{"payload":{"feedbackUrl":"https://github.com/orgs/community/discussions/53140","repo":{"id":683963097,"defaultBranch":"main","name":"mammoth","ownerLogin":"Helsinki-NLP","currentUserCanPush":false,"isFork":false,"isEmpty":false,"createdAt":"2023-08-28T06:48:14.000Z","ownerAvatar":"https://avatars.githubusercontent.com/u/35632158?v=4","public":true,"private":false,"isOrgOwned":true},"refInfo":{"name":"","listCacheKey":"v0:1722275254.0","currentOid":""},"activityList":{"items":[{"before":"70ac57da4cc155c3a4e5a0699f64f35dcc0256e9","after":"e6f564a3946ceb8da1c7fdd11fab078e8b056c22","ref":"refs/heads/feat/x-transformers","pushedAt":"2024-08-12T14:14:34.000Z","pushType":"push","commitsCount":3,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"},"commit":{"message":"WIP: translation is broken","shortMessageHtmlLink":"WIP: translation is broken"}},{"before":"89f31639d817b04b7469969ef67589c5918ee658","after":"70ac57da4cc155c3a4e5a0699f64f35dcc0256e9","ref":"refs/heads/feat/x-transformers","pushedAt":"2024-08-12T08:07:50.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"},"commit":{"message":"Model splitting and saving","shortMessageHtmlLink":"Model splitting and saving"}},{"before":"e09763f129a4aef098c78cb04d2fb84ec5929f86","after":"89f31639d817b04b7469969ef67589c5918ee658","ref":"refs/heads/feat/x-transformers","pushedAt":"2024-08-05T15:29:34.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"},"commit":{"message":"WIP: dimension ordering messed up","shortMessageHtmlLink":"WIP: dimension ordering messed up"}},{"before":null,"after":"e09763f129a4aef098c78cb04d2fb84ec5929f86","ref":"refs/heads/feat/x-transformers","pushedAt":"2024-07-29T17:47:34.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"},"commit":{"message":"WIP: Replace the custom Transformer implementation with x-transformers\n\nExternal dependencies for layer architectures #56","shortMessageHtmlLink":"WIP: Replace the custom Transformer implementation with x-transformers"}},{"before":"a147f7c712d22c6ed8ddd2861a2a0a041708acd2","after":null,"ref":"refs/heads/feats/pad_to_maxlen","pushedAt":"2024-06-13T09:18:10.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"}},{"before":"3d1e1a73d5208fbbf28ec1f1204f0e60b84e0886","after":null,"ref":"refs/heads/fix/optim-steps-saving","pushedAt":"2024-06-13T09:11:21.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"}},{"before":"20bdc2fd7597d0b25514d8f25a07c06b1d563848","after":"4a3277624ce44a056610e827f29ab28e100837c2","ref":"refs/heads/main","pushedAt":"2024-06-01T11:51:12.000Z","pushType":"pr_merge","commitsCount":2,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"},"commit":{"message":"Merge pull request #76 from Helsinki-NLP/fix/optim-steps-saving\n\noptimizer global step removal","shortMessageHtmlLink":"Merge pull request <a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2327504994\" data-permission-text=\"Title is private\" data-url=\"https://github.com/Helsinki-NLP/mammoth/issues/76\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/Helsinki-NLP/mammoth/pull/76/hovercard\" href=\"https://github.com/Helsinki-NLP/mammoth/pull/76\">#76</a> from Helsinki-NLP/fix/optim-steps-saving"}},{"before":null,"after":"3d1e1a73d5208fbbf28ec1f1204f0e60b84e0886","ref":"refs/heads/fix/optim-steps-saving","pushedAt":"2024-05-31T10:30:14.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"},"commit":{"message":"optimizer global step removal","shortMessageHtmlLink":"optimizer global step removal"}},{"before":"1c40c9b2c51249a7887cbca2ab03cf78aa45110f","after":"844f2d27b2b7cd7ea09087c2985545b338bf8b67","ref":"refs/heads/fix/gpu-assignment-split-corpora","pushedAt":"2024-05-27T14:00:19.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"},"commit":{"message":"When performing GPU assignment, keep track of split corpora","shortMessageHtmlLink":"When performing GPU assignment, keep track of split corpora"}},{"before":"20bdc2fd7597d0b25514d8f25a07c06b1d563848","after":"1c40c9b2c51249a7887cbca2ab03cf78aa45110f","ref":"refs/heads/fix/gpu-assignment-split-corpora","pushedAt":"2024-05-27T13:56:09.