Datasets:

---

renwei2024

/

nemo-github-issues

like 0

[ { "id": 1923495043, "node_id": "MDU6TGFiZWwxOTIzNDk1MDQz", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/core", "name": "core", "color": "f25e3a", "default": false, "description": "Changes to NeMo Core" } ]

[]

[ "jenkins", "Hi @titu1994,\r\n\r\nThank you for the review! \r\nI added the docstrings and comments.\r\n\r\nRegarding your question: If all checkpoints gets filtered out in Exp Manager, then depending on `resume_ignore_no_checkpoint` option - training starts from scratch (resume_ignore_no_checkpoint=True) or raises NotFoundError (resume_ignore_no_checkpoint=False).", "@jbieniusiewi thanks for the PR. PTL 2.1 creates a symlink for the last checkpoint so we wouldn't have an issue where saving last checkpoint is incomplete. Would this PR be still relevant with 2.1 ?\r\nCC @titu1994 @ericharper ", "Hi @athitten\r\nI think some parts of this PR might be still relevant even with symlinks to the latest checkpoint introduced in TPL 2.1 E.g., incomplete checkpoints cleanup. Certainly this PR will need to be reviewed for PTL 2.1. \r\n\r\n ", "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "This PR was closed because it has been inactive for 7 days since being marked as stale.", "jenkins", "jenkins", "@titu1994 is this good to be merged or do you have any comments to be resolved ?", "This will need to be rechecked for PTL 2.1 release when we update " ]

[ { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null }, { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "MCore code path ignores config of enabling or disabling TE, so this LGTM. Thanks.", "jenkins", "jenkins", "jenkins", "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "This PR was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null }, { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "This PR was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" } ]

[]

[ "\r\nI found that setting `num_sanity_val_steps: 0` and` limit_val_batches: 0.0` allows the training stage to run without the `Sanity Checking DataLoader\"`and `Validation DataLoader`. However, I still want to know if there is a better way to handle this, as I believe a time requirement of 66452 hours is not a normal value.", "Hi, its not a good idea to have same manifest for training and validation. The requirement is to have subset of speaker classes from training manifest. So prepare validation manifest accordingly. \r\n\r\nTraining time depends on multiple factors: dataset size, batchsize, GPU, number of GPUs, utterance length of each sample etc. ", "Hi, @nithinraok \r\n\r\nThank you for letting me know about the validation requirement. \r\n\r\nAlthough the training time depends on many factors, my training stage takes about 1 day, while the validation stage takes 66,452 hours. \r\n\r\nI believe there is a problem here, but I don't know how to deal with it. \r\n\r\nBesides my current practice of skipping the validation stage, do you have any other suggestions?", "My suggestion is to have 10% of training hrs of speakers in validation set. ", "ok, thanks" ]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" } ]

[]

[ "This happens when the job is killed while saving a checkpoint. Currently you need to manually cleanup the checkpoints folder to be able to resume. Figure out which is the most recent *complete* \"-last\" checkpoint, and delete the other(s).", "pls re-open if solution doesn;t work or if you face any other issue. ", "@szhengac To allow PTL to exit gracefully, if your job has some max runtime, you could tell PTL to exit earlier, so that the program has buffer time to clean up the old checkpoint. For example, if your job can only run for max 4 hours, you could set `++trainer.max_time=\"00:03:50:00\"`.", "Instead of trainer.max_time, you should use exp_manager.max_time_per_run \r\n\r\nhttps://github.com/NVIDIA/NeMo/blob/main/nemo/utils/exp_manager.py#L175" ]

[ { "id": 4418253970, "node_id": "LA_kwDOC_bI7s8AAAABB1k0kg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/TTS", "name": "TTS", "color": "49AF28", "default": false, "description": "" } ]

[]

[ "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "jenkins" ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" }, { "id": 4840216855, "node_id": "LA_kwDOC_bI7s8AAAABIH_ZFw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/CI", "name": "CI", "color": "1D76DB", "default": false, "description": "" } ]

[]

[ "jenkins", "jenkins", "jenkins", "jenkins", "jenkins", "jenkins" ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "jenkins", "jenkins" ]

[ { "id": 4840216855, "node_id": "LA_kwDOC_bI7s8AAAABIH_ZFw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/CI", "name": "CI", "color": "1D76DB", "default": false, "description": "" } ]

[]

[]

[ { "id": 4811884691, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ikw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/ASR", "name": "ASR", "color": "F3B33E", "default": false, "description": "" }, { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "jenkins", "tested and it looks good to me" ]

[ { "id": 1485815558, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU4", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/documentation", "name": "documentation", "color": "0075ca", "default": true, "description": "Improvements or additions to documentation" } ]

[ { "login": "Jorjeous", "id": 37293288, "node_id": "MDQ6VXNlcjM3MjkzMjg4", "avatar_url": "https://avatars.githubusercontent.com/u/37293288?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Jorjeous", "html_url": "https://github.com/Jorjeous", "followers_url": "https://api.github.com/users/Jorjeous/followers", "following_url": "https://api.github.com/users/Jorjeous/following{/other_user}", "gists_url": "https://api.github.com/users/Jorjeous/gists{/gist_id}", "starred_url": "https://api.github.com/users/Jorjeous/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Jorjeous/subscriptions", "organizations_url": "https://api.github.com/users/Jorjeous/orgs", "repos_url": "https://api.github.com/users/Jorjeous/repos", "events_url": "https://api.github.com/users/Jorjeous/events{/privacy}", "received_events_url": "https://api.github.com/users/Jorjeous/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "[celebrate] Georgy Zelenfroind reacted to your message:\r\n________________________________\r\nFrom: Somshubra Majumdar ***@***.***>\r\nSent: Tuesday, December 12, 2023 3:54:39 PM\r\nTo: NVIDIA/NeMo ***@***.***>\r\nCc: Georgy Zelenfroind ***@***.***>; Assign ***@***.***>\r\nSubject: Re: [NVIDIA/NeMo] Added a procedure for Windows users, README (PR #7942)\r\n\r\n\r\n@titu1994 approved this pull request.\r\n\r\nThis is very nice ! Thanks !\r\n\r\n—\r\nReply to this email directly, view it on GitHub<https://github.com/NVIDIA/NeMo/pull/7942#pullrequestreview-1777862998>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AI4QZ2DQQAQDV37562F3XX3YJB437AVCNFSM6AAAAAA74MRP2OVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMYTONZXHA3DEOJZHA>.\r\nYou are receiving this because you were assigned.Message ID: ***@***.***>\r\n", "[celebrate] Georgy Zelenfroind reacted to your message:\r\n________________________________\r\nFrom: Somshubra Majumdar ***@***.***>\r\nSent: Tuesday, December 12, 2023 3:54:52 PM\r\nTo: NVIDIA/NeMo ***@***.***>\r\nCc: Georgy Zelenfroind ***@***.***>; Assign ***@***.***>\r\nSubject: Re: [NVIDIA/NeMo] Added a procedure for Windows users, README (PR #7942)\r\n\r\n\r\nMerged #7942<https://github.com/NVIDIA/NeMo/pull/7942> into main.\r\n\r\n—\r\nReply to this email directly, view it on GitHub<https://github.com/NVIDIA/NeMo/pull/7942#event-11225619307>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AI4QZ2AJ4XXAHQWGFSYHF7DYJB44ZAVCNFSM6AAAAAA74MRP2OVHI2DSMVQWIX3LMV45UABCJFZXG5LFIV3GK3TUJZXXI2LGNFRWC5DJN5XDWMJRGIZDKNRRHEZTANY>.\r\nYou are receiving this because you were assigned.Message ID: ***@***.***>\r\n" ]

[ { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null } ]

[]

[ "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "This PR was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null } ]

[ { "login": "vsl9", "id": 4344862, "node_id": "MDQ6VXNlcjQzNDQ4NjI=", "avatar_url": "https://avatars.githubusercontent.com/u/4344862?v=4", "gravatar_id": "", "url": "https://api.github.com/users/vsl9", "html_url": "https://github.com/vsl9", "followers_url": "https://api.github.com/users/vsl9/followers", "following_url": "https://api.github.com/users/vsl9/following{/other_user}", "gists_url": "https://api.github.com/users/vsl9/gists{/gist_id}", "starred_url": "https://api.github.com/users/vsl9/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/vsl9/subscriptions", "organizations_url": "https://api.github.com/users/vsl9/orgs", "repos_url": "https://api.github.com/users/vsl9/repos", "events_url": "https://api.github.com/users/vsl9/events{/privacy}", "received_events_url": "https://api.github.com/users/vsl9/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "This PR was closed because it has been inactive for 7 days since being marked as stale." ]

[]

[ { "login": "ericharper", "id": 11999610, "node_id": "MDQ6VXNlcjExOTk5NjEw", "avatar_url": "https://avatars.githubusercontent.com/u/11999610?v=4", "gravatar_id": "", "url": "https://api.github.com/users/ericharper", "html_url": "https://github.com/ericharper", "followers_url": "https://api.github.com/users/ericharper/followers", "following_url": "https://api.github.com/users/ericharper/following{/other_user}", "gists_url": "https://api.github.com/users/ericharper/gists{/gist_id}", "starred_url": "https://api.github.com/users/ericharper/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/ericharper/subscriptions", "organizations_url": "https://api.github.com/users/ericharper/orgs", "repos_url": "https://api.github.com/users/ericharper/repos", "events_url": "https://api.github.com/users/ericharper/events{/privacy}", "received_events_url": "https://api.github.com/users/ericharper/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "Is `nemofw-training:23.08.03` subset of `nvcr.io/nvidia/nemo:23.08`? \r\n\r\nThe official guide [here](https://docs.nvidia.com/nemo-framework/user-guide/latest/playbooks/llama2sft.html) suggests to use `nemofw-training:23.08.03`, but I got the following error when I `docker pull nvcr.io/ea-bignlp/ga-participants/nemofw-training:23.08.03`:\r\n`Error response from daemon: pull access denied for nvcr.io/ea-bignlp/ga-participants/nemofw-training, repository does not exist or may require 'docker login': denied: requested access to the resource is denied`\r\n\r\nSame error even after I docker logged in. Do you guys have any idea what is the reason and which one is supposed for SFT?", "Hi\r\nAre you logging correctly?\r\nYou must register on the site below and use your created api-key to login.\r\nngc.nvidia.com", "@alireza0r Thank you for the reminder. Yes, it was related to the api-key. `docker login` command gave me login success info, but it didn't work. I have to use command `docker login nvcr.io -p api-key` to be able to successfully pull the image. ", "> @alireza0r Thank you for the reminder. Yes, it was related to the api-key. `docker login` command gave me login success info, but it didn't work. I have to use command `docker login nvcr.io -p api-key` to be able to successfully pull the image.\r\n\r\nYou actually need to have the proper access provided by NVIDIA. " ]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" }, { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null } ]

[]

