DeepSeek-V4-Flash-0731 UD-IQ3_S 12.5 tok/s on RTX 3090 +128GB DDR5
英文标题：DeepSeek-V4-Flash-0731 UD-IQ3_S 12.5 tok/s on RTX 3090 +128GB DDR5
I managed to run DeepSeek-V4-Flash-0731 UD-IQ3_S in text-generation-webui with: RTX 3090 24 GB 128 GB DDR5 overclocked to 5600 MHz using AMD EXPO llama.cpp loader First, I had to use a rather brutal workaround: I replaced the llama.cpp binaries included with text-generation-webui by the late
打开原文相关 2 · 机会 2
中文
I managed to run DeepSeek-V4-Flash-0731 UD-IQ3_S in text-generation-webui with: RTX 3090 24 GB 128 GB DDR5 overclocked to 5600 MHz using AMD EXPO llama.cpp loader First, I had to use a rather brutal workaround: I replaced the llama.cpp binaries included with text-generation-webui by the late
中文全文翻译正在补齐，可先查看下方英文原文。
English / 原文
I managed to run DeepSeek-V4-Flash-0731 UD-IQ3_S in text-generation-webui with: RTX 3090 24 GB 128 GB DDR5 overclocked to 5600 MHz using AMD EXPO llama.cpp loader First, I had to use a rather brutal workaround: I replaced the llama.cpp binaries included with text-generation-webui by the late