{"payload":{"feedbackUrl":"https://github.com/orgs/community/discussions/53140","repo":{"id":431039713,"defaultBranch":"master","name":"pytorch","ownerLogin":"awaelchli","currentUserCanPush":false,"isFork":true,"isEmpty":false,"createdAt":"2021-11-23T09:35:14.000Z","ownerAvatar":"https://avatars.githubusercontent.com/u/5495193?v=4","public":true,"private":false,"isOrgOwned":false},"refInfo":{"name":"","listCacheKey":"v0:1703915969.0","currentOid":""},"activityList":{"items":[{"before":"f488c790a854c092df0d5bdbb261126a022626d9","after":"8df7ba81358581f90ff4d3143a84a9fff5ef8c99","ref":"refs/heads/optimized-module-state-dict-new","pushedAt":"2024-01-02T00:01:28.000Z","pushType":"push","commitsCount":19,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"Merge branch 'main' into optimized-module-state-dict-new","shortMessageHtmlLink":"Merge branch 'main' into optimized-module-state-dict-new"}},{"before":"97891b184c12763f335fbe1ff63fab843edafab5","after":"abd80cbb15bdc67e829687ed99604c554b7ac1d2","ref":"refs/heads/master","pushedAt":"2024-01-02T00:00:38.000Z","pushType":"push","commitsCount":18,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"[Inductor] Decompose bmm if batch2's last dim size is 1 and coordinate_descent_tuning is enabled (#116582)\n\nWe found this perf optimization opportunity at https://github.com/pytorch-labs/gpt-fast/pull/71. This would bring 5%+ perf gain for Mixtral 8x7B on gpt-fast.\n\nPull Request resolved: https://github.com/pytorch/pytorch/pull/116582\nApproved by: https://github.com/lezcano","shortMessageHtmlLink":"[Inductor] Decompose bmm if batch2's last dim size is 1 and coordinat…"}},{"before":null,"after":"f488c790a854c092df0d5bdbb261126a022626d9","ref":"refs/heads/optimized-module-state-dict-new","pushedAt":"2023-12-30T05:59:29.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"Implement pass-through state_dict and load_state_dict for dynamo OptimizedModule","shortMessageHtmlLink":"Implement pass-through state_dict and load_state_dict for dynamo Opti…"}},{"before":"4af1c27fa806bfcbd76c93c2396953f2b1faf70e","after":"97891b184c12763f335fbe1ff63fab843edafab5","ref":"refs/heads/master","pushedAt":"2023-12-30T05:55:09.000Z","pushType":"push","commitsCount":23,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"[Dynamo] Trace autograd.function in dynamo when inputs require grad (#116358)\n\nFor training graphs (when inputs require grad), previously, we would speculate the forward and backward graph to determine if there are any graph breaks, side effect and etc but would not actually use these speculated graphs. We would just insert a call function node on the graph and later rely on autograd's tracing.\n\nThis approach does not work for more generalized graphs like graphs that include user defined triton kernels because autograd is not able to do the higher order function conversation.\n\nThis PR speculates the forward and backward functions and emits them in a HOF that later gets used via templating mechanism.\n\nWhile working on this PR, I have exposed some bugs in the current tracing due to trampoline functions losing the source information resulting in incorrect graphs being produced. I have fixed these source information bugs and killed the trampolines.\n\nPull Request resolved: https://github.com/pytorch/pytorch/pull/116358\nApproved by: https://github.com/jansel","shortMessageHtmlLink":"[Dynamo] Trace autograd.function in dynamo when inputs require grad (p…"}},{"before":"1b0bdee02ba38c920c91086ee3c219d1f7134070","after":null,"ref":"refs/heads/mmap-load-pathlib","pushedAt":"2023-12-28T23:37:58.