Qwen@Alibaba_Qwen
45AI 编辑部评分,满分 100
2026-08-14 23:50· 14分钟前
AI 导读

通义千问发布 Qwen3.8-27B,采用与 2.4T 旗舰相同的混合骨干架构,但为稠密而非 MoE,单张 Blackwell GPU 即可运行。原生 262K 上下文可扩展至 1M,GB300 上可同时处理 6 条全长序列;MTP 草稿头内置于检查点,短提示词接受率 BF16 达 92.2%、FP8 达 84.8%。

One GPU, 1M context, Day-0 ready. Big props to the vLLM team for the seamless integration!👍 Try Qwen3.8-27B on vLLM: @vllm_project https://recipes.vllm.ai/Qwen/Qwen3.8-27B

vLLM🎉 Qwen3.8-27B is here from @Alibaba_Qwen, and the whole thing fits on a single GPU. Same hybrid backbone as the 2.4T flagship, dense instead of MoE. Day-0 supp...

来源:Qwen · x.com

Qwen · @Alibaba_Qwen · X·2026-08-14 23:50·14分钟前
AI 导读

通义千问发布 Qwen3.8-27B,采用与 2.4T 旗舰相同的混合骨干架构,但为稠密而非 MoE,单张 Blackwell GPU 即可运行。原生 262K 上下文可扩展至 1M,GB300 上可同时处理 6 条全长序列;MTP 草稿头内置于检查点,短提示词接受率 BF16 达 92.2%、FP8 达 84.8%。

One GPU, 1M context, Day-0 ready. Big props to the vLLM team for the seamless integration!👍 Try Qwen3.8-27B on vLLM: @vllm_project https://recipes.vllm.ai/Qwen/Qwen3.8-27B

vLLM🎉 Qwen3.8-27B is here from @Alibaba_Qwen, and the whole thing fits on a single GPU. Same hybrid backbone as the 2.4T flagship, dense instead of MoE. Day-0 supp...

来源:Qwen· x.com