Devotd Github

aristo94/gsqhalo.cpp

在面向 GSQ 量化 Qwen3.8-Flash-Next 的 llama.cpp 分支文档中,恢复了“restore oracle”一节,用于验证从 SSD 读回的对话是否与原始对话完全一致地延续。

GitHub 项目说明: llama.cpp fork for GSQ-quantized Qwen3.8-Flash-Next on AMD Strix Halo (gfx1151): faster ROCm prefill, MTP and 4×256K chats with Q8_0 KV and one KV stream per slot, persistent KV cache on SSD. Fork of halo-box/strix-llama.cpp.

在 GitHub 查看 ↗
星标
★ 11
复刻
0
语言
C++
许可证
MIT

文章

1