Skip to content

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

New pull request New
28 Open 678 Closed
28 Open 678 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(dsv4/sparse_attn): per-row inverse-RoPE to avoid a5 gather lowering bug bug Something isn't working
#801 opened Jul 20, 2026 by lwDavid Contributor Loading…
@lwDavid
5
CI: run all self-hosted jobs on the host via one shared setup action
#800 opened Jul 20, 2026 by luohuan19 Contributor Loading…
1
8
perf(qwen3-14b decode): fold RoPE into the CANN FAI attention extern
#796 opened Jul 18, 2026 by ChaoWao Collaborator Loading…
8
feat(deepseek-v4): Define dynamic prefill attention contract
#793 opened Jul 17, 2026 by hashiqiqixian Contributor Loading…
20
perf(dsv4 decode swa): deferred attention RMSNorm and dep cleanup
#791 opened Jul 16, 2026 by zhangqi-chen Collaborator Loading…
4
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
Make Qwen3 device top-k selection exact
#787 opened Jul 16, 2026 by zmnobug Contributor Loading…
8
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
perf(qwen3-14b TQ decode): rotated-space attention, drop per-step KV …
#773 opened Jul 14, 2026 by sunghajung6688 Contributor Loading…
4
PoC: fold dsv4/moe dispatch/combine wait into syncall barriers enhancement New feature or request
#768 opened Jul 13, 2026 by lwDavid Contributor • 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
perf(dsv4/moe): launch routed expert stages once per expert
#738 opened Jul 9, 2026 by zhangqi-chen Collaborator • Draft
1
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
Share device-side LM head across decode and prefill
#711 opened Jul 7, 2026 by sjduan Contributor Loading…
2
perf(dsv4): qproj wq_b transpose + serving contract; sparse_attn/indexer tuning
#676 opened Jul 2, 2026 by Hzfengsy Member Loading…
6
perf(dsv4 qkv): precompute q-head rope factors
#648 opened Jun 30, 2026 by wangqin1723-max Contributor Loading…
4
perf(dsv4 prefill): avoid indexer pool token scan
#618 opened Jun 25, 2026 by zhaozhaozz Contributor Loading…
2
Add: vector (AIV) row_sum variant for qwen3-14b K/V projection
#586 opened Jun 23, 2026 by Inspiron-st Collaborator Loading…
7
Refactor: scope dsv4 decode+prefill kernels under auto_scope=False
#585 opened Jun 23, 2026 by zhangqi-chen Collaborator Loading…
9
feat(deepseek/v4): gather-free split-half RoPE for the decode path (draft, decode-only) enhancement New feature or request
#564 opened Jun 20, 2026 by Hzfengsy Member • Draft
@lwDavid
4
feat(world_model): Add T5, CLIP, Vae, DiT pypto3.0 kernels
#520 opened Jun 12, 2026 by xzhxzhxzh123 Collaborator Loading…
12
ProTip! Follow long discussions with comments:>50.