Issues

• Unsloth на Windows для fine-tuning.

  facebookresearch/xformers#1405 · Filter by author caprall116 opened on Aug 14, 2026

• Blackwell Support

  facebookresearch/xformers#1395 · Filter by author HowToDoThis opened on Apr 29, 2026

• 我的电脑需要这个版本xFormers PyTorch 2.11.0+cu130 Python 3.12.12的whl文件有哪位大佬编译成功的吗

  facebookresearch/xformers#1386 · Filter by author bbs888-Git opened on Mar 30, 2026

• 0.0.34 claims to support PyTorch>=2.10 but it doesn't?

  facebookresearch/xformers#1374 · Filter by author levicki opened on Jan 24, 2026
  15

• Provide official prebuilt CUDA-compatible wheels and fix common build/install failures for xFormers

  facebookresearch/xformers#1372 · Filter by author Shaheer66 opened on Jan 20, 2026

• Add a canonical long-context attention boundary benchmark (one command + metric)

  facebookresearch/xformers#1369 · Filter by author StanByriukov02 opened on Jan 9, 2026
  2

• 我的电脑需要这个版本xFormers PyTorch 2.8.0+cu129 Python 3.12.9的whl文件有哪位大佬编译成功的吗

  facebookresearch/xformers#1366 · Filter by author ggwwbb0310 opened on Jan 4, 2026
  2

• Wheels for rocm7.1

  facebookresearch/xformers#1365 · Filter by author Calandracas606 opened on Dec 19, 2025
  3

• Build flash attn on Windows failed

  facebookresearch/xformers#1363 · Filter by author silvertuanzi opened on Dec 3, 2025

• Incorrect CUDA_MINIMUM_COMPUTE_CAPABILITY check in fmha/flash3.py

  facebookresearch/xformers#1361 · Filter by author klondenberg-bioptimus opened on Dec 2, 2025
  1

• [BUG] cutlass attention completely blocked on Blackwell with no fallback path

  facebookresearch/xformers#1357 · Filter by author kbenkhaled opened on Nov 25, 2025
  1

• requires device with capability < (8, 0) but your GPU has capability (12, 0) (too new)

  facebookresearch/xformers#1356 · Filter by author lucasjinreal opened on Nov 17, 2025
  10

• Why not make it as fast as FlashInfer or FlashAttention-2?

  facebookresearch/xformers#1355 · Filter by author saymyname77 opened on Nov 11, 2025

• Suggestion for explicit compatibility matrix

  facebookresearch/xformers#1354 · Filter by author xiaoran007 opened on Nov 9, 2025
  9

• Flash Attention 3 torch extension namespace conflict

  facebookresearch/xformers#1348 · Filter by author georgkaleido opened on Oct 20, 2025

• Support for differentiable soft attention masks (float mask in [0,1], not in log-space)

  facebookresearch/xformers#1343 · Filter by author enszhou opened on Oct 7, 2025
  2

• Issue with 50-Series Graphics Card Compatibility for xFormers

  facebookresearch/xformers#1342 · Filter by author KATe-AGI opened on Oct 5, 2025
  4

• python 3.13 cuda 12.9 torch 2.8 Compilation failed

  facebookresearch/xformers#1341 · Filter by author wzgrx opened on Oct 4, 2025

• Windows build is broken

  facebookresearch/xformers#1340 · Filter by author levicki opened on Sep 29, 2025
  1

• flash_attn version selection for GPU under SM_89 (e.g., L20)

  facebookresearch/xformers#1338 · Filter by author Luchuanzhao opened on Sep 29, 2025

• It seems xformers/csrc/sparse24/sparse24_gemm_sm90.cu not follow TMA alignment

  facebookresearch/xformers#1335 · Filter by author jasl opened on Sep 23, 2025
  1

• Tracker for migrating to torch stable ABI

  facebookresearch/xformers#1333 · Filter by author mikaylagawarecki opened on Sep 18, 2025

• When xFormers is compiled with Flash Attention it fails on RTX 5000 series - CUDA error (/workspace/xfrm2/third_party/flash-attention/hopper/flash_fwd_launch_template.h:188): no kernel image is available for execution on the device

  facebookresearch/xformers#1329 · Filter by author FurkanGozukara opened on Sep 6, 2025
  7

• Which CUDA archs are supported in pypi release? 0.0.32.post2-cp39-abi3-win_amd64

  facebookresearch/xformers#1328 · Filter by author FurkanGozukara opened on Sep 6, 2025

• Python 3.10.11, CUDA 12.9 and Torch 2.8 - error : 'ptxas' died with status 0xC0000005 (ACCESS_VIOLATION)

  facebookresearch/xformers#1327 · Filter by author FurkanGozukara opened on Sep 6, 2025
  4