#
hipblaslt
Here are 2 public repositories matching this topic...
A complete GEMM tuning loop on AMD MI325X — profile → hotspot ranking → demand extraction → tuning → load verification → paired A/B. +4.1~5.1% request throughput on Qwen3-8B/SGLang, plus the six checks that make the number defensible.
performance-engineering benchmarking profiling gemm rocm amd-gpu kernel-tuning mi300 llm-inference sglang hipblaslt
-
Updated
Oct 3, 2026 - Python
Add this topic to your repo
To associate your repository with the hipblaslt topic, visit your repo's landing page and select "manage topics."