[ "@KunalDhawan @nithinraok Have you seen anything like this?", "I had something similar, not sure if exactly the same though.\r\n\r\nWhat I did was: \r\n1. Not use the container: Install everything manually inside a virtualenv.\r\n2. Do not use `dali`. \r\n3. Check the callbacks you are using. \r\n", "The cluster I'm running on does not allow bare metal installations. However, I'm also running the most recent NeMo container that is available on NGC (23.08), more precisely the example recipe `examples/asr/asr_ctc/speech_to_text_ctc_bpe.py`. All of this under the assumption that apart from NCCL parameters due to local network specifics, it should work as is. \r\n \r\nIn a 2 x 2 setting with \r\n```\r\nexport TORCH_CPP_LOG_LEVEL=INFO\r\nexport TORCH_DISTRIBUTED_DEBUG=DETAIL\r\n```\r\n\r\nI get the following error:\r\n```py\r\n...\r\nRuntimeError: Detected mismatch between collectives on ranks. Rank 1 is running collective: CollectiveFingerPrint(SequenceNumber=76157OpType=BARRIER), but Rank 2 is running collective: CollectiveFingerPrint(SequenceNumber=0OpType=GATHER).Collectives differ in the following aspects: Sequence number: 76157vs 0 Op type: BARRIERvs GATHER\r\n...\r\nRuntimeError: Detected mismatch between collectives on ranks. Rank 1 is running collective: CollectiveFingerPrint(SequenceNumber=76157OpType=BARRIER), but Rank 2 is running collective: CollectiveFingerPrint(SequenceNumber=0OpType=GATHER).Collectives differ in the following aspects: Sequence number: 76157vs 0 Op type: BARRIERvs GATHER\r\n...\r\nRuntimeError: Detected mismatch between collectives on ranks. Rank 0 is running collective: CollectiveFingerPrint(SequenceNumber=76157OpType=BARRIER), but Rank 2 is running collective: CollectiveFingerPrint(SequenceNumber=0OpType=GATHER).Collectives differ in the following aspects: Sequence number: 76157vs 0 Op type: BARRIERvs GATHER\r\n...\r\nRuntimeError: Detected mismatch between collectives on ranks. Rank 3 is running collective: CollectiveFingerPrint(SequenceNumber=76157, OpType=ALLREDUCE, TensorShape=[1], TensorDtypes=Float, TensorDeviceTypes=TensorOptions(dtype=float (default), device=cuda, layout=Strided (default), requires_grad=false (default), pinned_memory=false (default), memory_format=(nullopt))), but Rank 0 is running collective: CollectiveFingerPrint(SequenceNumber=0OpType=REDUCE).Collectives differ in the following aspects: Sequence number: 76157vs 0 Op type: ALLREDUCEvs REDUCE\r\n```", "in your script, do you prepend your command with `srun`?\r\n\r\nAnd I suppose you also added `SLURMEnvironment` to the `trainer`:\r\n\r\n```py\r\nfrom pytorch_lightning.plugins.environments import SLURMEnvironment\r\n[...]\r\n trainer = pl.Trainer(\r\n **cfg.trainer,\r\n plugins=[SLURMEnvironment()],\r\n )\r\n```", "Yes, I do run with srun, and the SLURMEnvironment is properly set. \r\n\r\nI've discovered that the culprit is the addition of the parameter\r\n```\r\n+exp_manager.checkpoint_callback_params.save_best_model=true\r\n```\r\nas soon as I set `save_best_model` to `true` the above mentioned mismatch errors appear, I guess there must be something wrong with \r\nhttps://github.com/NVIDIA/NeMo/blob/937df1a4899d4a0ab9236ded491abd44a86e597c/nemo/utils/callbacks/nemo_model_checkpoint.py#L159-L184\r\n\r\nPerhaps related to https://github.com/Lightning-AI/pytorch-lightning/issues/14302#issuecomment-1220525529, which has been corrected in `NeMoModelCheckpoint.on_tain_end()`, but doesn't seem to have been addressed here. Indeed I've noticed log warnings of the like\r\n```\r\n[NeMo W 2023-12-22 23:04:20 nemo_logging:349] /usr/local/lib/python3.10/dist-packages/pytorch_lightning/trainer/connectors/logger_connector/result.py:433: PossibleUserWarning: It is recommended to use `self.log('global_step', ..., sync_dist=True)` when logging on epoch level in distributed setting to accumulate the metric across devices.\r\n\r\n[NeMo W 2023-12-22 23:04:20 nemo_logging:349] /usr/local/lib/python3.10/dist-packages/pytorch_lightning/trainer/connectors/logger_connector/result.py:433: PossibleUserWarning: It is recommended to use `self.log('val_loss', ..., sync_dist=True)` when logging on epoch level in distributed setting to accumulate the metric across devices.\r\n ```\r\n Could this be related?", "@itzsimpl is the above issue you described with `save_best_model` in exp_manager only in case of multi gpu runs ? Can it be reproduced with single GPU ?", "As said, single-node single-gpu, single-node multi-gpu works fine; the issue is present when running multi-node multi-gpu. I have the system busy at the moment, but for the sake of completeness I will re-check the case of multi-node single-gpu, asap, but if I recall correctly from previous experiments, it worked fine as well.\r\n\r\n@titu1994 any ideas?", "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "Turns out this issue pops out depending on the setup, the principal cause being the checkpoint reload that happens when `save_best_model` is used in combination with `always_save_nemo` (the error will/may happen on checkpoint):\r\n\r\nhttps://github.com/NVIDIA/NeMo/blob/937df1a4899d4a0ab9236ded491abd44a86e597c/nemo/utils/callbacks/nemo_model_checkpoint.py#L169\r\n\r\nand/or `save_last`(the error will/may happen on train end):\r\n\r\nhttps://github.com/NVIDIA/NeMo/blob/937df1a4899d4a0ab9236ded491abd44a86e597c/nemo/utils/callbacks/nemo_model_checkpoint.py#L216\r\n\r\nThe checkpoint reload requires the files to exist on all nodes, as all nodes will reload it, even if only RANK 0 will actually save the `.nemo` archive (at least in the case of ASR and non-distributed checkpoints). \r\n\r\nWhen using a setup based on NFS shared storage, it is thus important to have the proper mounting parameters set on all nodes. NFS by default will cache directory entries and file attributes, which may prevent a file that was created on one node from being visible on another until the cache gets invalidated. In my testing the addition of `lookupcache=positive`, which sets the directory entry cache to always revalidate negative entries (non-existing files) was enough, but I haven’t tested with distributed-checkpoints. Others (like in #7756) suggest setting `lookupcache=none`, `noac` and/or `actimeo=0`, which disables directory entry and file attributes caching entirely, albeit with some performance penalty. \r\n\r\n
If changing the mounting parameters is not possible, one workaround is also to have `save_best_model` set to `false` (this resolves the issue at least for non-distributed checkpoints like in ASR training), but with the drawback that one needs to create the `.nemo` archive in a post-processing step.", "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "This issue was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null }, { "id": 4811884691, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ikw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/ASR", "name": "ASR", "color": "F3B33E", "default": false, "description": "" } ]

[ { "login": "KunalDhawan", "id": 15515031, "node_id": "MDQ6VXNlcjE1NTE1MDMx", "avatar_url": "https://avatars.githubusercontent.com/u/15515031?v=4", "gravatar_id": "", "url": "https://api.github.com/users/KunalDhawan", "html_url": "https://github.com/KunalDhawan", "followers_url": "https://api.github.com/users/KunalDhawan/followers", "following_url": "https://api.github.com/users/KunalDhawan/following{/other_user}", "gists_url": "https://api.github.com/users/KunalDhawan/gists{/gist_id}", "starred_url": "https://api.github.com/users/KunalDhawan/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/KunalDhawan/subscriptions", "organizations_url": "https://api.github.com/users/KunalDhawan/orgs", "repos_url": "https://api.github.com/users/KunalDhawan/repos", "events_url": "https://api.github.com/users/KunalDhawan/events{/privacy}", "received_events_url": "https://api.github.com/users/KunalDhawan/received_events", "type": "User", "user_view_type": "public", "site_admin": false }, { "login": "okuchaiev", "id": 22035961, "node_id": "MDQ6VXNlcjIyMDM1OTYx", "avatar_url": "https://avatars.githubusercontent.com/u/22035961?v=4", "gravatar_id": "", "url": "https://api.github.com/users/okuchaiev", "html_url": "https://github.com/okuchaiev", "followers_url": "https://api.github.com/users/okuchaiev/followers", "following_url": "https://api.github.com/users/okuchaiev/following{/other_user}", "gists_url": "https://api.github.com/users/okuchaiev/gists{/gist_id}", "starred_url": "https://api.github.com/users/okuchaiev/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/okuchaiev/subscriptions", "organizations_url": "https://api.github.com/users/okuchaiev/orgs", "repos_url": "https://api.github.com/users/okuchaiev/repos", "events_url": "https://api.github.com/users/okuchaiev/events{/privacy}", "received_events_url": "https://api.github.com/users/okuchaiev/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "@okuchaiev A gentle reminder for the same", "@titu1994 can you kindly please answer this", "@KunalDhawan could you answer this?", "Hi @mukherjeesougata, we do provide support for early stopping based on [PTL's EarlyStopping callback](https://lightning.ai/docs/pytorch/stable/common/early_stopping.html). The config parameters **create_checkpoint_callback** and **checkpoint_callback_params** in [ExpManager](https://github.com/NVIDIA/NeMo/blob/main/nemo/utils/exp_manager.py) might be of interest to you - https://github.com/NVIDIA/NeMo/blob/main/nemo/utils/exp_manager.py#L156-L157", "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "This issue was closed because it has been inactive for 7 days since being marked as stale." ]

