Add optimized CUDA kernels (RMSNorm, MLP) and patched Gemma4 modeling files 42e38d8 verified Khanin Udomchoksakul commited on 4 days ago
