DeepSeek-V4.1-Flash-1017 at 41 tok/s on one 3090
英文标题：DeepSeek-V4.1-Flash-1017 at 41 tok/s on one 3090
No, this is not a lobotomized 1-bit squeeze, its full Q2 nazeshinjite's DeepSeek-V4.1-Flash-1017 Q2 GGUF on a single 3090, natively on Windows no WSL. On an RTX 3090 with 192 GB of RAM I'm writing at 41 tok/s (average across multiple prompts) and the output agrees with DeepSeek's own API on th
打开原文相关 4 · 机会 2
中文
No, this is not a lobotomized 1-bit squeeze, its full Q2 nazeshinjite's DeepSeek-V4.1-Flash-1017 Q2 GGUF on a single 3090, natively on Windows no WSL. On an RTX 3090 with 192 GB of RAM I'm writing at 41 tok/s (average across multiple prompts) and the output agrees with DeepSeek's own API on th
中文全文翻译正在补齐，可先查看下方英文原文。
English / 原文
No, this is not a lobotomized 1-bit squeeze, its full Q2 nazeshinjite's DeepSeek-V4.1-Flash-1017 Q2 GGUF on a single 3090, natively on Windows no WSL. On an RTX 3090 with 192 GB of RAM I'm writing at 41 tok/s (average across multiple prompts) and the output agrees with DeepSeek's own API on th