← sources

CUDA: enable sparse fa for qwen4 by am17an · Pull Request #28770 · ggml-org/llama.cpp

source: reddit (reddit:LocalLLaMA)published 2026-09-20 16:03ingested 2026-09-21 02:38

retained text

Another day, another Qwen Flash Next speedup 

document fetched 2026-09-21 00:36 · canonical https://reddit.com/r/LocalLLaMA/comments/1wlkxjw/cuda_enable_sparse_fa_for_qwen4_by_am17an_pull