Skip to content
GitLab
Explore
Sign in
bug全改队
2025pra-bug全改队
Repository
Branches
Overview
Active
Stale
All
cai
default
b4a218f8
·
优化方案说明文档
·
Jul 15, 2026
lyc_fill_functor_opt_baseline
6c4df0e4
·
fix: prevent NaN/crash from uninitialized memory and OOB reads in GDN kernels
·
Jul 11, 2026
lyc_fill_functor_opt
066c4ad6
·
feat: optimize Linear Attention by pre-allocating buffer
·
Jul 11, 2026
lyc_qk_norm_rope_fusion
5556baad
·
feat: enable QK Norm + RoPE fusion for BW1000
·
Jul 11, 2026
dcu-vllm-deploy
5cb2ab00
·
refactor: 简化 per-head scale 初始化(移除冗余 _dev)
·
Jul 10, 2026
main
17350ead
·
feat: FLA_NUM_WARPS env var
·
Jul 10, 2026
cai_bk
7ec940a4
·
opt: fused_recurrent decode kernel num_warps 1→8 (3.5x 加速)
·
Jul 09, 2026
cai_gdn_runner
b1c15d12
·
opt: runner 激进优化 - 缓存 (_la, _prefix) 元组
·
Jul 08, 2026
cai_gdn
68e46d95
·
Revert "cai_gdn: 修复 per_slot_ssm_state 索引越界 bug"
·
Jul 07, 2026
cai_gdn_aggressive
ed134b65
·
aggressive + 方案G: 全局标志 + 最小化 runner
·
Jul 07, 2026
cai_gdn_runner_opt
687c2ca7
·
runner_opt: 保守版优化 - 缓存引用 + decode 时遍历
·
Jul 07, 2026
cai_gdn_diag4
f134fe24
·
diag4 fix: _forward_core decode 调用改用 state_indices
·
Jul 07, 2026
lyc_attn
64542c81
·
attn: hardcode prefill BLOCK_Q=8/num_stages=2 for BW1000 (remove os.environ)
·
Jul 06, 2026
cai_gdn_qwen3_no_triton
38a15134
·
fix: 移除 forward 路径中的 logger 调用
·
Jul 06, 2026
cai_triton_verify
7fcf0a56
·
fix: 恢复 @CustomOp.register 装饰器
·
Jul 06, 2026
cai_quant_v2_fix
24388168
·
chore: 移除意外提交的 .nfs 文件
·
Jul 06, 2026
cai_quant_v2
b76b9621
·
chore: 移除意外提交的 .nfs 文件
·
Jul 06, 2026
hwb
bd5f0cb5
·
roll back
·
Jul 06, 2026
cai_quant
1f0c442c
·
fix: 移除 forward 中的 logging 调用
·
Jul 05, 2026
cai_final
451d8c6b
·
fix: start_vllm.sh 清理系统 vllm 和 pyc 缓存
·
Jul 05, 2026
Prev
1
2
Next