{"payload":{"feedbackUrl":"https://github.com/orgs/community/discussions/53140","repo":{"id":718781516,"defaultBranch":"main","name":"pytorch","ownerLogin":"alexjwilliams","currentUserCanPush":false,"isFork":true,"isEmpty":false,"createdAt":"2023-11-14T19:38:49.000Z","ownerAvatar":"https://avatars.githubusercontent.com/u/15970774?v=4","public":true,"private":false,"isOrgOwned":false},"refInfo":{"name":"","listCacheKey":"v0:1712006048.0","currentOid":""},"activityList":{"items":[{"before":null,"after":"5847b207c4ff8118cd1f947e8baa5124ee25018d","ref":"refs/heads/v2.2.2-bfloat16-rnn","pushedAt":"2024-04-01T21:14:08.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"Add bfloat16 CUDA support to RNN (#116927)\n\nFixes #116925\nFixes #116763\n\nPull Request resolved: https://github.com/pytorch/pytorch/pull/116927\nApproved by: https://github.com/malfet","shortMessageHtmlLink":"Add bfloat16 CUDA support to RNN (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2068752495\" data-permission-text=\"Title is private\" data-url=\"https://github.com/pytorch/pytorch/issues/116927\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/pytorch/pytorch/pull/116927/hovercard\" href=\"https://github.com/pytorch/pytorch/pull/116927\">pytorch#116927</a>)"}},{"before":"a8e7c98cb95ff97bb30a728c6b2a1ce6bff946eb","after":null,"ref":"refs/heads/v2.1.2-fix-autocast","pushedAt":"2024-03-18T21:27:29.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"}},{"before":null,"after":"8d5474853030373257a6d21b796fa87be5e0700d","ref":"refs/heads/v2.2.1-bfloat16-rnn","pushedAt":"2024-03-18T21:25:41.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"Add bfloat16 CUDA support to RNN (#116927)\n\nFixes #116925\nFixes #116763\n\nPull Request resolved: https://github.com/pytorch/pytorch/pull/116927\nApproved by: https://github.com/malfet","shortMessageHtmlLink":"Add bfloat16 CUDA support to RNN (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2068752495\" data-permission-text=\"Title is private\" data-url=\"https://github.com/pytorch/pytorch/issues/116927\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/pytorch/pytorch/pull/116927/hovercard\" href=\"https://github.com/pytorch/pytorch/pull/116927\">pytorch#116927</a>)"}},{"before":null,"after":"a8e7c98cb95ff97bb30a728c6b2a1ce6bff946eb","ref":"refs/heads/v2.1.2-fix-autocast","pushedAt":"2024-03-18T21:25:41.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"Revert \"Require less alignment for attn bias (#114173) (#114837)\"\n\nThis reverts commit 59656491f3b1da809312942872cce010337504b0.","shortMessageHtmlLink":"Revert \"Require less alignment for attn bias (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2003063768\" data-permission-text=\"Title is private\" data-url=\"https://github.com/pytorch/pytorch/issues/114173\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/pytorch/pytorch/pull/114173/hovercard\" href=\"https://github.com/pytorch/pytorch/pull/114173\">pytorch#114173</a>) (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2017830826\" data-permission-text=\"Title is private\" data-url=\"https://github.com/pytorch/pytorch/issues/114837\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/pytorch/pytorch/pull/114837/hovercard\" href=\"https://github.com/pytorch/pytorch/pull/114837\">pytorc…</a>"}},{"before":"b18d8d4595aa6e0768eedd5fc7d4a4402c567181","after":"773ae817f7bd1d428bc1a0bc83ca56f84e0d6a87","ref":"refs/heads/main","pushedAt":"2024-03-18T21:08:21.000Z","pushType":"push","commitsCount":2819,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"Batch Norm Consolidation (#116092)\n\n**Summary:**\n\nThis commit simplifies the existing decomposition hierarchy\nof batch norm ops by adding a single, backend agnostic op:\n`batch_norm_with_update`. The existing hierarchy looks like:\n\n```\naten.batch_norm ->\naten._batch_norm_impl_index ->\n[\n  aten.native_batch_norm ->\n  aten._native_batch_norm_legit (export only) ->\n  _batch_norm_legit_cpu/cuda (kernels, export only) ->\n  _batch_norm_cpu/cuda (kernels)\n] OR\n[ aten.cudnn_batch_norm ] OR\n[ aten.miopen_batch_norm ]\n```\n\nAside from complexity, an important problem with the\nabove decomposition hierarchy is cuda numerics in\nexport flows. We observed significantly worse convergence\nwhen training a mobilenetv2-like model when using the\n`_batch_norm_cuda` kernel instead of the `cudnn_batch_norm`\nkernel. This means users who export their models on CPU\nfirst then move the models to cuda later may silently\nsee worse accuracies even when cudnn is installed,\nbecause they are using the worse kernel. This issue is\nsummarized in https://github.com/pytorch/pytorch/issues/111384.\n\nInstead, the new hierarchy proposed by consolidating\nexisting batch norm ops will look like:\n\n```\naten.batch_norm ->\naten.batch_norm_with_update ->\n[ _batch_norm_cpu (kernel) ] OR\n[ _batch_norm_cuda (kernel) ] OR\n[ cudnn_batch_norm (kernel) ] OR\n[ miopen_batch_norm (kernel) ]\n```\n\nThe new op `batch_norm_with_update` hides backend\nimplementation details and automatically picks the right\nkernel based on what is installed. This commit also adds\nthe following variants to this op:\n\n```\nbatch_norm_with_update_functional\nbatch_norm_with_update.out\nbatch_norm_no_update\nbatch_norm_no_update.out\nbatch_norm_backward\n```\n\nNote that this commit only adds this op and its variants,\nbut does not actually change the decomps to produce these\nops in the graph. This will be done after the 2 week FC\nwindow, and the ops used in the old stack is planned to\nbe removed after the 6 month BC window.\n\nTest Plan: `OpInfo` tests for `batch_norm_with_update`.\n\nReviewers: albanD, bdhirsh\n\nSubscribers: albanD, bdhirsh, supriyar\n\nTasks: https://github.com/pytorch/pytorch/issues/111384\n\nDifferential Revision: [D54805279](https://our.internmc.facebook.com/intern/diff/D54805279)\nCo-authored-by: Tugsbayasgalan Manlaibaatar <tmanlaibaatar@fb.com>\nPull Request resolved: https://github.com/pytorch/pytorch/pull/116092\nApproved by: https://github.com/bdhirsh, https://github.com/albanD","shortMessageHtmlLink":"Batch Norm Consolidation (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2048225865\" data-permission-text=\"Title is private\" data-url=\"https://github.com/pytorch/pytorch/issues/116092\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/pytorch/pytorch/pull/116092/hovercard\" href=\"https://github.com/pytorch/pytorch/pull/116092\">pytorch#116092</a>)"}},{"before":"2a8a7425be6227bc6edd513244d72687df20896f","after":"b18d8d4595aa6e0768eedd5fc7d4a4402c567181","ref":"refs/heads/main","pushedAt":"2024-01-02T19:02:22.000Z","pushType":"push","commitsCount":1546,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"Add a wrapper to transform a NumPy function into a PyTorch function (#114610)\n\nA less general version of this wrapper was used in the keynote on\n`torch.compile(numpy)`. We expose a generic version of the wrapper\nthat works seamlessly with `torch.compile`.\n\nPull Request resolved: https://github.com/pytorch/pytorch/pull/114610\nApproved by: https://github.com/albanD","shortMessageHtmlLink":"Add a wrapper to transform a NumPy function into a PyTorch function (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2012833338\" data-permission-text=\"Title is private\" data-url=\"https://github.com/pytorch/pytorch/issues/114610\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/pytorch/pytorch/pull/114610/hovercard\" href=\"https://github.com/pytorch/pytorch/pull/114610\">p…</a>"}},{"before":"6d2994d1f3c04cd87030e7a3f1643507e7404f85","after":"6f69861c77931b3fdc64b94cd63bf8b3cf692881","ref":"refs/heads/v2.1.0-modify-CUDAGraph","pushedAt":"2023-11-17T05:08:17.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"add get_reference method","shortMessageHtmlLink":"add get_reference method"}},{"before":"554778e4e51b882bb3de817bdaa23334513aed47","after":"e75427b42ac7d734e3c1b34b922777c0eb382185","ref":"refs/heads/modify-CUDAGraph","pushedAt":"2023-11-17T05:08:17.