Download vllm_check_nvfp4.json from Mapika/decider-35b-a3b-nvfp4: direct link, hf CLI and curl.
- Browser
- Download file 873 Bytes
-
https://huggingface.co/Mapika/decider-35b-a3b-nvfp4/resolve/main/vllm_check_nvfp4.json
- Command line
-
hf download hf://Mapika/decider-35b-a3b-nvfp4/vllm_check_nvfp4.json
-
curl -L -o vllm_check_nvfp4.json https://huggingface.co/Mapika/decider-35b-a3b-nvfp4/resolve/main/vllm_check_nvfp4.json
873 Bytes
| { | |
| "model": "/mnt/data/mark/projects/decider/runs/moe_optimizer_ab_v1/muon/model_nvfp4", | |
| "temperature": 1.08, | |
| "quantization": "NVFP4 (ModelOpt 0.46.1, NVFP4_DEFAULT_CFG, 512 calibration prompts); for vLLM / TensorRT-LLM on Blackwell; the bf16 checkpoint is the reference", | |
| "fixtures": { | |
| "typesafe": { | |
| "rows": 102, | |
| "accuracy": 0.8431372549019608, | |
| "nll": 0.3648713501022975, | |
| "argmax_agreement_with_bf16": 0.9607843137254902, | |
| "mean_total_variation_to_bf16": 0.050491656887313836, | |
| "seconds": 1.993523359298706, | |
| "rows_per_second": 51.1656908980902 | |
| }, | |
| "general_validation": { | |
| "rows": 847, | |
| "accuracy": 0.8819362455726092, | |
| "nll": 0.33909610998107875, | |
| "argmax_agreement_with_bf16": 0.9669421487603306, | |
| "mean_total_variation_to_bf16": 0.02974127205877468, | |
| "seconds": 3.0871644020080566, | |
| "rows_per_second": 274.3618057558146 | |
| } | |
| } | |
| } |