Qwen3.8-27B-TWIN-TURBO-Fable-Cold-Fusion-709-L-Uncensored (MXFP4)

Official Solstice-AI OCP Microscaling FP4 Suite

Original Model & GAIN Merge by DavidAU • MXFP4 Quantization & Packaging by Solstice-AI

Overview

OCP Microscaling FP4 (MXFP4) format via compressed-tensors with dynamic group scaling, optimized for high throughput inference across modern accelerators.

Serving with vLLM

vllm serve Solstice-AI/Qwen3.8-27B-TWIN-TURBO-Fable-Cold-Fusion-709-L-Uncensored-MXFP4 \
  --tensor-parallel-size 1 \
  --max-model-len 262144