Tag
Red Hat AI released an NVFP4-quantized 35B MoE version of Qwen3.6 that retains 96.28% GSM8K accuracy while enabling 4-bit inference via vLLM.