---
license: apache-2.0
base_model: orcarouter/Qwen3.8-27B-Uncensored-FP8
pipeline_tag: image-text-to-text
tags:
- transformers
- safetensors
- qwen3_5
- fp8
- block-fp8
- vllm
- nvidia
---

# Qwen3.8-27B-Uncensored-FP8 (NVIDIA / FP8)

Unofficial mirror of [orcarouter/Qwen3.8-27B-Uncensored-FP8](https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-FP8) in the original **FP8 safetensors** format, intended for **NVIDIA GPUs** (vLLM / transformers with FP8 support).

## Format
- `transformers` checkpoint, `qwen3_5` architecture (`Qwen3_5ForConditionalGeneration`)
- 7 safetensors shards, FP8 (block-FP8) quantized weights
- Multimodal (image-text-to-text)

## Usage (NVIDIA / vLLM)
```bash
# vLLM (FP8 native)
vllm serve id-2/Qwen3.8-27B-Uncensored-FP8 --quantization fp8

# transformers
from transformers import AutoModelForCausalLM, AutoProcessor
model = AutoModelForCausalLM.from_pretrained("id-2/Qwen3.8-27B-Uncensored-FP8")
```

> Note: MLX format (Apple Silicon) is NOT provided here — this is the NVIDIA/FP8 build.