💥 TensorFold 1.2.2 just sped up Qwen3.8-27B EXL3: 38 to 42% faster prefill on … · @QwenAIFans · TwKit

@QwenAIFans

💥 TensorFold 1.2.2 just sped up Qwen3.8-27B EXL3: 38 to 42% faster prefill on GB10 (2K to 64K). Flash Next on M5 Max also +36 to 39% decode with SSD-backed embeddings. Caveat: release notes, not our bench. Want that prefill bump? via GitHub 👇 https://github.com/ashhart/TensorFold/releases

曝光 67 · 评论 0 · 点赞 0 · 书签 1