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"add get_reference method","shortMessageHtmlLink":"add get_reference method"}},{"before":"038b71d219de2152b0032d3d456a3d6d90a07bc5","after":"554778e4e51b882bb3de817bdaa23334513aed47","ref":"refs/heads/modify-CUDAGraph","pushedAt":"2023-11-14T23:57:48.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"turn graph_exec_ attribute into getter","shortMessageHtmlLink":"turn graph_exec_ attribute into getter"}},{"before":"cffa6ff4ed087e403b1564187f3cff7278f51615","after":"6d2994d1f3c04cd87030e7a3f1643507e7404f85","ref":"refs/heads/v2.1.0-modify-CUDAGraph","pushedAt":"2023-11-14T23:57:48.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"turn graph_exec_ attribute into getter","shortMessageHtmlLink":"turn graph_exec_ attribute into getter"}},{"before":null,"after":"038b71d219de2152b0032d3d456a3d6d90a07bc5","ref":"refs/heads/modify-CUDAGraph","pushedAt":"2023-11-14T23:39:36.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"turn capture_stream_ attribute into getter method","shortMessageHtmlLink":"turn capture_stream_ attribute into getter method"}},{"before":"4d5cfe262df147e9ccfc6efa8523891b2ad2ce99","after":null,"ref":"refs/heads/modify-CUDAGraph","pushedAt":"2023-11-14T23:39:23.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"}},{"before":null,"after":"cffa6ff4ed087e403b1564187f3cff7278f51615","ref":"refs/heads/v2.1.0-modify-CUDAGraph","pushedAt":"2023-11-14T23:39:04.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"turn capture_stream_ attribute into getter method","shortMessageHtmlLink":"turn capture_stream_ attribute into getter method"}},{"before":"05edc34190a6b9b6fdcf22c3ce57822b965d53fa","after":null,"ref":"refs/heads/v2.1.0-modify-CUDAGraph","pushedAt":"2023-11-14T23:38:49.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"}},{"before":"22b6092acbd7e2b8ef403b70267e5b08825e132f","after":"05edc34190a6b9b6fdcf22c3ce57822b965d53fa","ref":"refs/heads/v2.1.0-modify-CUDAGraph","pushedAt":"2023-11-14T23:26:07.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"turn capture_stream_ attribute into getter method","shortMessageHtmlLink":"turn capture_stream_ attribute into getter method"}},{"before":"c65472de12b63b7290a57812330703e7fe663952","after":"4d5cfe262df147e9ccfc6efa8523891b2ad2ce99","ref":"refs/heads/modify-CUDAGraph","pushedAt":"2023-11-14T23:26:07.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"turn capture_stream_ attribute into getter method","shortMessageHtmlLink":"turn capture_stream_ attribute into getter method"}},{"before":"637fe06571fe2a69e63fb00ba3c3fa8eb80b639d","after":"22b6092acbd7e2b8ef403b70267e5b08825e132f","ref":"refs/heads/v2.1.0-modify-CUDAGraph","pushedAt":"2023-11-14T22:56:57.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"expose other members to python","shortMessageHtmlLink":"expose other members to python"}},{"before":"f1beaf0e25738ca9ea0649d6e2ece0c860b2927d","after":"c65472de12b63b7290a57812330703e7fe663952","ref":"refs/heads/modify-CUDAGraph","pushedAt":"2023-11-14T22:53:10.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"expose other members to python","shortMessageHtmlLink":"expose other members to python"}},{"before":null,"after":"637fe06571fe2a69e63fb00ba3c3fa8eb80b639d","ref":"refs/heads/v2.1.0-modify-CUDAGraph","pushedAt":"2023-11-14T20:17:08.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"expose has_graph_exec_ to python","shortMessageHtmlLink":"expose has_graph_exec_ to python"}},{"before":null,"after":"f1beaf0e25738ca9ea0649d6e2ece0c860b2927d","ref":"refs/heads/modify-CUDAGraph","pushedAt":"2023-11-14T20:17:08.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"alexjwilliams","name":null,"path":"/alexjwilliams","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15970774?s=80&v=4"},"commit":{"message":"expose has_graph_exec_ to python","shortMessageHtmlLink":"expose has_graph_exec_ to python"}}],"hasNextPage":false,"hasPreviousPage":false,"activityType":"all","actor":null,"timePeriod":"all","sort":"DESC","perPage":30,"cursor":"djE6ks8AAAAEJN0-sAA","startCursor":null,"endCursor":null}},"title":"Activity · alexjwilliams/pytorch"}