Skip to content
GitLab
Explore
Sign in
算得又快又队
2025pra-QwenOpt
Repository
Branches
Overview
Active
Stale
All
feature/selective-gemv-3shape
370b7b3a
·
Optimize Qwen3.5 selective GEMV shapes
·
Jul 09, 2026
feature/decode-only-v2-prefill-gemv
de12954a
·
Merge prefill GEMV optimizations into decode-only v2
·
Jul 07, 2026
feature/decode-only-v2
0c1d91fb
·
fix(decode): gate B/E opts on actual seq_lens not max_model_len during cudagraph capture
·
Jul 07, 2026
feat/KVcache-merge
4989269f
·
Merge branch 'feat/KVcache1' into 'feat/KVcache-merge'
·
Jul 06, 2026
feature/prefill-gemv-opt
688df4c5
·
Optimize ROCm prefill attention query tiling
·
Jul 05, 2026
feature/prefill-decode-opt
1e9e86bf
·
Guard Qwen3.5 decode tile sizing during init
·
Jul 05, 2026
feature/decode-bw-opt
cbed807d
·
Optimize ROCm decode KV tile size for Qwen3.5
·
Jul 05, 2026
feature/prefill-query-tiling
199e2527
·
Optimize ROCm prefill attention query tiling
·
Jul 03, 2026
main
default
protected
b6266ac4
·
chore: disable dependabot
·
Jun 27, 2026