llm-d / llm-d
State of the art inference performance with modern accelerators on Kubernetes
llm-d / llm-d
julianmb / halofpx
High-performance Lemonade alternative for AMD Strix Halo & Radeon — latest MoE models (Ling-3.0-Flash, Qwen 3.8 Flash Next, Ornith 1.5), bleeding-edge RDNA 3.5 Wave64/ROCmFP4 kernels, and silicon-tuned profiles.