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"},"commit":{"message":"When performing GPU assignment, keep track of split corpora","shortMessageHtmlLink":"When performing GPU assignment, keep track of split corpora"}},{"before":null,"after":"20bdc2fd7597d0b25514d8f25a07c06b1d563848","ref":"refs/heads/fix/gpu-assignment-split-corpora","pushedAt":"2024-05-27T13:50:43.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"},"commit":{"message":"Fix conflict between fp16 and deterministic sampling\n\nDue to the removal of the grad hook, MultipleOptimizer no longer has a\nmethod step, it has been replaced with externally_managed_step which\ntakes information about which optimizers need to be stepped. This means\nthat it is no longer compatible with torch.cuda.amp.GradScaler.\n\nWhile fixing this issue, the MultipleOptimizer system was also\nrefactored.\n- MultipleOptimizer and the OpenNMT Optimizer wrapper switched places:\n  MultipleOptimizer now wraps the other one, instead of the reverse.\n- The OpenNMT Optimizer was renamed to SubOptimizer for clarity.\n- SubOptimizer handles learning rate scheduling and grad clipping.\n- MultipleOptimizer handles creation of multiple optimizers, grad scaling,\n  restoring from checkpoint, backward, zero_grad, deciding which\n  suboptimizers to step, and reporting.\n- Each optimizer now individually controls its learning rate schedule.\n  When new components with freshly initialized parameters are introduced\n  by the curriculum, they now apply warmup to the LR of these\n  parameters. This should improve stability.\n- As each optimizer has its own learning rate, it is not obvious what to\n  log in the report_training one-liner. Learning rate was removed.\n  Instead, all optimizers log their learning rates. This is currently\n  log spam, but will be lowered to debug in #70.\n\nEach sub-optimizer having its own GradScaler leads to multiple backward\npasses and RuntimeError. There can only be one GradScaler, which must\ntherefore be the responsibility of MultipleOptimizer.\n\nCloses: #71","shortMessageHtmlLink":"Fix conflict between fp16 and deterministic sampling"}},{"before":"8040d3bc5654626633872ba46bda24f3b011ad5c","after":"d359ed1ab8624340c8d2d3a06dfc9b5b0e424ad0","ref":"refs/heads/crude-queue-time-profiling","pushedAt":"2024-05-27T13:14:17.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"},"commit":{"message":"Crude profiling of time waiting for the multiprocessing queue\n\nA semicolon separated csv format can be extracted from logs using the\nmarker string \"QUEUE_PERF\".\nBoth the semaphore and the put/get are measured, for both producer and\nconsumer.","shortMessageHtmlLink":"Crude profiling of time waiting for the multiprocessing queue"}},{"before":"e06ac631732fef22431d42d46516ad8b8bdc42d3","after":null,"ref":"refs/heads/feat/multiple_optimizer","pushedAt":"2024-05-27T13:11:37.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"}},{"before":"4b9b5445b80e1170c7dbad9eea236b66d9a884af","after":"20bdc2fd7597d0b25514d8f25a07c06b1d563848","ref":"refs/heads/main","pushedAt":"2024-05-27T13:11:32.000Z","pushType":"pr_merge","commitsCount":3,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"},"commit":{"message":"Fix conflict between fp16 and deterministic sampling\n\nDue to the removal of the grad hook, MultipleOptimizer no longer has a\nmethod step, it has been replaced with externally_managed_step which\ntakes information about which optimizers need to be stepped. This means\nthat it is no longer compatible with torch.cuda.amp.GradScaler.\n\nWhile fixing this issue, the MultipleOptimizer system was also\nrefactored.\n- MultipleOptimizer and the OpenNMT Optimizer wrapper switched places:\n  MultipleOptimizer now wraps the other one, instead of the reverse.\n- The OpenNMT Optimizer was renamed to SubOptimizer for clarity.\n- SubOptimizer handles learning rate scheduling and grad clipping.\n- MultipleOptimizer handles creation of multiple optimizers, grad scaling,\n  restoring from checkpoint, backward, zero_grad, deciding which\n  suboptimizers to step, and reporting.