[]

[ { "login": "KunalDhawan", "id": 15515031, "node_id": "MDQ6VXNlcjE1NTE1MDMx", "avatar_url": "https://avatars.githubusercontent.com/u/15515031?v=4", "gravatar_id": "", "url": "https://api.github.com/users/KunalDhawan", "html_url": "https://github.com/KunalDhawan", "followers_url": "https://api.github.com/users/KunalDhawan/followers", "following_url": "https://api.github.com/users/KunalDhawan/following{/other_user}", "gists_url": "https://api.github.com/users/KunalDhawan/gists{/gist_id}", "starred_url": "https://api.github.com/users/KunalDhawan/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/KunalDhawan/subscriptions", "organizations_url": "https://api.github.com/users/KunalDhawan/orgs", "repos_url": "https://api.github.com/users/KunalDhawan/repos", "events_url": "https://api.github.com/users/KunalDhawan/events{/privacy}", "received_events_url": "https://api.github.com/users/KunalDhawan/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "Hi @abarcovschi, we don't support P&C stripping currently but would be adding this feature very soon! The easiest way would be to filter-out/ convert (capitalization) the tokens in post-processing. We would soon be adding text post-processors for ASR, initial implementation (not tested/merged) is available here - [TextPostProcessor](https://github.com/NVIDIA/NeMo/blob/asr_normalize/nemo/collections/asr/parts/postprocessing/asr_text_postprocessor.py) class, [language specific normalizations](https://github.com/NVIDIA/NeMo/tree/asr_normalize/nemo/collections/asr/parts/postprocessing/language_normalizations)." ]

[ { "id": 1923495043, "node_id": "MDU6TGFiZWwxOTIzNDk1MDQz", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/core", "name": "core", "color": "f25e3a", "default": false, "description": "Changes to NeMo Core" }, { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "Closing to follow up elsewhere - see the description above for details." ]

[ { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null }, { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "jenkins", "jenkins", "jenkins", "The FIM part looks good to me. Thank you!", "jenkins", "jenkins", "jenkins", "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "This PR was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" }, { "id": 4840216855, "node_id": "LA_kwDOC_bI7s8AAAABIH_ZFw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/CI", "name": "CI", "color": "1D76DB", "default": false, "description": "" } ]

[]

[ "jenkins", "jenkins", "This PR incremented the version of Megatron Core from `e122536b7645edcb7ebf099b5c92a443f7dbf8e7` to `973330e9c3681604703bf1eb6b5a265d1b9b9b38`, that is from 0.3.0 to 0.4.0, but did not change the [requirements file](https://github.com/NVIDIA/NeMo/blob/main/requirements/requirements_nlp.txt#L13) accordingly" ]

[]

[ { "login": "okuchaiev", "id": 22035961, "node_id": "MDQ6VXNlcjIyMDM1OTYx", "avatar_url": "https://avatars.githubusercontent.com/u/22035961?v=4", "gravatar_id": "", "url": "https://api.github.com/users/okuchaiev", "html_url": "https://github.com/okuchaiev", "followers_url": "https://api.github.com/users/okuchaiev/followers", "following_url": "https://api.github.com/users/okuchaiev/following{/other_user}", "gists_url": "https://api.github.com/users/okuchaiev/gists{/gist_id}", "starred_url": "https://api.github.com/users/okuchaiev/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/okuchaiev/subscriptions", "organizations_url": "https://api.github.com/users/okuchaiev/orgs", "repos_url": "https://api.github.com/users/okuchaiev/repos", "events_url": "https://api.github.com/users/okuchaiev/events{/privacy}", "received_events_url": "https://api.github.com/users/okuchaiev/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "@okuchaiev a gentle reminder for the same", "If you have the two resume flags set, simply calling the script multiple times with the same paths will be sufficient ", "Yes, I have tried that but it is restarting from the 0th epoch.", "Please check the folder called nemo_experiments and if it has a folder inside it which contains logs and a checkpoints dir, then add the following line to your training script \r\n\r\nexp_manager.explicit_log_dir=[prefix]/nemo_experiments/[name of the folder]\r\n\r\nIn the future you can add exp_manager.name=\"xyz\" in order for NeMo to automatically detect the folder. ", "Try\r\n```\r\n++exp_manager.resume_from_checkpoint=\"<path of your .nemo file>\"\r\n```", "> exp_manager.resume_from_checkpoint\r\nI have used the following .sh script :\r\n```\r\nif [[ ${tokenizer_type}=bpe ]]; then\r\n\r\nCUDA_VISIBLE_DEVICES=3,4 python /home/conformer/vakyansh-nemo-experimentation/src/finetuning/speech_to_text_ctc_bpe.py \\\r\n --config-path=${config_path} \\\r\n --config-name=${config_name} \\\r\n model.train_ds.manifest_filepath=${train_manifest_path} \\\r\n model.validation_ds.manifest_filepath=${valid_manifest_path} \\\r\n model.test_ds.manifest_filepath=${test_manifest_path} \\\r\n model.tokenizer.dir=${tokenizer_dir} \\\r\n model.tokenizer.type=${tokenizer_type} \\\r\n trainer.devices=-1 \\\r\n trainer.accelerator=\"gpu\" \\\r\n trainer.strategy=\"ddp\" \\\r\n exp_manager.resume_if_exists=true \\\r\n exp_manager.resume_ignore_no_checkpoint=true \\\r\n exp_manager.create_wandb_logger=True \\\r\n exp_manager.wandb_logger_kwargs.name=${wandb_run_name} \\\r\n exp_manager.wandb_logger_kwargs.project=${wandb_project_name} \\\r\n +exp_manager.explicit_log_dir=${ckpt_dir} \\\r\n +init_from_pretrained_model=${pretrained_model}\r\n\r\nfi\r\n```\r\nBut, it is showing the following error:-\r\n```\r\nCould not override 'exp_manager.explicit_log_dir'.\r\nTo append to your config use +exp_manager.explicit_log_dir=/home/conformer/nemo_experiments/Conformer-CTC-BPE/2023-11-22_17-23-06/\r\nKey 'explicit_log_dir' is not in struct\r\n full_key: exp_manager.explicit_log_dir\r\n object_type=dict\r\n```", "You can add the key by using the provided instruction - \r\n \r\nTo append to your config use ++exp_manager.explicit_log_dir=/home/conformer/nemo_experiments/Conformer-CTC-BPE/2023-11-22_17-23-06/\r\n", "This worked for me\r\n\r\n> You can add the key by using the provided instruction -\r\n> \r\n> To append to your config use ++exp_manager.explicit_log_dir=/home/conformer/nemo_experiments/Conformer-CTC-BPE/2023-11-22_17-23-06/\r\n\r\n" ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "jenkins", "This PR aims to reverse changes from this PR: https://github.com/NVIDIA/NeMo/pull/7003\r\n\"Added a YAML parsing for UB/TP overlap configurations.\" doesn't describe the problem that you want to solve. Please be more verbose.\r\nThe https://github.com/NVIDIA/NeMo/pull/7003 removed module loading, opening file and parsing arguments from the model class. All of this is what hydra is for." ]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" }, { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null } ]

