Betelgeuse running DeepSeek V4.1 Flash Q8_XL (1017) at 19 tok/s decode & 62 tok/s prefill on an RTX 3060 12GB + 16GB RAM: How we finally fixed the prefill bottleneck on out-of-core MoEs

🌐 链接: https://www.reddit.com/r/LocalLLaMA/comments/1xqfekg/betelgeuse_running_deepseek_v41_flash_q8_xl_1017/
🏷️ 分组: Reddit｜AI·自托管·VPS
🕒 时间: 2026-11-08 03:58:54
Reddit
From the LocalLLaMA community on Reddit: Betelgeuse running DeepSeek V4.1 Flash Q8_XL (1017) at 19 tok/s decode & 62 tok/s prefill…
From the LocalLLaMA community on Reddit: Betelgeuse running DeepSeek V4.1 Flash Q8_XL (1017) at 19 tok/s decode & 62 tok/s prefill…
Explore this post and more from the LocalLLaMA community