Skip to content

Pull requests: deepseek-ai/FlashMLA

New pull request New
49 Open 53 Closed
49 Open 53 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 duplicate TRAP_ONLY_DEVICE_ASSERT definition
#209 opened Aug 7, 2026 by tianchongchong Loading…
[Draft]SM100 persistent sparse attention for V4 CSA/HCA
#207 opened Aug 4, 2026 by uchihatmtkinu Collaborator Loading…
Mark auxiliary LSE output non-differentiable
#203 opened Aug 3, 2026 by Mr-Neutr0n Loading…
Preserve input validation under Python optimization
#202 opened Jul 31, 2026 by Mr-Neutr0n Loading…
Fix scheduler state after backend failure
#201 opened Jul 31, 2026 by Mr-Neutr0n Loading…
[SM90] Add packed selected-KV sparse decode
#200 opened Jul 26, 2026 by xyz2606 Loading…
Add sparse MLA backward (DSA top-k), sliding-window attention + attention sink, and block-sparse MLA forward/backward for SM100
#198 opened Jul 16, 2026 by kmswin1 Loading…
sparse_prefill_fwd (sm100): skip trailing fully-invalid top-k blocks when topk_length is not provided
#196 opened Jul 10, 2026 by zkyue Loading…
Add pyproject.toml with build-system metadata
#195 opened Jul 9, 2026 by Functionhx Loading…
Become ABI Stable with LibTorch 2.11+
#193 opened Jul 2, 2026 by janeyx99 Loading…
1
1
fix typo
#191 opened Jun 25, 2026 by wangtianxia-sjtu Loading…
Dsv4 sparse prefill optimization by skipping odd tail block
#187 opened Jun 9, 2026 by Njuapp • Draft
Raise on unsupported head_dim in SM100 dense FMHA dispatch instead of silently returning uninitialized output
#185 opened May 30, 2026 by toffee-desuwa Loading…
3
Fix SM100 V32 sparse decode FP8 handling
#184 opened May 10, 2026 by jianglan89 Loading…
Add SM80 (Ampere/A100) dense MLA decode kernel
#183 opened May 1, 2026 by bzantium Loading…
6 tasks done
docs: fix sign of online softmax rescaling factors
#177 opened Apr 23, 2026 by tangpanyu Loading…
Install CUDAGuard before Arch() in api entry points (fixes #158)
#176 opened Apr 19, 2026 by yurekami • Draft
2 tasks
1
fix: tcgen05.mma.ws assert typo
#175 opened Apr 16, 2026 by KevinZeng08 Loading…
Fix potential memory leak in sparse_attn_decode_interface
#174 opened Apr 14, 2026 by NJX-njx Loading…
3 tasks
2
[sm90][sparse-prefill] optimize phase1 producer scheduling
#170 opened Mar 11, 2026 by huangzhilin-hzl Loading…
Fix cuda 13.0 compile
#162 opened Feb 9, 2026 by MARD1NO Loading…
Fix grammar in docstring: 'has' -> 'have' and 'affect' -> 'effect'
#157 opened Jan 24, 2026 by bodhisilberling Loading…
Fix grammar in docstring: 'has' -> 'have' and 'affect' -> 'effect'
#156 opened Jan 22, 2026 by bodhisilberling Loading…
docs: add troubleshooting guide, tensor shapes reference, and FAQ
#147 opened Jan 2, 2026 by yurekami Loading…
3 tasks done
fix: add SM120 (RTX PRO 6000 Blackwell) support (#124)
#146 opened Dec 29, 2025 by yurekami Loading…
2 tasks done
1
ProTip! Follow long discussions with comments:>50.