Skip to content
GitLab
Explore
Sign in
洪城码力
2026pra-qwen
Repository
Branches
Overview
Active
Stale
All
candidate/segments32-official-ab-20260629
b62a667d
·
Tune Triton attention softmax segments for Qwen3.5 ROCm
·
Jun 29, 2026
candidate/segments32-midonly-20260701
6f35bd27
·
Gate Triton softmax segments for mid-length decode
·
Jul 01, 2026
candidate/greedy-top-token-20260701
3c231770
·
Add guarded greedy top-token fast path for Qwen3.5
·
Jul 01, 2026
candidate/bf16-gemv-whitelist-20260702
3dc26968
·
Add Qwen3.5 BF16 GEMV decode whitelist
·
Jul 02, 2026
candidate/bf16-gemv-bk512-20260704
5cfd57ae
·
Tune Qwen3.5 BF16 GEMV block size
·
Jul 05, 2026
candidate/attn-prefill-tile16-20260706
abbe4ec7
·
Tune Qwen3.5 ROCm prefill attention tile size
·
Jul 06, 2026
candidate/gemv-shape-meta-20260708
d6e8fd75
·
Tune Qwen3.5 BF16 GEMV shape meta
·
Jul 08, 2026
candidate/gemv-gdn-meta-20260708
42701ee5
·
Tune GDN BF16 GEMV launch meta
·
Jul 09, 2026
candidate/llmm1-fp32pair-v4-restricted-20260710
16dfd53b
·
Use ROCm LLMM1 FP32-pair for Qwen3.5 GEMV
·
Jul 10, 2026
candidate/llmm1-fp32pair-rpb2-20260710
79ce6035
·
Tune ROCm LLMM1 row blocking for Qwen3.5
·
Jul 10, 2026
candidate/flash-page1024-fa2-20260711
cea77d91
·
Use vendor FlashAttention2 for Qwen3.5 prefill
·
Jul 11, 2026
candidate/attention3d-pagecontig784-20260712
5600ecac
·
Optimize Qwen3.5 page-contiguous decode attention
·
Jul 12, 2026
candidate/attention2d-pagecontig784-prefill-20260712
5417714d
·
Add page-major Qwen3.5 prefill attention
·
Jul 12, 2026
candidate/gdn-nosplit-riskprobe-20260713
b7daefcd
·
Test GDN outer graph split removal
·
Jul 13, 2026
candidate/a23-longonly-page784-20260713
01f37a85
·
Add long-context page784 prefill attention path
·
Jul 13, 2026
candidate/qk-row8-v5-20260713
b79392d9
·
Fuse QK norm and rotary embedding for Qwen3.5 ROCm
·
Jul 13, 2026
candidate/qk-row8-v5-ba-20260713
ecb7c97b
·
Combine QK row8 fusion with GDN BA LLMM1
·
Jul 13, 2026
candidate/best88-a23-longonly-20260714
80d80155
·
Add long-only page784 prefill attention path
·
Jul 14, 2026
candidate/best88-page832-vendor-prefill-20260714
c734944a
·
Add page832 vendor prefill fast path
·
Jul 14, 2026
candidate/final-compliance-best888294-20260714
ab180f02
·
Polish Chinese review documentation and code comments
·
Jul 14, 2026
Prev
1
2
Next