Neooooo/qf-integration-test

1---2base_model: Qwen/Qwen3-30B-A3B3datasets:4- HuggingFaceH4/ultrachat_200k5library_name: transformers6pipeline_tag: text-generation7quantized_by: QuantForge8tags:9- quantforge10- quantized11- nvfp412---13 14# Neooooo/qf-integration-test15 16## QuantForge Metadata17 18- Base model: `Qwen/Qwen3-30B-A3B`19- Quantization scheme: `nvfp4`20- Calibration dataset: `HuggingFaceH4/ultrachat_200k`21- Calibration samples: `32`22- Max sequence length: `512`23- Ignored layers: `lm_head, re:.*\.mlp\.gate$, re:.*\.mlp\.router$`24 25## Accuracy (BF16 vs NVFP4)26 27| Task | Metric | BF16 | NVFP4 | Recovery |28|---|---:|---:|---:|---:|29| arc_challenge | acc,none | 0.4000 | 0.3000 | 0.750 |30| hellaswag | acc,none | 0.4000 | 0.4000 | 1.000 |31 32Aggregate macro recovery: **0.875**33 34> **Note:** Scores estimated from subset.35 36## Performance37 38_Performance benchmark unavailable: evaluate.skip_perf=true_39 40## Usage (vLLM)41 42```bash43vllm serve Neooooo/qf-integration-test44```45