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"}},{"before":"0bacd6096645423caa2b4034e2b63a94ff5db046","after":"1b0bdee02ba38c920c91086ee3c219d1f7134070","ref":"refs/heads/mmap-load-pathlib","pushedAt":"2023-12-28T20:00:22.000Z","pushType":"push","commitsCount":73,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"Merge branch 'master' into mmap-load-pathlib","shortMessageHtmlLink":"Merge branch 'master' into mmap-load-pathlib"}},{"before":"015bd0e0a189f929e469c6bc75fe1541c18a014d","after":"4af1c27fa806bfcbd76c93c2396953f2b1faf70e","ref":"refs/heads/master","pushedAt":"2023-12-28T19:59:30.000Z","pushType":"push","commitsCount":72,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"Migrate repr, deterministic state_dict test to OptimizerInfo (#116496)\n\nPull Request resolved: https://github.com/pytorch/pytorch/pull/116496\nApproved by: https://github.com/albanD\nghstack dependencies: #116471","shortMessageHtmlLink":"Migrate repr, deterministic state_dict test to OptimizerInfo (pytorch…"}},{"before":"97c16fa0f9c2480508ec5fe99ccc4a1f97b26ebe","after":"0bacd6096645423caa2b4034e2b63a94ff5db046","ref":"refs/heads/mmap-load-pathlib","pushedAt":"2023-12-24T11:55:31.000Z","pushType":"push","commitsCount":65,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"Merge branch 'master' into mmap-load-pathlib","shortMessageHtmlLink":"Merge branch 'master' into mmap-load-pathlib"}},{"before":"a734085a63f8b840c2ab36ab188cb842d731ddd8","after":"015bd0e0a189f929e469c6bc75fe1541c18a014d","ref":"refs/heads/master","pushedAt":"2023-12-23T19:39:42.000Z","pushType":"push","commitsCount":64,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"[Dynamo][10/N] Remove TorchVariable and is_allowed (#116312)\n\nAfter this refactor:\n* ```TorchVariable``` definition and all references are removed.\n* All ```is_allowed``` references except one are removed.\n - The only left one is in ```torch/_dynamo/decorators:_disallow_in_graph_helper```. It was called when users put ```disallow_in_graph``` decorator on a function. Since we use the lists in ```trace_rules``` to decide the function's trace rule, so the decorator would only be used as customer function rather than torch functions. I'll defer this to a separate decorator refactor PR.\n\nPull Request resolved: https://github.com/pytorch/pytorch/pull/116312\nApproved by: https://github.com/jansel","shortMessageHtmlLink":"[Dynamo][10/N] Remove TorchVariable and is_allowed (pytorch#116312)"}},{"before":"78d8fa23f05cdbd13b1aa2be93281d198e9beacb","after":"97c16fa0f9c2480508ec5fe99ccc4a1f97b26ebe","ref":"refs/heads/mmap-load-pathlib","pushedAt":"2023-12-21T01:28:25.000Z","pushType":"push","commitsCount":98,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"Merge branch 'master' into mmap-load-pathlib","shortMessageHtmlLink":"Merge branch 'master' into mmap-load-pathlib"}},{"before":"92998693a9455af6259cae468265f01cfff8810e","after":"a734085a63f8b840c2ab36ab188cb842d731ddd8","ref":"refs/heads/master","pushedAt":"2023-12-21T01:27:51.000Z","pushType":"push","commitsCount":97,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"[ONNX][Dort] Fix bug preventing running with OrtValueVector (#116124)\n\nPull Request resolved: https://github.com/pytorch/pytorch/pull/116124\nApproved by: https://github.com/justinchuby, https://github.com/titaiwangms\nghstack dependencies: #115945","shortMessageHtmlLink":"[ONNX][Dort] Fix bug preventing running with OrtValueVector (pytorch#…"}},{"before":"d87ff541055e3579bc633fe3ada88daaf8fd1712","after":"78d8fa23f05cdbd13b1aa2be93281d198e9beacb","ref":"refs/heads/mmap-load-pathlib","pushedAt":"2023-12-20T22:24:02.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"type guard","shortMessageHtmlLink":"type guard"}},{"before":"451b54b0807f88a1e2cf71b87252d2ade260b8f9","after":"d87ff541055e3579bc633fe3ada88daaf8fd1712","ref":"refs/heads/mmap-load-pathlib","pushedAt":"2023-12-19T15:06:56.