OtherU

介绍

DeepSeek-V4-Flash inference for AMD Strix Halo

翻译准备中

取自仓库 README

README

正在加载完整 README。 以下为已保存的摘录。

Ember is designed exclusively for AMD Strix Halo (gfx1151). It is an intentionally focused engine for one APU platform and one model family, allowing its CPU, GPU, unified memory, and optional XDNA2 NPU path to be tuned together. It is not a general-purpose inference engine and does not support other GPU or NPU architectures. Ember is a C inference server for DeepSeek-V4-Flash. It provides OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, and legacy Completions APIs on one local endpoint.