返回Ssignal1708/21 15:00·6天前·1 signalsArchived topic · 归档话题，来源已停止追踪Qwen3.8-27B at 262K context on a Strix Halo + RTX 3090 Ti: 9.5 -> 153 tok/s, and it beats a dual-3090 vLLM box on HumanEval收藏分享导出 MarkdownNVIDIA端侧推理qwen3热度趋势趋势数据积累中百分比基于当前可用热度信号，而非评论数或独立用户人数。