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"add test","shortMessageHtmlLink":"add test"}},{"before":"f8d466cbc75329c241b12736d61f8489778d55c6","after":"451b54b0807f88a1e2cf71b87252d2ade260b8f9","ref":"refs/heads/mmap-load-pathlib","pushedAt":"2023-12-19T15:00:57.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"update","shortMessageHtmlLink":"update"}},{"before":null,"after":"f8d466cbc75329c241b12736d61f8489778d55c6","ref":"refs/heads/mmap-load-pathlib","pushedAt":"2023-12-19T14:58:12.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"support pathlib with mmap","shortMessageHtmlLink":"support pathlib with mmap"}},{"before":"4e2e0437ea483a02aebefeab60e8870658990a5b","after":"92998693a9455af6259cae468265f01cfff8810e","ref":"refs/heads/master","pushedAt":"2023-12-19T14:41:08.000Z","pushType":"push","commitsCount":1437,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"[inductor] Avoid bool being upcast to int (#109913)\n\nCurrently the inductor code for `x.any(-1)` does a this strange dance:\n```python\ntmp0 = tl.load(in_ptr0 + (r1 + (128*x0)), rmask & xmask)\ntmp1 = tmp0.to(tl.int64)\ntmp2 = (tmp1 != 0)\n```\n\nThis happens because `register_lowering` is doing type promotion with the\ndimension argument, and so promotes to `int64` which we then cast back to bool.\nA better fix would be to fix `register_lowering` but for now I just remove\nthe unnecessary type promotion from `aten.any`.\n\nIn the current code we also see:\n```python\n tmp5 = tl.where(rmask & xmask, tmp3, 0)\n```\nwhich promotes the boolean value to int since `0` is an int32 in triton.\nThis fixes it to generate a boolean constant instead.\n\nFinally there is also a triton bug where the `tl.load` itself upcasts to\n`tl.int8`. I fix this by adding an explicit cast to `tl.int1`. The final\nkernel code looks like:\n\n```python\ntmp0 = tl.load(in_ptr0 + (r1 + (128*x0)), rmask & xmask).to(tl.int1)\ntmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])\ntmp3 = tl.full([1, 1], 0, tl.int1)\ntmp4 = tl.where(rmask & xmask, tmp1, tmp3)\ntmp5 = triton_helpers.any(tmp4, 1)[:, None]\n\n```\n\nPull Request resolved: https://github.com/pytorch/pytorch/pull/109913\nApproved by: https://github.com/lezcano","shortMessageHtmlLink":"[inductor] Avoid bool being upcast to int (pytorch#109913)"}},{"before":"7e31ff01d431355e3205a59a31847c7cbb10c359","after":null,"ref":"refs/heads/feature/optimized-module-state-dict","pushedAt":"2023-12-13T06:40:21.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"}},{"before":"285375bd8315db88fba4118d9dfa88db0c044d9a","after":"7e31ff01d431355e3205a59a31847c7cbb10c359","ref":"refs/heads/feature/optimized-module-state-dict","pushedAt":"2023-11-10T18:26:49.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"avoid +1","shortMessageHtmlLink":"avoid +1"}},{"before":"32751db404626283a00a9d3db909208b0edb9d19","after":"285375bd8315db88fba4118d9dfa88db0c044d9a","ref":"refs/heads/feature/optimized-module-state-dict","pushedAt":"2023-11-10T18:22:25.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"extend test","shortMessageHtmlLink":"extend test"}},{"before":"8cd0d11589c3dd89927f3190e161155081874b4e","after":"32751db404626283a00a9d3db909208b0edb9d19","ref":"refs/heads/feature/optimized-module-state-dict","pushedAt":"2023-11-10T04:04:32.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"lintrunner","shortMessageHtmlLink":"lintrunner"}},{"before":"a8c9ac2d0b6bf439b768bb2c8c224c586c9c5394","after":null,"ref":"refs/heads/docathon/fsdp","pushedAt":"2023-11-10T03:20:54.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"}},{"before":"0d35cf986e802eb25410fc5ce7a1b00d2335612f","after":"8cd0d11589c3dd89927f3190e161155081874b4e","ref":"refs/heads/feature/optimized-module-state-dict","pushedAt":"2023-11-10T03:20:21.