---
license: mit
license_link: https://opensource.org/license/mit
base_model: microsoft/Phi-4-reasoning
tags:
  - quantized
  - 4-bit
  - int4
  - awq
language:
  - en
library_name: transformers
pipeline_tag: text-generation
---

# Phi-4-reasoning-NVFP4

INT4 weight-only quantization of [`microsoft/Phi-4-reasoning`](https://huggingface.co/microsoft/Phi-4-reasoning).

Microsoft Phi-4-reasoning in NVFP4 W4A4. Native vLLM compressed-tensors. About 8 GB on disk.

| Property | Value |
|---|---|
| Base model | [microsoft/Phi-4-reasoning](https://huggingface.co/microsoft/Phi-4-reasoning) |
| Quantization | INT4 weight-only |
| Approx. on-disk size | ~9.7 GB |
| License | MIT License |
| Languages | English |

## Load (vLLM)

```bash
vllm serve drawais/Phi-4-reasoning-NVFP4 \
  --max-model-len 32768 \
  --gpu-memory-utilization 0.94
```

```python
from vllm import LLM, SamplingParams
llm = LLM(model="drawais/Phi-4-reasoning-NVFP4", max_model_len=32768)
print(llm.generate(["Hello!"], SamplingParams(max_tokens=128))[0].outputs[0].text)
```

## Footprint

~9.7 GB on disk. Recommended VRAM: enough headroom for KV cache.

## License & attribution

This artifact is a derivative work of [`microsoft/Phi-4-reasoning`](https://huggingface.co/microsoft/Phi-4-reasoning),
released by its original authors under the **MIT License**.

This artifact is distributed under the same license. The full license text is
included in [`LICENSE`](LICENSE), and required attribution is in [`NOTICE`](NOTICE).

License text: https://opensource.org/license/mit
Source model: https://huggingface.co/microsoft/Phi-4-reasoning