[ { "login": "ericharper", "id": 11999610, "node_id": "MDQ6VXNlcjExOTk5NjEw", "avatar_url": "https://avatars.githubusercontent.com/u/11999610?v=4", "gravatar_id": "", "url": "https://api.github.com/users/ericharper", "html_url": "https://github.com/ericharper", "followers_url": "https://api.github.com/users/ericharper/followers", "following_url": "https://api.github.com/users/ericharper/following{/other_user}", "gists_url": "https://api.github.com/users/ericharper/gists{/gist_id}", "starred_url": "https://api.github.com/users/ericharper/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/ericharper/subscriptions", "organizations_url": "https://api.github.com/users/ericharper/orgs", "repos_url": "https://api.github.com/users/ericharper/repos", "events_url": "https://api.github.com/users/ericharper/events{/privacy}", "received_events_url": "https://api.github.com/users/ericharper/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "Fix for this is #7943 \r\n\r\n", "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "This issue was closed because it has been inactive for 7 days since being marked as stale.", "```\r\nTraceback (most recent call last):\r\n File \"/opt/NeMo/examples/nlp/language_modeling/megatron_gpt_pretraining.py\", line 46, in <module>\r\n main()\r\n File \"/opt/NeMo/nemo/core/config/hydra_runner.py\", line 129, in wrapper\r\n _run_hydra(\r\n File \"/usr/local/lib/python3.10/dist-packages/hydra/_internal/utils.py\", line 389, in _run_hydra\r\n _run_app(\r\n File \"/usr/local/lib/python3.10/dist-packages/hydra/_internal/utils.py\", line 452, in _run_app\r\n run_and_report(\r\n File \"/usr/local/lib/python3.10/dist-packages/hydra/_internal/utils.py\", line 216, in run_and_report\r\n raise ex\r\n File \"/usr/local/lib/python3.10/dist-packages/hydra/_internal/utils.py\", line 213, in run_and_report\r\n return func()\r\n File \"/usr/local/lib/python3.10/dist-packages/hydra/_internal/utils.py\", line 453, in <lambda>\r\n lambda: hydra.run(\r\n File \"/usr/local/lib/python3.10/dist-packages/hydra/_internal/hydra.py\", line 132, in run\r\n _ = ret.return_value\r\n File \"/usr/local/lib/python3.10/dist-packages/hydra/core/utils.py\", line 260, in return_value\r\n raise self._return_value\r\n File \"/usr/local/lib/python3.10/dist-packages/hydra/core/utils.py\", line 186, in run_job\r\n ret.return_value = task_function(task_cfg)\r\n File \"/opt/NeMo/examples/nlp/language_modeling/megatron_gpt_pretraining.py\", line 40, in main\r\n model = MegatronGPTModel(cfg.model, trainer)\r\n File \"/opt/NeMo/nemo/collections/nlp/models/language_modeling/megatron_gpt_model.py\", line 273, in __init__\r\n super().__init__(cfg, trainer=trainer, no_lm_init=True)\r\n File \"/opt/NeMo/nemo/collections/nlp/models/language_modeling/megatron_base_model.py\", line 221, in __init__\r\n self._build_tokenizer()\r\n File \"/opt/NeMo/nemo/collections/nlp/models/language_modeling/megatron_base_model.py\", line 421, in _build_tokenizer\r\n self.tokenizer = get_nmt_tokenizer(\r\n File \"/opt/NeMo/nemo/collections/nlp/modules/common/tokenizer_utils.py\", line 175, in get_nmt_tokenizer\r\n raise ValueError(\"No Tokenizer path provided or file does not exist!\")\r\nValueError: No Tokenizer path provided or file does not exist!\r\n```\r\n\r\nGetting this error when trying to run nemotron. Has this issue been resolved? How can I ensure the tokenizer is loaded?" ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "LGTM. ", "jenkins" ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "jenkins", "jenkins", "jenkins" ]

[ { "id": 4811884691, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ikw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/ASR", "name": "ASR", "color": "F3B33E", "default": false, "description": "" } ]

[]

