Red Hat AI(@RedHat_AI):Qwen3.8-2.4T-A95B, compressed two ways at once: 25% of the experts pruned with REAP, and the rest quantized to NVFP4. Even with a quarter of the experts gone and 4-bit weights, GPQA Diamond holds at 91.5 vs 92.6 for the full-precision base. ~99% recovery. Serve on @vllm

Red Hat AI

@RedHat_AI

Accelerating AI innovation with open platforms and community. The future of AI is open.

加入 May 2018

2.1K 正在關注 12.3K 粉絲

Red Hat AI@RedHat_AI

2026.08.20 13:33

Qwen3.8-2.4T-A95B, compressed two ways at once: 25% of the experts pruned with REAP, and the rest quantized to NVFP4. Even with a quarter of the experts gone and 4-bit weights, GPQA Diamond holds at 91.5 vs 92.6 for the full-precision base. ~99% recovery. Serve on @vllm_project: