Skip to content

Issues: Dao-AILab/flash-attention

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

632 Open 562 Closed
632 Open 562 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
Assignee
Filter by who’s assigned
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Issues list

Running flash_attn/flash_attn_triton_amd/bench.py with sequence length > 4096 causes RuntimeError: Triton Error [CUDA]: an illegal memory access was encountered
#1440 opened Jan 12, 2025 by jiqimaoke
IncompatibleTypeErrorImpl('invalid operands of type pointer<int64> and triton.language.int32')
#1439 opened Jan 11, 2025 by wuyouliaoxi
FA3 forward performance regression on H200
#1438 opened Jan 10, 2025 by complexfilter
3
FA3 does not work with torch.compile
#1435 opened Jan 10, 2025 by nighting0le01
FA3 regression on H100 80GB?
#1432 opened Jan 9, 2025 by bastianhagedorn
8
[flash attn v2] Why V uses no-swizzle layout for registers?
#1429 opened Jan 8, 2025 by phantaurus
1
version `GLIBCXX_3.4.29' not found
#1428 opened Jan 8, 2025 by zhanghanxing2022
ERROR: No matching distribution found for flash-attn==2.6.3+cu123torch2.4cxx11abifalse
#1423 opened Jan 6, 2025 by carolynsoo
1
Unable to install flash_attn on H100 with CUDA 12.5
#1422 opened Jan 6, 2025 by ghadiaravi13
Unable to install flash-attn even if I first install torch alone
#1421 opened Jan 3, 2025 by ytxmobile98
4
Is there a plan to support flash_attn_varlen_backward with fp8
#1420 opened Jan 3, 2025 by gaodaheng
1
Encounter some problems when building wheel
#1418 opened Jan 2, 2025 by ZarkPanda
flash_attn_with_kvcache discrepancy slicing kv_cache / cache_seqlens
#1417 opened Jan 1, 2025 by jeromeku
RuntimeError: Error compiling objects for extension
#1415 opened Dec 27, 2024 by ProgramerSalar
2
looking for a test to verify cache correctness in flash_attn_with_kvcache
#1414 opened Dec 26, 2024 by chakpongchung
2
Performance Impact of Using Three Warps per Group (WG) in FA3 Compared to Two WGs
#1413 opened Dec 24, 2024 by ziyuhuang123
1
UnboundLocalError: local variable 'out' referenced before assignment
#1412 opened Dec 24, 2024 by chuangzhidan
3
Can't intall it
#1411 opened Dec 24, 2024 by TherrenceF
1
Impact of Register Spills on FA3 Kernel Performance
#1410 opened Dec 24, 2024 by ziyuhuang123
1
FA 2.4.2 is falling unitest on A6000 and A5880
#1409 opened Dec 23, 2024 by BoxiangW
4
Why Does FA3 Use Registers Instead of Directly Accessing SMEM with WGMMA on SM90?
#1407 opened Dec 23, 2024 by ziyuhuang123
1
4 Failing test_flash_attn_output_fp8 tests on H100
#1404 opened Dec 20, 2024 by BioGeek
3
Understanding sync and arrive in FA3 Store Function
#1401 opened Dec 19, 2024 by ziyuhuang123
Understanding the Role of arrive in NamedBarrier Synchronization
#1400 opened Dec 19, 2024 by ziyuhuang123
1
The execution order between GEMM0 of the next iteration and GEMM1 of the current iteration in Pingpong scheduling pipeline for overlapping gemms and softmax between warpgroups
#1398 opened Dec 19, 2024 by tengdecheng
ProTip! Mix and match filters to narrow down what you’re looking for.