microsoft
/
ptgnn
зеркало из https://github.com/microsoft/ptgnn.git
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность

Граф коммитов

  • 9a44061ff1 added conda recipe master Sarthak Pati 2022-01-31 08:46:15 -0500
  • d1e80e3870
         added conda recipe Sarthak Pati 2022-01-31 08:46:15 -0500
  • 5d3ef87287
         Update oom.py Miltos 2022-01-05 09:09:57 +0000
  • 5e173e9fbf Fix invalid version number for torch-scatter in GH workflow mir-am 2021-12-17 22:52:37 +0100
  • f45dc7b062 Specify version of torch-scatter in GH workflow to fix the compilation error mir-am 2021-12-16 23:58:10 +0100
  • ff5034398e Fix the required PyTorch version in dependencies and docs mir-am 2021-12-08 14:21:20 +0100
  • 54b6f8a641 Fix invalid version number for torch-scatter in GH workflow mir-am 2021-12-17 22:52:37 +0100
  • 93ca3c9a64 Specify version of torch-scatter in GH workflow to fix the compilation error mir-am 2021-12-16 23:58:10 +0100
  • 56dbd9a38e Fix the required PyTorch version in dependencies and docs mir-am 2021-12-08 14:21:20 +0100
  • 53af8080f1 Use trick shown here https://pytorch.org/tutorials/recipes/recipes/tuning_guide.html#use-parameter-grad-none-instead-of-model-zero-grad-or-optimizer-zero-grad v0.10.4 Miltos Allamanis 2021-10-21 20:10:52 +0100
  • ef13a9fec0 Option to catch occasional CUDA OOMs to allow for more robust training. v0.10.3 Miltos Allamanis 2021-09-10 18:59:55 +0100
  • 8871865645 Option to catch occasional CUDA OOMs to allow for more robust training. Miltos Allamanis 2021-09-10 18:59:55 +0100
  • bf72e7f18a
         Improve einsum notation Miltos 2021-09-03 11:30:18 +0100
  • 8da3c1d91e Remove redundant `np.array` invocation. Miltos Allamanis 2021-07-28 13:38:46 +0100
  • 6c55894406 Fix reported training loss when using AMP v0.10.2 Miltos Allamanis 2021-07-25 13:39:13 +0000
  • 2ba4358c53 Fix reported training loss when using AMP Miltos Allamanis 2021-07-25 13:39:13 +0000
  • f62a6165dd Fixes. v0.10.1 Miltos Allamanis 2021-07-21 11:27:05 +0000
  • 03ab80499b Explicitly set the starting epoch index to the restored one. Miltos Allamanis 2021-07-21 10:23:09 +0100
  • 764a7038ff Fix bugs in distributed saving of optimizer state. Miltos Allamanis 2021-07-20 14:19:35 +0000
  • 3bd5966e88 Allow restoring optimizer/scheduler state. Miltos Allamanis 2021-07-20 13:47:42 +0100
  • adc19d7691 Fixes. Miltos Allamanis 2021-07-21 11:27:05 +0000
  • 656a9184b1 Explicitly set the starting epoch index to the restored one. Miltos Allamanis 2021-07-21 10:23:09 +0100
  • 5020bb7f6f Improve some code comments about message-passing layers. Miltos Allamanis 2021-07-20 17:21:38 +0100
  • 658ea91bdb Improve some code comments about message-passing layers. Miltos Allamanis 2021-07-20 17:21:38 +0100
  • 97e327d3c2 Fix bugs in distributed saving of optimizer state. Miltos Allamanis 2021-07-20 14:19:35 +0000
  • 3c7c066d46 Allow restoring optimizer/scheduler state. Miltos Allamanis 2021-07-20 13:47:42 +0100
  • 27b3102184
         Merge pull request #13 from microsoft/dev/miallama/distributedtrainer v0.10.0 Miltos 2021-07-14 14:43:11 +0100
  • ce0d8f6814 Fix AMP which was inconsistently implemented: https://pytorch.org/docs/stable/notes/amp_examples.html dev/miallama/distributedtrainer Miltos Allamanis 2021-07-14 12:40:59 +0100
  • 6edb8aeeba Reduce gpu->cpu transfers when progress bar is disabled. Miltos Allamanis 2021-07-14 11:36:08 +0000
  • c724767872 Remove unneeded gpu->cpu transfers during metrics update. Miltos Allamanis 2021-07-14 12:04:03 +0100
