CUDA: fuse FFN gate/up matmuls and GLU in MMQ with interleaved gate/up rows by anujj · Pull Request #29948 · ggml-org/llama.cpp

🌐 链接: https://www.reddit.com/r/LocalLLaMA/comments/1xey0uw/cuda_fuse_ffn_gateup_matmuls_and_glu_in_mmq_with/
🏷️ 分组: Reddit｜AI·自托管·VPS
🕒 时间: 2026-10-25 17:35:12
Reddit
From the LocalLLaMA community on Reddit: CUDA: fuse FFN gate/up matmuls and GLU in MMQ with interleaved gate/up rows by anujj ·…
From the LocalLLaMA community on Reddit: CUDA: fuse FFN gate/up matmuls and GLU in MMQ with interleaved gate/up rows by anujj ·…
Explore this post and more from the LocalLLaMA community