[ "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "jenkins", "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "Waiting for torch.jit support #8027 ", "jenkins", "jenkins", "jenkins", "jenkins", "jenkins", "jenkins", "jenkins", "jenkins", "Pasting here the discussion from Slack about Joint refactoring, `joint_after_projection`, non-inplace `unsqueeze`.\r\n\r\n## Main points:\r\n\r\n- (1) **separating projections in Joint** from other operations is not only needed for the \"memory vs speed\" tradeoff. It is also helpful for speed optimization without additional memory usage (this is done not only for (2))\r\n- (2) **The immediate projection of encoder output** is a tiny overhead, and I'm sure it is negligible compared to other operations for the RNN-T system\r\n- (3) **inplace `unsqueeze_` does not save memory**, memory consumption is the same with unsqueeze, and there is no overhead for the change\r\n- (4) i**mplementation of separation of projections** – I tried to preserve compatibility, readability, and usability for inheritance.\r\n\r\n\r\n### 1) separating projections in Joint from other operations is helpful in many cases.\r\nEven in the original encoder algorithm, when we loop over encoder frames, we can project the frame immediately (one-by-one => no memory overhead), but this will save computations: for each encoder frame, multiple evaluations for Joint are used => we waste time when recalculating the encoder vector's projection.\r\nThe new algorithm is even more sensitive to operations in Joint, and I see a substantial speedup for separating projections\r\n\r\n### 2) The immediate projection of encoder output is a tiny overhead\r\n\r\nI see the speedup from projecting the encoder output immediately.\r\nSo, what's the overhead, and is it significant?\r\n* This could be considered a significant overhead when we used tiny encoders with linear memory/time complexity. For modern encoders with quadratic complexity (due to attention)\r\n* for bs 128, 30 sec, subsampling 8, joint_hidden=640, fp32, the size of tensor will be ~118MB, for bf16 – ~59MB\r\n * To compare with the memory consumption of one piece of the encoder, I tried a MultiHeadAttention block (used by Conformer). It uses ~2129MB memory (one block!) due to quadratic complexity. I'm sure that 118MB or even 59MB is a tiny piece of memory compared to modern encoders (it's from Conformer large, not x+large!)\r\n* From a practical point of view, I can easily fit bs 256, fp32 to my desktop GPU (LibriSpeech test-other, Fast Conformer Large), and we are targeting bf16, bs 128, and smaller.\r\n* Comparison with CTC system: We project to the final output with vocabulary size dimension (not one-by-one!), which is larger than RNN-T projection, and do not optimize this for better memory usage, sacrificing the speed\r\n\r\nGiven all these facts, it is acceptable to project the encoder output immediately. If we need a robust memory consumption optimization, we can use a separate flag (`preserve_memory`), but I don't think it is now required.\r\n\r\n### 3) in-place `unsqueeze_` does not save memory (no overhead after separating projections)\r\n\r\nDue to separating projections, I needed to replace the in-place `unsqueeze_` operation with `unsqueeze`. There is no overhead in memory.\r\nAccording to the documentation https://pytorch.org/docs/stable/generated/torch.unsqueeze.html\r\n> The returned tensor shares the same underlying data with this tensor.\r\n\r\n### 4) Implementation of separation of projections.\r\nI think the acceptable solution should:\r\n* expose projections as the public API (`AbstractRNNTJoint`)\r\n* should be developer-friendly\r\n* should not lead to unnecessary code duplication\r\n* should not break checkpoints\r\n* should not introduce any significant overhead\r\n\r\nI considered several possibilities.\r\na) we can duplicate the code for `joint` in `joint_after_projection`, but there I do not think it is a good practice to maintain the same code in 2 places (it must be the same except applying projections)\r\n\r\nb) using `enc()` and `pred()` as functions: undesirable, since it will break the checkpoints.\r\n\r\nc) use `enc` and `pred` with type annotations in public API\r\n```python\r\nclass AbstractRNNTJoint(NeuralModule, ABC):\r\n enc: Callable # can be Identity\r\n pred: Callable # can be Identity\r\n \r\n @abstractmethod\r\n def joint_after_projection(self, f, g):\r\n \"\"\"This is the main method that one should implement for Joint\"\"\" \r\n raise NotImplementedError()\r\n \r\n def joint(self, f, g):\r\n # not abstract anymore!\r\n return self.joint_after_projection(self.enc(f), self.pred(g))\r\n```\r\n\r\nd) separate abstract `project_prednet` and `project_encoder` methods – the current solution\r\nI think it is better since if `project_prednet` and `project_encoder` are not implemented, a clear error will indicate this.\r\n\r\nI prefer the last one because there is no overhead for the current implementation (see (3))." ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" }, { "id": 4840216855, "node_id": "LA_kwDOC_bI7s8AAAABIH_ZFw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/CI", "name": "CI", "color": "1D76DB", "default": false, "description": "" } ]

[]

[ "jenkins", "jenkins" ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "@athitten this change is similar to yours, could you review?", "@titu1994, can you review or assign somebody?", "@athitten can you review this?", "jenkins", "@ericharper, why my jenkins hangs forever? I don't have rights to see on it", "jenkins", "@titu1994, jenkins passed\r\n", "@athitten, @ericharper what about this pr? All checks passed" ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "jenkins", "@athitten Please review and merge" ]

[]

[]

[ "Do we have a test for `convert_nemo_llama_to_hf.py` anywhere?\r\n\r\nLike we have, for example, for the script `convert_hf_llama_to_nemo.py` working the other way around [here](https://github.com/NVIDIA/NeMo/blob/521cfb4c9359c4033f34776da4b73fabf3e94240/Jenkinsfile#L131-L138).", "@janekl I'm not sure if there's such a test since I'm not the original author of `convert_nemo_llama_to_hf.py`. The changes here are made based on trying to load the files into hf (which didn't work previously if you use the pytorch_model.bin directly, since all model weights are contained within ['state_dict'] which the corresponding pytorch_model.bin from HF do not, and instead expose the keys directly). The only other change here is ensuring `model_config.tensor_model_parallel_size = 1`, which makes the conversion fail otherwise when running with `--cpu-only ` Maybe Utkarsh Uppal, the author of https://github.com/NVIDIA/NeMo/pull/7770 might know.", "jenkins" ]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" }, { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null } ]

[]

