Skip to content

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

New pull request New
39 Open 752 Closed
39 Open 752 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: stabilize MTP logits validation
#890 opened Aug 3, 2026 by Little-oil Contributor Loading…
5
Feat: add DeepSeek V4 Flash prefill context parallelism
#889 opened Aug 3, 2026 by sjduan Contributor Loading…
6
Feat: add operator scheduling analysis skills
#888 opened Aug 3, 2026 by Little-oil Contributor Loading…
10
Refactor: flatten models/ to one directory per model build
#886 opened Aug 3, 2026 by zhangqi-chen Collaborator Loading…
1
feat(dsv4): make MTP prefill chunk-safe
#883 opened Aug 1, 2026 by vegetabledoww Contributor Loading…
1
Perf: assemble SWA decode window in L1, drop the GM gather
#882 opened Jul 31, 2026 by Hzfengsy Member Loading…
5
CI: bench decode_attention_csa on fixed even/odd cards
#881 opened Jul 31, 2026 by luohuan19 Contributor Loading…
5
Fix: stop calling pypto's removed compile_for_test()
#880 opened Jul 31, 2026 by lyfne123 Contributor Loading…
1
Fix(deepseek-v4 pro): let MoE post_ffn use automatic dependency tracking
#879 opened Jul 31, 2026 by yanghaoran29 Contributor Loading…
2 of 3 tasks
1
[WIP] Add: L3 communication-computation example
#869 opened Jul 30, 2026 by Shenggan • Draft
1
Test: baseline prefill MTP CI control
#861 opened Jul 30, 2026 by Little-oil Contributor • Draft
1
Perf: defer DSv4-flash attention RMSNorm reduce past the QKV projection
#846 opened Jul 27, 2026 by Hzfengsy Member Loading…
3
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 Collaborator 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 Collaborator 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 Collaborator 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
ProTip! Adding no:label will show everything without a label.