Skip to content

Pull requests: SmallDoges/flash-dmattn

New pull request New
15 Open 153 Closed
15 Open 153 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

Optimize triton version: GQA, mask/bias broadcasting, skip inactive tiles, and stability fixes
#200 opened Oct 27, 2025 by LoserCheems Loading…
2 of 5 tasks
@ftgreat @Thanksyy
@juliohsu @SNHuan @Evanwu1125 @wubingheng111 @zacliu2023
13
[FEATURE] Support total_k-head broadcastable mask/bias in varlen batch inference
#187 opened Oct 10, 2025 by Copilot AI • Draft
1
@LoserCheems
[BUG FIX] Optimize top-k mask construction: prevent unsafe gradient flow and eliminate unnecessary memory allocations
#184 opened Oct 4, 2025 by LoserCheems Loading…
8 tasks done
1
@ftgreat @Thanksyy
@juliohsu @SNHuan @Evanwu1125 @wubingheng111 @zacliu2023
4
Fix INF issue in bf16 backward pass with safer value clamping
#181 opened Sep 26, 2025 by Copilot AI • Draft
1
@LoserCheems
Implement Unified Sparse Mask Strategy with Block-Level Skipping
#164 opened Sep 11, 2025 by Copilot AI • Draft
1
@LoserCheems
Add optional mask & bias inputs with adaptive computation skipping
#162 opened Sep 10, 2025 by Copilot AI • Draft
1
@LoserCheems
Fix installation error with improved backend fallback and error messages
#151 opened Sep 6, 2025 by Copilot AI • Draft
1
@LoserCheems
[PERFORMANCE] Implement Linear KV Cache optimization for inference acceleration
#143 opened Sep 1, 2025 by Copilot AI • Draft
1
@LoserCheems
[FEATURE] Reduce compute bubbles in backward skip path
#133 opened Aug 29, 2025 by Copilot AI • Draft
1
@LoserCheems
Fix NaN/Inf values in dV gradients during backward pass
#122 opened Aug 24, 2025 by Copilot AI • Draft
1
@LoserCheems
Add comprehensive documentation for efficient attention mask handling in long sequences
#120 opened Aug 23, 2025 by Copilot AI • Draft
1
@LoserCheems
2
Document query-agnostic design and implications for associative recall tasks
#119 opened Aug 23, 2025 by Copilot AI • Draft
1
@LoserCheems
[PERFORMANCE] Implement Backward Launch Template Optimization for SM 8.9 and SM 9.0
#112 opened Aug 17, 2025 by Copilot AI • Draft
1
@LoserCheems
Optimize sparse logic with block-level Tensor Core utilization
#110 opened Aug 15, 2025 by Copilot AI • Draft
1
@LoserCheems
1
Implement dBias gradient computation in CUDA backward kernel
#73 opened Jul 22, 2025 by Copilot AI • Draft
1
@LoserCheems
ProTip! Filter pull requests by the default branch with base:main.