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"test","shortMessageHtmlLink":"test"}},{"before":null,"after":"0d35cf986e802eb25410fc5ce7a1b00d2335612f","ref":"refs/heads/feature/optimized-module-state-dict","pushedAt":"2023-11-10T02:35:29.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"Merge branch 'master' into feature/optimized-module-state-dict","shortMessageHtmlLink":"Merge branch 'master' into feature/optimized-module-state-dict"}},{"before":"4305435a4aa293a1de3c762fb2a04167760c8916","after":"a8c9ac2d0b6bf439b768bb2c8c224c586c9c5394","ref":"refs/heads/docathon/fsdp","pushedAt":"2023-11-09T21:59:59.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"undo space","shortMessageHtmlLink":"undo space"}},{"before":"c1750f410a64ee6a905ea4b0a7450634f1af2e95","after":"4305435a4aa293a1de3c762fb2a04167760c8916","ref":"refs/heads/docathon/fsdp","pushedAt":"2023-11-09T21:54:38.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"fix typo","shortMessageHtmlLink":"fix typo"}},{"before":"fbf7866ac98258a8130e14de3021d3160bc28af3","after":"4e2e0437ea483a02aebefeab60e8870658990a5b","ref":"refs/heads/master","pushedAt":"2023-11-09T21:51:12.000Z","pushType":"push","commitsCount":25,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"[fx] stylistic improvements for fx.split_module (#113373)\n\nWas overly verbose before. Less qualified / long names = more clarity\n\nPull Request resolved: https://github.com/pytorch/pytorch/pull/113373\nApproved by: https://github.com/wconstab","shortMessageHtmlLink":"[fx] stylistic improvements for fx.split_module (pytorch#113373)"}},{"before":"96e5718eea5c15f1d1c29291305e4498e031c72b","after":"c1750f410a64ee6a905ea4b0a7450634f1af2e95","ref":"refs/heads/docathon/fsdp","pushedAt":"2023-11-09T21:47:04.000Z","pushType":"push","commitsCount":103,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"Merge branch 'pytorch:main' into docathon/fsdp","shortMessageHtmlLink":"Merge branch 'pytorch:main' into docathon/fsdp"}},{"before":"acadb5500e0a16fd046772101a151eb5a92f51ef","after":"96e5718eea5c15f1d1c29291305e4498e031c72b","ref":"refs/heads/docathon/fsdp","pushedAt":"2023-11-09T21:46:34.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"Update torch/distributed/fsdp/fully_sharded_data_parallel.py\n\nCo-authored-by: Svetlana Karslioglu ","shortMessageHtmlLink":"Update torch/distributed/fsdp/fully_sharded_data_parallel.py"}},{"before":"a3a2486be842b20a72c48067d957b5b8d6913b13","after":"fbf7866ac98258a8130e14de3021d3160bc28af3","ref":"refs/heads/master","pushedAt":"2023-11-09T03:44:42.000Z","pushType":"push","commitsCount":78,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"[Inductor] Fallback scatter when src dtype is bf16 (#113204)\n\nbasic_gnn_gcn, basic_gnn_gin, basic_gnn_sage now pass\n\nPull Request resolved: https://github.com/pytorch/pytorch/pull/113204\nApproved by: https://github.com/eellison","shortMessageHtmlLink":"[Inductor] Fallback scatter when src dtype is bf16 (pytorch#113204)"}},{"before":"f314eef05bf85cc04cfd659791dd799ed1de215f","after":"acadb5500e0a16fd046772101a151eb5a92f51ef","ref":"refs/heads/docathon/fsdp","pushedAt":"2023-11-09T02:44:57.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"awaelchli","name":null,"path":"/awaelchli","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/5495193?s=80&v=4"},"commit":{"message":"lintrunner","shortMessageHtmlLink":"lintrunner"}}],"hasNextPage":true,"hasPreviousPage":false,"activityType":"all","actor":null,"timePeriod":"all","sort":"DESC","perPage":30,"cursor":"djE6ks8AAAAD1aM0bgA","startCursor":null,"endCursor":null}},"title":"Activity · awaelchli/pytorch"}