新发现
aristo94/gsqhalo.cpp — SSD 提示缓存恢复预言机文档
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。
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 ↗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。
没有符合条件的动态。