  • cd02705dab Fix how validation-improved hooks work. Miltos Allamanis 2021-07-09 12:08:15 +0100
  • 66f0046517 Improved hooks/AML support fixes. Miltos Allamanis 2021-07-09 12:00:11 +0100
  • 80e8879469 Reduce insanely long buffering. Miltos Allamanis 2021-07-08 19:16:04 +0100
  • ed8aac4bb2 Merge branch 'dev/miallama/distributedtrainer' of github.com:microsoft/ptgnn into dev/miallama/distributedtrainer Miltos Allamanis 2021-07-08 17:23:23 +0100
  • 95727b9802 Minor AML fixes. Miltos Allamanis 2021-07-08 17:22:10 +0100
  • b22b91fca7 Restore models in-cpu. Miltos Allamanis 2021-07-08 15:45:47 +0000
  • 060765d6d6 Remove uncessary `enumerate` Miltos Allamanis 2021-07-08 10:29:10 +0100
  • b37f3830c0 Fix typos Miltos Allamanis 2021-07-08 10:28:34 +0100
  • 948ba35646 Bypass deadlock by throwing on join() Miltos Allamanis 2021-07-07 22:22:39 +0000
  • a3a841676a Repair bug. Miltos Allamanis 2021-07-07 13:55:33 +0000
  • 25e9b29e28 Polish distributed trainer. Miltos Allamanis 2021-07-07 08:03:23 +0000
  • 12496e4ee4 WIP implementation of distributed data parallel training. Miltos Allamanis 2021-07-06 22:26:59 +0000
  • 8df6d85368
         Merge pull request #12 from microsoft/dev/miallam/ecg-implementation Miltos 2021-04-16 08:05:45 +0100
  • 5e7e146cc0 An implementation of ECG Miltos Allamanis 2021-04-12 13:54:48 +0100
  • d30cb0d585
         Merge pull request #10 from microsoft/dev/miallama/better-names Miltos 2021-03-16 14:24:30 +0000
  • b36fd52f37 Add more informative aliases for the GnnOutput Miltos Allamanis 2021-03-16 13:46:33 +0000
  • 9f0c0f0019 Fix AMP issue in PNA. Miltos Allamanis 2021-03-12 21:13:57 +0000
  • 207074d246
         Merge pull request #8 from microsoft/dev/miallama/pna-layer v0.9.2 Miltos 2021-03-12 07:55:42 +0000
  • 6a00b4a65c Simplify PNA aggregation/layers. Miltos Allamanis 2021-03-11 17:12:06 +0000
  • 4d45e10a4f Merge branch 'master' into dev/miallama/pna-layer Miltos Allamanis 2021-03-11 16:21:22 +0000
  • 35b45d5f8b Update AML files. Miltos Allamanis 2021-03-11 16:19:29 +0000
  • e627ca507d Update AML files. Miltos Allamanis 2021-03-11 16:19:29 +0000
  • 66cf9eae29
         Update setup.py v0.9.1 Miltos 2021-02-23 12:47:01 +0000
  • 7f59fb0a84 Merge branch 'master' into dev/miallama/pna-layer Miltos Allamanis 2021-02-22 10:15:47 +0000
  • 7a23ce4910
         Merge pull request #9 from microsoft/dev/miallama/edge-features v0.9.0 Miltos 2021-02-22 10:12:23 +0000
  • 8dca26f8cc Address Marc's feedback. Miltos Allamanis 2021-02-22 09:47:35 +0000
  • b42df7765d Bug fix on no edge features. Miltos Allamanis 2021-02-15 12:57:02 +0000
  • e5e1aff090 Fixes to the edge-attrs. Miltos Allamanis 2021-02-15 12:41:49 +0000
  • d5702cda39 Allow adding features to edges. Miltos Allamanis 2021-02-15 12:16:01 +0000
  • bc5323d2fd Minor AML updates. Miltos Allamanis 2021-01-25 08:32:49 +0000
  • b4525bec3d Add PNA layer. Miltos Allamanis 2021-01-12 14:59:34 +0000
  • 16c7ce6c7d
         Update pre-commit.yml v0.8.7 Miltos 2020-12-07 18:12:37 +0000
  • b2bbd7d2b5 Update how torch-scatter is loaded, given upstream changes. Miltos Allamanis 2020-12-07 17:47:20 +0000
  • f0340bc47a Always save after metadata loading. Miltos Allamanis 2020-12-07 17:41:30 +0000
  • 2ef1159b4b
         Merge pull request #7 from microsoft/mallamanis/simplify-saving v0.8.6 Miltos 2020-11-27 14:15:17 +0000
  • cceac45eb3 Add import... mallamanis/simplify-saving Miltos Allamanis 2020-11-27 12:35:20 +0000
  • ab3e3091d7 Simplify saving, when parent folder doesn't exist. Miltos Allamanis 2020-11-27 12:29:50 +0000