[ "A quick first win is to use `.size()` instead of `.shape[]` here https://github.com/NVIDIA/NeMo/blob/08937c8e7dd2e782fac99c6c230ae0170c277700/nemo/collections/asr/parts/preprocessing/features.py#L150 :\r\n\r\n```py\r\n mask = (\r\n torch.arange(like.size(time_dim), device=like.device)\r\n .repeat(lengths.size(0), 1)\r\n .lt(lengths.view(-1, 1))\r\n )\r\n```\r\n\r\nBut then the loops can't pass:\r\n```\r\nTraceback (most recent call last):\r\n File \"/Users/msis/Projects/tarteel/nemo2ios/poc/nemo_tracing_issue.py\", line 30, in <module>\r\n model = ct.convert(\r\n ^^^^^^^^^^^\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/_converters_entry.py\", line 574, in convert\r\n mlmodel = mil_convert(\r\n ^^^^^^^^^^^^\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/mil/converter.py\", line 188, in mil_convert\r\n return _mil_convert(model, convert_from, convert_to, ConverterRegistry, MLModel, compute_units, **kwargs)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/mil/converter.py\", line 212, in _mil_convert\r\n proto, mil_program = mil_convert_to_proto(\r\n ^^^^^^^^^^^^^^^^^^^^^\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/mil/converter.py\", line 286, in mil_convert_to_proto\r\n prog = frontend_converter(model, **kwargs)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/mil/converter.py\", line 108, in __call__\r\n return load(*args, **kwargs)\r\n ^^^^^^^^^^^^^^^^^^^^^\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/mil/frontend/torch/load.py\", line 80, in load\r\n return _perform_torch_convert(converter, debug)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/mil/frontend/torch/load.py\", line 99, in _perform_torch_convert\r\n prog = converter.convert()\r\n ^^^^^^^^^^^^^^^^^^^\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/mil/frontend/torch/converter.py\", line 519, in convert\r\n convert_nodes(self.context, self.graph)\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/mil/frontend/torch/ops.py\", line 88, in convert_nodes\r\n add_op(context, node)\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/mil/frontend/torch/ops.py\", line 3272, in loop\r\n loop = mb.while_loop(\r\n ^^^^^^^^^^^^^^\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/mil/mil/ops/registry.py\", line 182, in add_op\r\n return cls._add_op(op_cls_to_add, **kwargs)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/mil/mil/builder.py\", line 183, in _add_op\r\n new_op.build_nested_blocks()\r\n File \"/Users/msis/.virtualenvs/nemo2ios/lib/python3.11/site-packages/coremltools/converters/mil/mil/ops/defs/iOS15/control_flow.py\", line 452, in build_nested_blocks\r\n raise ValueError(msg.format(\r\nValueError: loop_vars 'mask.1_x0' changes in the body of while_loop 'mask.45':\r\n <class 'coremltools.converters.mil.mil.types.type_tensor.tensor.<locals>.tensor'> -> <class 'coremltools.converters.mil.mil.types.type_tensor.tensor.<locals>.tensor'>\r\n```", "[7921.zip](https://github.com/NVIDIA/NeMo/files/13419605/7921.zip)\r\nA zip of the tensor used in the example.", "There a few more hurdles:\r\n1. Need to manually unroll loop\r\n2. `coreml` does not support the logical not\r\n3. `coreml` does not support bool type and will cast to fp32\r\n\r\nGiven that it seems like these are `coreml` short-comings, I'm not sure the code should be changed (I wrote this function to be `torch.jit`-friendly and reasonably generalized).\r\n\r\nWith some assumptions, these ad hoc patches export with `coreml` (though I didn't check for correctness, since I am unfamiliar with `coreml`):\r\n\r\n```python\r\[email protected]_if_tracing\r\ndef make_seq_mask_like(\r\n lengths: torch.Tensor, like: torch.Tensor, time_dim: int = -1, valid_ones: bool = True\r\n) -> torch.Tensor:\r\n # use `ge` instead of `lt` if `valid_ones=False`\r\n mask = torch.arange(like.size(time_dim), device=like.device).repeat(lengths.size(0), 1).ge(lengths.view(-1, 1))\r\n\r\n # unroll loop [B, T] -> [B, 1, T] \r\n mask = mask.unsqueeze(1)\r\n # for _ in range(like.dim() - mask.dim()):\r\n # mask = mask.unsqueeze(1)\r\n\r\n # RuntimeError: PyTorch convert function for op '__not__' not implemented.\r\n # we addressed this with `ge` vs `lt`\r\n # if not valid_ones:\r\n # mask = ~mask\r\n\r\n # coreml doesn't support bool output\r\n mask = mask.int()\r\n return mask\r\n```", "My goal is to export a NeMo model to Core ML. \r\nThat function was the last issue I had.\r\nWith the new modification, I can now export to Core ML without casting to bool:\r\n```py\r\[email protected]_if_tracing\r\ndef make_seq_mask_like(\r\n lengths: torch.Tensor, like: torch.Tensor, time_dim: int = -1, valid_ones: bool = True\r\n) -> torch.Tensor:\r\n # use `ge` instead of `lt` if `valid_ones=False`\r\n mask = torch.arange(like.size(time_dim), device=like.device).repeat(lengths.size(0), 1).ge(lengths.view(-1, 1))\r\n\r\n # unroll loop [B, T] -> [B, 1, T] \r\n mask = mask.unsqueeze(1)\r\n # for _ in range(like.dim() - mask.dim()):\r\n # mask = mask.unsqueeze(1)\r\n\r\n # RuntimeError: PyTorch convert function for op '__not__' not implemented.\r\n # we addressed this with `ge` vs `lt`\r\n # if not valid_ones:\r\n # mask = ~mask\r\n\r\n # # coreml doesn't support bool output\r\n # mask = mask.int()\r\n return mask\r\n```\r\n\r\nThis still works well with `torch.jit` and I think is just as generalized if no casting is done.\r\nSo can it be updated?", "It's really not my call, but I think the function has lost generality and should be used only for this purpose. E.g., the unrolled loop assumes `like`'s shape is `[B, D, T]`, `valid_ones` is ignored, and the output dtype is unintuitive. \r\n\r\nIf you grep this codebase for something like `egrep '^\\s*def .*mask(_|\\()'` you'll see about 10-20 different functions that create a similar mask in an ad hoc way... my intent was a reusable one.\r\n\r\nGiven that these are basic and common operations, I think you'd be doing the coreml converter developers a favor by opening an issue with this function as an MWE.", "Core ML only supports tracing and not scripting... See the issue I linked above. \r\n\r\nWhat other shapes does `like` take? \r\n\r\nWe can add conditions for `valid_ones`, no?\r\n", "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "?", "E.g., `like` takes a 4-dimensional shape `(B, M, F, N)` here \r\nhttps://github.com/NVIDIA/NeMo/blob/4fd1c74299d5c2c06c8b04e7a64c83fd9ab5a5da/nemo/collections/asr/modules/audio_modules.py#L461-L463\r\nI don't even know what the dimensions `(B, M, F, N)` are, but it works, which is the benefit of keeping it so general.\r\n\r\nConditions for `valid_ones` already exist in the original implementation... it's the coreml converter that cannot capture the behavior. This behavior is used, e.g., here to invert the default value: https://github.com/NVIDIA/NeMo/blob/4fd1c74299d5c2c06c8b04e7a64c83fd9ab5a5da/nemo/collections/asr/losses/audio_losses.py#L56\r\n\r\nI've verified that the original code works when exported to ONNX:\r\n\r\n<details><summary>ONNX test</summary>\r\n\r\n```python\r\ntorch.onnx.export(\r\n model=model,\r\n args=(lengths, like),\r\n f=\"tmp.onnx\",\r\n input_names=[\"lengths\", \"like\"],\r\n opset_version=17,\r\n dynamic_axes={\"lengths\": [0], \"like\": [0, 1, 2]},\r\n)\r\nsession = onnxruntime.InferenceSession(\"tmp.onnx\")\r\noutputs = session.run(\r\n None,\r\n {\"lengths\": np.array([5, 3, 1]).astype(np.float32), \"like\": np.random.random(size=[3, 2, 5]).astype(np.float32)},\r\n)\r\nprint(outputs[0])\r\n# output:\r\n# [[[ True True True True True]]\r\n# [[ True True True False False]]\r\n# [[ True False False False False]]]\r\n```\r\n\r\n</details> \r\n\r\nSo it works for JIT and ONNX, just not the coreml converter (which I think is immature).\r\n\r\nI still think it should be locally patched for the coreml converter for now, until they support more operators. @titu1994 would make the final call, I'm just a random guy who contributes for fun, not a maintainer.", "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "This issue was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "jenkins", "jenkins", "jenkins", "jenkins" ]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" }, { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null } ]