\n- Each optimizer now individually controls its learning rate schedule.\n  When new components with freshly initialized parameters are introduced\n  by the curriculum, they now apply warmup to the LR of these\n  parameters. This should improve stability.\n- As each optimizer has its own learning rate, it is not obvious what to\n  log in the report_training one-liner. Learning rate was removed.\n  Instead, all optimizers log their learning rates. This is currently\n  log spam, but will be lowered to debug in #70.\n\nEach sub-optimizer having its own GradScaler leads to multiple backward\npasses and RuntimeError. There can only be one GradScaler, which must\ntherefore be the responsibility of MultipleOptimizer.\n\nCloses: #71","shortMessageHtmlLink":"Fix conflict between fp16 and deterministic sampling"}},{"before":null,"after":"e06ac631732fef22431d42d46516ad8b8bdc42d3","ref":"refs/heads/feat/multiple_optimizer","pushedAt":"2024-05-27T12:17:44.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"},"commit":{"message":"Fix conflict between fp16 and deterministic sampling\n\nDue to the removal of the grad hook, MultipleOptimizer no longer has a\nmethod step, it has been replaced with externally_managed_step which\ntakes information about which optimizers need to be stepped. This means\nthat it is no longer compatible with torch.cuda.amp.GradScaler.\n\nWhile fixing this issue, the MultipleOptimizer system was also\nrefactored.\n- MultipleOptimizer and the OpenNMT Optimizer wrapper switched places:\n  MultipleOptimizer now wraps the other one, instead of the reverse.\n- The OpenNMT Optimizer was renamed to SubOptimizer for clarity.\n- SubOptimizer handles learning rate scheduling and grad clipping.\n- MultipleOptimizer handles creation of multiple optimizers, grad scaling,\n  restoring from checkpoint, backward, zero_grad, deciding which\n  suboptimizers to step, and reporting.\n- Each optimizer now individually controls its learning rate schedule.\n  When new components with freshly initialized parameters are introduced\n  by the curriculum, they now apply warmup to the LR of these\n  parameters. This should improve stability.\n- As each optimizer has its own learning rate, it is not obvious what to\n  log in the report_training one-liner. Learning rate was removed.\n  Instead, all optimizers log their learning rates. This is currently\n  log spam, but will be lowered to debug in #70.\n\nEach sub-optimizer having its own GradScaler leads to multiple backward\npasses and RuntimeError. There can only be one GradScaler, which must\ntherefore be the responsibility of MultipleOptimizer.\n\nCloses: #71","shortMessageHtmlLink":"Fix conflict between fp16 and deterministic sampling"}},{"before":"7a14eaa5c76a33d1aa697d5c10bd66785a5719b5","after":null,"ref":"refs/heads/feat/translation-server","pushedAt":"2024-05-27T07:47:55.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"}},{"before":"503a9989096b89f728b178388ca1ddd66e2817be","after":"4b9b5445b80e1170c7dbad9eea236b66d9a884af","ref":"refs/heads/main","pushedAt":"2024-05-27T07:46:54.000Z","pushType":"pr_merge","commitsCount":12,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"},"commit":{"message":"flake8","shortMessageHtmlLink":"flake8"}},{"before":"89186bb68665770f00eba33e039293500fe54537","after":"7a14eaa5c76a33d1aa697d5c10bd66785a5719b5","ref":"refs/heads/feat/translation-server","pushedAt":"2024-05-27T07:44:03.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"},"commit":{"message":"flake8","shortMessageHtmlLink":"flake8"}},{"before":"dc010391407a5d0deb8a31d04ff24e3056a5ca07","after":"503a9989096b89f728b178388ca1ddd66e2817be","ref":"refs/heads/main","pushedAt":"2024-05-26T09:38:36.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"},"commit":{"message":"bugfix","shortMessageHtmlLink":"bugfix"}},{"before":"ba3abb577adb48ae00414dfedb162d640f781fd1","after":null,"ref":"refs/heads/feats/data-restoration","pushedAt":"2024-05-22T14:53:31.