I merged fixes for quantized KV cache into my DeepSeek V4 branch
英文标题：I merged fixes for quantized KV cache into my DeepSeek V4 branch
Check it out: https://github.com/fairydreaming/llama.cpp/tree/dsv4 They are PRs #25247, #25303 (mine) and #25202 (from am17an) but I omitted some padding changes from the last one that I think are not necessary. So if it crashes for you let me know. You can now fit the antirez IQ2XXS model with 1M c
打开原文相关 2 · 机会 2
中文
Check it out: https://github.com/fairydreaming/llama.cpp/tree/dsv4 They are PRs #25247, #25303 (mine) and #25202 (from am17an) but I omitted some padding changes from the last one that I think are not necessary. So if it crashes for you let me know. You can now fit the antirez IQ2XXS model with 1M c
中文全文翻译正在补齐，可先查看下方英文原文。
English / 原文
Check it out: https://github.com/fairydreaming/llama.cpp/tree/dsv4 They are PRs #25247, #25303 (mine) and #25202 (from am17an) but I omitted some padding changes from the last one that I think are not necessary. So if it crashes for you let me know. You can now fit the antirez IQ2XXS model with 1M c