hyukudan/strix-halo-engine-lab
Exact LLM inference engine written from scratch for AMD Strix Halo (gfx1151, ROCm/HIP). Qwen3.8-Flash-Next on one node or two replicas; DeepSeek V4.1 Flash across two nodes over USB4 with an optional RTX PRO. Speculative decoding bit-identical to greedy, up to 255 tok/s on coding chat. MiniMax and GLM next.
GitHub repository with 30 stars and 5 forks.
Language: C++
Topics: amd, cuda, deepseek, distributed-inference, gfx1151, gguf, hip, llm-inference, mixture-of-experts, mtp