"KOI team on AMD Developer Hackathon: ACT III Hackathon"
Team Idea
KOI is a CDNA-first adaptive inference engine for AMD Instinct GPUs. It profiles real LLM workloads, identifies whether performance is being lost in host dispatch, kernels, scheduling, or memory access, then automatically benchmarks and selects the fastest execution path. Starting with MI300X and ROCm, KOI is built around one goal: more tokens, lower latency, and higher GPU utilization from the same hardware. Find the wait. Remove the wait
Team Members
.png&w=256&q=75)
Unknown Hacker
thanakritwongkha25095
Submission
Team Leader hasn't made a submission yet