{"full_name":"Dao-AILab/flash-attention","owner":"Dao-AILab","name":"flash-attention","shard":"1","url":"https://github.com/Dao-AILab/flash-attention","description":"Fast and memory-efficient exact attention","homepage":"","topics":[],"tag_pages":[],"language":"Python","license":"BSD-3-Clause","avatar":"https://avatars.githubusercontent.com/u/139507659?v=4","created_at":"2022-05-19T21:22:06Z","pushed_at":"2026-11-15T06:15:15Z","metrics":{"stars":25299,"forks":3231,"watchers":144,"issues":1860,"prs":1240,"releases":131,"issues_open":1051,"issues_closed":809,"prs_open":354,"prs_closed":274,"prs_merged":612,"discussions":25},"growth":{"day":{"stars":6,"forks":2,"base_day":"2026-11-15"},"week":{"stars":54,"forks":27,"base_day":"2026-11-04"},"month":{"stars":152,"forks":75,"base_day":"2026-10-17"}},"history":{"dates":["2026-09-06","2026-09-07","2026-09-08","2026-09-09","2026-09-10","2026-09-11","2026-09-12","2026-09-13","2026-09-14","2026-09-15","2026-09-16","2026-09-17","2026-09-18","2026-09-19","2026-09-20","2026-09-21","2026-09-22","2026-09-23","2026-09-24","2026-09-25","2026-09-26","2026-09-27","2026-09-28","2026-09-29","2026-09-30","2026-10-01","2026-10-02","2026-10-03","2026-10-04","2026-10-05","2026-10-06","2026-10-07","2026-10-08","2026-10-09","2026-10-10","2026-10-11","2026-10-12","2026-10-13","2026-10-14","2026-10-15","2026-10-16","2026-10-17","2026-10-18","2026-10-19","2026-10-20","2026-10-21","2026-10-22","2026-10-23","2026-10-24","2026-10-25","2026-10-26","2026-10-27","2026-10-28","2026-10-29","2026-10-30","2026-10-31","2026-11-01","2026-11-02","2026-11-03","2026-11-04","2026-11-10","2026-11-11","2026-11-12","2026-11-13","2026-11-14","2026-11-15","2026-11-16"],"stars":[24838,24841,24847,24856,24857,24868,24882,24890,24890,24899,24906,24920,24930,24949,24968,24976,24984,24992,25000,25004,25008,25012,25017,25022,25027,25035,25042,25053,25063,25070,25083,25091,25096,25100,25105,25108,25115,25115,25121,25132,25137,25147,25152,25155,25161,25174,25183,25191,25202,25206,25215,25219,25219,25217,25221,25225,25233,25235,25239,25245,25264,25275,25283,25284,25292,25293,25299],"forks":[3039,3040,3044,3047,3052,3058,3058,3060,3065,3068,3070,3070,3077,3080,3085,3087,3089,3094,3094,3096,3099,3100,3104,3104,3105,3108,3111,3116,3119,3125,3127,3137,3139,3143,3145,3146,3147,3149,3150,3151,3152,3156,3162,3163,3174,3175,3178,3184,3184,3186,3188,3192,3190,3193,3194,3197,3198,3200,3202,3204,3214,3217,3220,3221,3224,3229,3231],"watchers":[149,149,150,150,150,150,150,150,150,150,150,150,150,150,150,149,149,149,149,149,149,149,148,148,148,148,148,148,148,149,149,149,149,148,148,148,148,148,148,148,148,148,148,148,148,148,148,148,148,146,146,145,145,145,145,145,145,145,145,145,null,145,145,145,145,144,144],"issues_open":[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,1051,1051,1051,1052,1054,1051],"prs_open":[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,351,352,352,354,355,354]},"boards":{},"same_owner":[],"issues":[{"number":595,"title":"Might be a solution to get built/compiles Flash Attention 2 on Windows","comments":87,"created_at":"2023-10-08T16:10:49Z"},{"number":1038,"title":"Build flash-attn takes a lot of time ","comments":77,"created_at":"2024-07-10T10:40:08Z"},{"number":246,"title":"No Module Named 'torch'","comments":77,"created_at":"2023-05-28T13:11:46Z"},{"number":1464,"title":"Add support for Cuda 12.8 and B200 GPUs","comments":55,"created_at":"2025-01-25T19:27:03Z"},{"number":1708,"title":"ImportError: /lib64/libc.so.6: version `GLIBC_2.32' not found","comments":44,"created_at":"2025-06-11T05:29:38Z"},{"number":945,"title":"[bug] build is verrrrrrrrrrrrrrrrrrrry slow","comments":42,"created_at":"2024-05-11T09:10:47Z"},{"number":358,"title":"Not able to install 2.0","comments":33,"created_at":"2023-07-20T11:04:15Z"},{"number":879,"title":"Has anyone successfully compiled this on ARM Linux (aarch64)?","comments":31,"created_at":"2024-03-07T23:42:08Z"}],"discussions":[{"number":2767,"title":"What should long-context support prove beyond maximum length?","comments":0,"category":"Q&A","created_at":"2026-08-07T09:51:18Z"},{"number":2693,"title":"Does FA4 support split-kv flashdecode with cudagraph?","comments":1,"category":"Q&A","created_at":"2026-07-02T06:45:59Z"},{"number":2644,"title":"文章寻求","comments":0,"category":"General","created_at":"2026-06-11T03:29:41Z"},{"number":2612,"title":"Source commit for FlashAttention 3.0.0b1","comments":1,"category":"Q&A","created_at":"2026-06-01T09:23:09Z"},{"number":2240,"title":"Routed Attention: Learning when to use O(N) conv vs O(N²) attention (75-99% compute savings)","comments":0,"category":"Show and tell","created_at":"2026-02-07T16:35:11Z"},{"number":2114,"title":"Why does FA3 use persistent blocks while FA2 doesn't?","comments":1,"category":"Q&A","created_at":"2025-12-31T07:49:03Z"},{"number":2112,"title":"When can I use the logsumexp I get from return_attn_probs?","comments":1,"category":"Q&A","created_at":"2025-12-30T18:25:11Z"},{"number":2012,"title":"Clarification: baseline GLIBC version for 2.8.x releases","comments":1,"category":"Q&A","created_at":"2025-11-14T02:55:31Z"}],"latest_release":{"name":"v2.8.3.post2","tag":"v2.8.3.post2","published_at":"2026-10-13T05:41:17Z","body":"Maintenance release of FlashAttention 2 on top of v2.8.3.post1. No kernel feature changes.\n\n- Backport #2622: `torch.load(..., weights_only=True)` in `flash_attn/utils/distributed.py` and `flash_attn/models` (CVE-2026-31253 / GHSA-7g5w-pq96-8c5w). #2974\n- Backport #2702: safe `tarfile` extraction in `hopper/setup.py` (CVE-2026-62239 / GHSA-w9r7-4gwr-j959). #2974\n- Backport #2260: fix int32 overflow in the FA2 backward `dq_accum` offsets, which caused an illegal memory access or NaN gradients when `batch * round128(seqlen_q) * nheads * headdim` exceeded 2^31 (#2987). #2994\n- `setup.py`: request cu13 prebuilt wheels for CUDA 13 torch.\n","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}