Skip to content

Pull requests: hw-native-sys/pypto-lib

New pull request New
34 Open 720 Closed
34 Open 720 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Fix: parameterize DeepSeek V4-Pro cases with the Pro preset
#851 opened Jul 28, 2026 by lwDavid Collaborator Loading…
6
Fix: clear LM-head signals after forward
#850 opened Jul 28, 2026 by high-cloud Collaborator Loading…
1
Fix(es): align expert_shared golden with kernel int32 matmul order enhancement New feature or request
#847 opened Jul 28, 2026 by lwDavid Collaborator Loading…
@lwDavid
3
Perf: defer DSv4-flash attention RMSNorm reduce past the QKV projection
#846 opened Jul 27, 2026 by Hzfengsy Member Loading…
3
Perf: hoist DSv4-flash RoPE interleave tables out of per-block scopes
#845 opened Jul 27, 2026 by Hzfengsy Member Loading…
4
Add: dynamic public batch (1..16) for Qwen3-14B decode
#841 opened Jul 27, 2026 by lyfne123 Contributor Loading…
11
Perf: accelerate DeepSeek V4 Flash prefill attention
#838 opened Jul 27, 2026 by MaxwellF1 Loading…
5
WIP: dsv4-pro Hybrid MXFP8–MXFP4 operator rewrite
#836 opened Jul 25, 2026 by yanghaoran29 Contributor • Draft
2 of 4 tasks
1
fix(deepseek-v4-flash): fit decode buffers for PTOAS planning
#827 opened Jul 24, 2026 by Little-oil Contributor Loading…
1
[Docs] PyPTO 性能优化案例：DeepSeek-V4 MoE 预处理算子融合与同步开销复盘
#825 opened Jul 24, 2026 by Little-oil Contributor Loading…
45 tasks
1
CI: disable all jobs, add A5 runner bring-up smoke job
#821 opened Jul 23, 2026 by luohuan19 Contributor Loading…
1
dsv4 decode_fwd eplb test
#814 opened Jul 22, 2026 by zhangqi-chen Collaborator • Draft
4
perf(dsv4/moe): predicate empty routed-expert stages
#809 opened Jul 21, 2026 by wangqin1723-max Contributor Loading…
5
fix(qwen3-14b): guard negative slot_mapping KV writes
#806 opened Jul 21, 2026 by Leaf-Salix Contributor • Draft
16
perf(dsv4 hc_pre): widen linear split-K LINEAR_OK 4->8 (align with hc_head)
#804 opened Jul 20, 2026 by wangqin1723-max Contributor Loading…
3
feat(deepseek-v4): Define dynamic prefill attention contract
#793 opened Jul 17, 2026 by hashiqiqixian Contributor Loading…
20
Fuse QKV prologue into Qwen3-14B paged attention
#789 opened Jul 16, 2026 by Little-oil Contributor Loading…
7
test ci
#788 opened Jul 16, 2026 by high-cloud Collaborator Loading…
1
perf(dsv4 hca decode): early-resolve gather_kv/merge_norm + L2 swimlane level flag
#785 opened Jul 15, 2026 by wangqin1723-max Contributor Loading…
5
PoC: fold dsv4/moe dispatch/combine wait into syncall barriers enhancement New feature or request
#768 opened Jul 13, 2026 by lwDavid Collaborator • Draft
@lwDavid
4
refactor(dsv4): port #734 rope_cs collapse to HCA/SWA decode sparse kernels
#763 opened Jul 13, 2026 by wangqin1723-max Contributor Loading…
4
Add direct CANN FAI decode attention for Qwen3
#755 opened Jul 12, 2026 by ChaoWao Collaborator Loading…
10
fix(dsv4): zero inactive prefill MoE tails
#744 opened Jul 10, 2026 by high-cloud Collaborator Loading…
3
Merge DSV4 compressor ratio modules
#730 opened Jul 8, 2026 by zhaozhaozz Contributor Loading…
12
Fix Qwen3 attention mask NaNs
#714 opened Jul 7, 2026 by superxf Contributor Loading…
1
6
ProTip! Find all pull requests that aren't related to any open issues with -linked:issue.