2 projects
isat-tuner
ISAT: Inference Stack Auto-Tuner — 91-command production inference engine with KV cache compression, disaggregated prefill-decode, MoE expert parallelism, model routing/cascade, multi-modal pipelines, RAG engine, long context (100K+), inference compiler, SLO scheduling, prompt caching, AI watermarking, token economics, session management, shadow deployment, and edge-cloud hybrid inference.
stxh-setup-amd
Automated setup for AMD Strix Halo APU: ROCm, MIGraphX, custom ORT, and AI model deployment.
