": "integrations/transformers/usage.md", | |
| "format": "runnable-markdown", | |
| "requires": [ | |
| "transformers>=5", | |
| "torch", | |
| "Pillow", | |
| "torchvision", | |
| "accelerate" | |
| ], | |
| "targets": [ | |
| "fixtures" | |
| ], | |
| "variants": { | |
| "default": {}, | |
| "eager": { | |
| "attn_implementation": "eager" | |
| }, | |
| "bfloat16": { | |
| "dtype": "bfloat16" | |
| } | |
| }, | |
| "training": true, | |
| "resources": { | |
| "fixture": { | |
| "memory_gb": 4, | |
| "download_mb": 50 | |
| }, | |
| "MiniMaxAI/MiniMax-M3-MXFP8": { | |
| "devices": [ | |
| "cuda" | |
| ], | |
| "gpus": 4, | |
| "gpu_memory_gb": 120, | |
| "memory_gb": 64, | |
| "download_mb": 450000 | |
| } | |
| } | |
| } | |
| }, | |
| "outputs": { | |
| "runtime": { | |
| "compare": "exact" | |
| }, | |
| "logits_probe": { | |
| "compare": "allclose", | |
| "atol": 0.0001, | |
| "rtol": 0.0001 | |
| }, | |
| "training.loss": { | |
| "compare": "allclose", | |
| "atol": 0.0001, | |
| "rtol": 0.0001 | |
| }, | |
| "training.gradients": { | |
| "compare": "all_true" | |
| }, | |
| "training.updated": { | |
| "compare": "true" | |
| }, | |
| "training.frozen": { | |
| "compare": "true" | |
| } | |
| } | |
| } | |