  • 0ba830866e
         Merge pull request #6 from microsoft/dev/mallamanis/amp v0.8.5 Miltos 2020-11-10 14:53:17 +0000
  • 3bfe1f8f7a Update test matrix to allow support for AMP dev/mallamanis/amp Miltos Allamanis 2020-11-10 14:13:39 +0000
  • f12abb4c01 Support AMP training in GNNs and other neural models. Miltos Allamanis 2020-11-10 14:04:56 +0000
  • e1c507f399
         Merge pull request #5 from microsoft/dev/mallamanis/expose-shuffling-option v0.8.4 Miltos 2020-09-29 07:45:11 +0100
  • 7d19a30ad4 Expose option to shuffle training tensor iterator. Miltos Allamanis 2020-09-28 23:33:50 +0100
  • 128f01c040 Migrate to a fixed version of black/isort to avoid discrepancies in CI. Miltos Allamanis 2020-09-28 09:43:12 +0100
  • 2cb7ed63fd
         Merge pull request #4 from microsoft/dev/mallamanis/expose-ifaces v0.8.3 Miltos 2020-09-28 09:24:25 +0100
  • 5eeedfb00d Expose `load_metadata_and_create_network` dev/mallamanis/expose-ifaces Miltos Allamanis 2020-09-28 09:15:37 +0100
  • 1f1001f3bf Try to fix pre-commit + black + GitHub Action again Miltos Allamanis 2020-09-10 09:57:49 +0100
  • 34cb860e0f try to fix pre-commit action and update Docker def Miltos Allamanis 2020-09-10 09:34:34 +0100
  • 9b0e876630
         Merge pull request #3 from microsoft/mallamanis/graphnorm Miltos 2020-09-10 09:23:32 +0100
  • 688391a2cc A GraphNorm implementation. mallamanis/graphnorm Miltos Allamanis 2020-09-10 08:30:12 +0100
  • 69da86ac88 Bug fix: actually invoke metadata-finalized hooks. Miltos Allamanis 2020-09-01 16:26:59 +0100
  • 20668b8bfb
         Merge pull request #2 from DrShushen/DrShushen/fix-train-restore-checkpoint Marc Brockschmidt 2020-08-20 08:41:27 +0100
  • c04398bea0 Add missing device arg. in __restore_checkpoint() Evgeny Saveliev 2020-08-19 18:15:54 +0000
  • 05217d2fc4 Create an accessor to the message-passing layers. Miltos Allamanis 2020-07-16 13:38:29 +0100
  • cbba5ad4ae Fix self-att pseudo-layer Miltos Allamanis 2020-07-09 08:14:03 +0100
  • ed1d4de286
         Update tutorial.md Miltos 2020-07-01 10:08:35 +0100
  • ebe3b35f40 Add missing implementations. v0.8.2 Miltos Allamanis 2020-06-24 11:12:57 +0100
  • d69379a59e Improve clarity of code in `MultiheadSelfAttentionVarSizedElementReduce`. v0.8.1 v0.8.0 Miltos Allamanis 2020-06-11 18:45:44 +0100
  • 696c2aa9b7 Ensure optional dependencies are loaded only when necessary. Miltos Allamanis 2020-06-06 08:26:05 +0100
  • e43234bebd Address some LGTM.com alerts and other minor edits. Miltos Allamanis 2020-05-17 10:53:53 +0100
  • 995b818f7b Add release instructions in README. Miltos Allamanis 2020-05-13 11:13:40 +0100
  • 21da7192d0 Update README to include pip installation instructions. Miltos Allamanis 2020-05-13 08:28:55 +0100
  • dc750bdef2
         Update publish.yml v0.7.8 Miltos 2020-05-12 22:38:53 +0100
  • 6864f34b7d
         Update publish.yml Miltos 2020-05-12 22:16:48 +0100
  • aed943f475
         Update publish.yml Miltos 2020-05-12 22:12:27 +0100
  • 45ec96264d Still trying to get GH action. v0.7.4 Miltos Allamanis 2020-05-12 21:32:08 +0100
  • c4f54ef1f6 Try fix PyPi pipeline, again. v0.7.3 Miltos Allamanis 2020-05-12 20:57:18 +0100
  • e818d5474a Keep trying to fix the publishig pipeline. v0.7.2 Miltos Allamanis 2020-05-12 19:55:49 +0100
  • 9dafc8ec93 Try fix publish workflow. v0.7.1 Miltos Allamanis 2020-05-12 19:21:32 +0100
  • 2ae47e7117
         Try publishing to PyPi v0.7 Miltos 2020-05-12 19:01:29 +0100
  • c19beee08b Update dpu-utils dependency. Miltos Allamanis 2020-05-12 13:57:18 +0100