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"}},{"before":"c6995b7a57a11566ab9c69b4836f2be6b125e29d","after":"dc010391407a5d0deb8a31d04ff24e3056a5ca07","ref":"refs/heads/main","pushedAt":"2024-05-22T14:53:27.000Z","pushType":"pr_merge","commitsCount":10,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"},"commit":{"message":"Merge pull request #64 from Helsinki-NLP/feats/data-restoration\n\ndata state restoration","shortMessageHtmlLink":"Merge pull request <a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2207433040\" data-permission-text=\"Title is private\" data-url=\"https://github.com/Helsinki-NLP/mammoth/issues/64\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/Helsinki-NLP/mammoth/pull/64/hovercard\" href=\"https://github.com/Helsinki-NLP/mammoth/pull/64\">#64</a> from Helsinki-NLP/feats/data-restoration"}},{"before":"af4b2123c1e04a667c4ed82e5eb4d83b1103e866","after":"ba3abb577adb48ae00414dfedb162d640f781fd1","ref":"refs/heads/feats/data-restoration","pushedAt":"2024-05-22T14:48:32.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"},"commit":{"message":"oop,s translator was broken","shortMessageHtmlLink":"oop,s translator was broken"}},{"before":"b6595225c7dfc56e3e183abe8a549337b6bc0f85","after":"af4b2123c1e04a667c4ed82e5eb4d83b1103e866","ref":"refs/heads/feats/data-restoration","pushedAt":"2024-05-22T14:35:11.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"},"commit":{"message":"minor fixes","shortMessageHtmlLink":"minor fixes"}},{"before":"9308e7dbd729d19242bfb444eb007986ffdfebbc","after":"b6595225c7dfc56e3e183abe8a549337b6bc0f85","ref":"refs/heads/feats/data-restoration","pushedAt":"2024-05-21T07:46:31.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"},"commit":{"message":"smoketesting ok?","shortMessageHtmlLink":"smoketesting ok?"}},{"before":"cc807c721ce05cf33f37922622b4569f2deefe13","after":"9308e7dbd729d19242bfb444eb007986ffdfebbc","ref":"refs/heads/feats/data-restoration","pushedAt":"2024-05-21T07:14:59.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"},"commit":{"message":"unplug model vocab (remnants of opts cleaning?)","shortMessageHtmlLink":"unplug model vocab (remnants of opts cleaning?)"}},{"before":"279876e75341bb935fd816a92db61018980bb97b","after":"cc807c721ce05cf33f37922622b4569f2deefe13","ref":"refs/heads/feats/data-restoration","pushedAt":"2024-05-21T06:58:05.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"},"commit":{"message":"runable","shortMessageHtmlLink":"runable"}},{"before":"149795218ecfe1282a81d451c8fbd2f90b83de35","after":"279876e75341bb935fd816a92db61018980bb97b","ref":"refs/heads/feats/data-restoration","pushedAt":"2024-05-20T16:45:42.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"},"commit":{"message":"fix after merge","shortMessageHtmlLink":"fix after merge"}},{"before":"ec347bca0200cf7650b39748c2e8d01b3eb2ddb7","after":"149795218ecfe1282a81d451c8fbd2f90b83de35","ref":"refs/heads/feats/data-restoration","pushedAt":"2024-05-20T15:50:04.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"},"commit":{"message":"making tests valid","shortMessageHtmlLink":"making tests valid"}},{"before":"edaa78b3d4748a6f87da4731f287f4878988b87d","after":"ec347bca0200cf7650b39748c2e8d01b3eb2ddb7","ref":"refs/heads/feats/data-restoration","pushedAt":"2024-05-20T15:44:12.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"TimotheeMickus","name":"Timothee Mickus","path":"/TimotheeMickus","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/45401828?s=80&v=4"},"commit":{"message":"fix test expectations (add line index support)","shortMessageHtmlLink":"fix test expectations (add line index support)"}},{"before":"bda983a445c94aa3c74c7160875301b347f2bc4b","after":null,"ref":"refs/heads/feat/reimplement-fixed-batching","pushedAt":"2024-05-20T15:22:30.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"Waino","name":"Stig-Arne Grönroos","path":"/Waino","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/1087483?s=80&v=4"}}],"hasNextPage":true,"hasPreviousPage":false,"activityType":"all","actor":null,"timePeriod":"all","sort":"DESC","perPage":30,"cursor":"djE6ks8AAAAEmDH8wwA","startCursor":null,"endCursor":null}},"title":"Activity · Helsinki-NLP/mammoth"}