Feature Extraction
Transformers
Safetensors
qwen3_5
matilda
jev
fp4
quantized
maincode
8-bit precision
Instructions to use Maincode/matilda-jev-fp4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Maincode/matilda-jev-fp4 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("feature-extraction", model="Maincode/matilda-jev-fp4")# pip install -U transformers accelerate # Load model directly from transformers import AutoProcessor, AutoModel processor = AutoProcessor.from_pretrained("Maincode/matilda-jev-fp4") model = AutoModel.from_pretrained("Maincode/matilda-jev-fp4", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Download MODEL_INFO.json from Maincode/matilda-jev-fp4: direct link, hf CLI and curl.
- Browser
- Download file 1.5 kB
-
https://huggingface.co/Maincode/matilda-jev-fp4/resolve/main/MODEL_INFO.json
- Command line
-
hf download hf://Maincode/matilda-jev-fp4/MODEL_INFO.json
-
curl -L -o MODEL_INFO.json https://huggingface.co/Maincode/matilda-jev-fp4/resolve/main/MODEL_INFO.json
1.5 kB
| { | |
| "created_utc": "2026-10-06T16:01:38.503748+00:00", | |
| "status": "FP4 W4A16; full Decision Index and native JEV inference validated", | |
| "model_path": "Maincode/matilda-jev-fp4", | |
| "source_checkpoint": "/shared/yue/jev-best-tritask-20261006/runs/balanced/checkpoints/step-00800", | |
| "source_decision_index": 62.43, | |
| "quantized_decision_index": 61.77, | |
| "format": "FP4 E2M1 packed weights; FP8 E4M3 scale per16weights; FP32global scale; BF16activations and GEMM", | |
| "weight_tensor_GB": 17.20065232, | |
| "original_weight_file_GB": 52.170802736, | |
| "native_fp4_tensor_core_GEMM": false, | |
| "readout_and_tokenizer_identical": true, | |
| "quantized_modules": 400, | |
| "paired_checks": { | |
| "development": { | |
| "n": 608, | |
| "agreement_percent": 95.88815789473684, | |
| "bf16_accuracy_percent": 80.92105263157895, | |
| "fp4_accuracy_percent": 81.57894736842105, | |
| "mean_KL_bf16_to_fp4": 0.015593464629358985 | |
| }, | |
| "confirmation": { | |
| "n": 544, | |
| "agreement_percent": 97.24264705882354, | |
| "bf16_accuracy_percent": 77.57352941176471, | |
| "fp4_accuracy_percent": 76.83823529411765, | |
| "mean_KL_bf16_to_fp4": 0.011581205562651279 | |
| } | |
| }, | |
| "hardware": "AMD Instinct MI355X", | |
| "serving_endpoints_changed": false, | |
| "notes": [ | |
| "Full Decision Index edition0.2.1:150317successful requests across44benchmarks; all native scores independently recomputed.", | |
| "Requires included FP4DecisionModel adapter. BF16 activations and GEMM; no native FP4 Tensor Core acceleration." | |
| ] | |
| } | |