[ { "login": "ericharper", "id": 11999610, "node_id": "MDQ6VXNlcjExOTk5NjEw", "avatar_url": "https://avatars.githubusercontent.com/u/11999610?v=4", "gravatar_id": "", "url": "https://api.github.com/users/ericharper", "html_url": "https://github.com/ericharper", "followers_url": "https://api.github.com/users/ericharper/followers", "following_url": "https://api.github.com/users/ericharper/following{/other_user}", "gists_url": "https://api.github.com/users/ericharper/gists{/gist_id}", "starred_url": "https://api.github.com/users/ericharper/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/ericharper/subscriptions", "organizations_url": "https://api.github.com/users/ericharper/orgs", "repos_url": "https://api.github.com/users/ericharper/repos", "events_url": "https://api.github.com/users/ericharper/events{/privacy}", "received_events_url": "https://api.github.com/users/ericharper/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "This issue was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "the PR changes look fine but we might want to note that this only changes things for NeMo local GPT models, not m-core based ones which are the ones we majorly use now", "I think the main issue is that user are used to configuring this from the config. Now TE expects an environment variable to be set. Mcore makes sure that the env var and the config are consistent. Can NeMo set the env var automatically so that the user doesn't have to set the env var?", "yes, thats right. It would probably be better if we automatically set the TE env variable if someone passes it in the config", "Ideally, I would prefer if this env var setting happens in m-core since m-core also relies on arg passing format and is the place where we actually interface with TE and setup TE/local layers. But if we feel that it might need broader discussion it might make sense to add this in nemo for now so that it unblocks the issue?", "@ericharper @aklife97 can you approve this for now in nemo so it unblocks? ", "jenkins", "jenkins" ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" } ]

[]

[ "Hi @MachineGunLin \r\n\r\nCan you please let us know how you resolved the issue?" ]

[]

[ { "login": "okuchaiev", "id": 22035961, "node_id": "MDQ6VXNlcjIyMDM1OTYx", "avatar_url": "https://avatars.githubusercontent.com/u/22035961?v=4", "gravatar_id": "", "url": "https://api.github.com/users/okuchaiev", "html_url": "https://github.com/okuchaiev", "followers_url": "https://api.github.com/users/okuchaiev/followers", "following_url": "https://api.github.com/users/okuchaiev/following{/other_user}", "gists_url": "https://api.github.com/users/okuchaiev/gists{/gist_id}", "starred_url": "https://api.github.com/users/okuchaiev/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/okuchaiev/subscriptions", "organizations_url": "https://api.github.com/users/okuchaiev/orgs", "repos_url": "https://api.github.com/users/okuchaiev/repos", "events_url": "https://api.github.com/users/okuchaiev/events{/privacy}", "received_events_url": "https://api.github.com/users/okuchaiev/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "@SefaZeng there is a issue with links, its not open sourced through NeMo. \r\nCurrently you can request for access to NeMo framework at https://developer.nvidia.com/nemo-framework and run scripts", "@SefaZeng closing this issue, feel free to reopen if you are unable to run the scripts after downloading the nemo-framework container.", "> @SefaZeng there is a issue with links, its not open sourced through NeMo. Currently you can request for access to NeMo framework at https://developer.nvidia.com/nemo-framework and run scripts\r\n\r\nThank you for your reply. Could you please tell what is the difference between NeMo and NeMo-Framework?", "(As of current)\r\n[NeMo](https://nvidia.github.io/NeMo/) is open source conversational toolkit for training ASR, TTS, NLP and Speaker Diarization models. \r\n[NeMo framework](https://developer.nvidia.com/nemo-framework) is a open access container based framework for building, customizing and deploying LLM/multi-modal models. " ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "jenkins", "jenkins" ]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" }, { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null } ]

[ { "login": "tango4j", "id": 13211483, "node_id": "MDQ6VXNlcjEzMjExNDgz", "avatar_url": "https://avatars.githubusercontent.com/u/13211483?v=4", "gravatar_id": "", "url": "https://api.github.com/users/tango4j", "html_url": "https://github.com/tango4j", "followers_url": "https://api.github.com/users/tango4j/followers", "following_url": "https://api.github.com/users/tango4j/following{/other_user}", "gists_url": "https://api.github.com/users/tango4j/gists{/gist_id}", "starred_url": "https://api.github.com/users/tango4j/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/tango4j/subscriptions", "organizations_url": "https://api.github.com/users/tango4j/orgs", "repos_url": "https://api.github.com/users/tango4j/repos", "events_url": "https://api.github.com/users/tango4j/events{/privacy}", "received_events_url": "https://api.github.com/users/tango4j/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "Thanks for raising this issue, could you please attach full log here. ", "Attached the log from running NeuralDiarizer. \r\n\r\n[nemo.log](https://github.com/NVIDIA/NeMo/files/13428499/nemo.log)\r\n", "I'm also having this issue, hangs at this step, similar settings as op. Versions: Nemo 1.21.0, python 3.10\r\n\r\n```\r\n[NeMo I 2023-12-05 18:46:27 collections:302] Dataset loaded with 43 items, total duration of 0.60 hours.\r\n[NeMo I 2023-12-05 18:46:27 collections:304] # 43 files loaded accounting to # 1 labels\r\nvad: 100%|██████████| 43/43 [00:07<00:00, 6.06it/s]\r\n[NeMo I 2023-12-05 18:46:34 clustering_diarizer:250] Generating predictions with overlapping input segments\r\n```", "@remenberl \r\nThank you for uploading the samples. I will test it and get back to you.\r\nLooking into the log you shared, I suppose 64GB RAM is not enough to handle 4 hours of diarization in an offline manner.\r\nI will confirm the RAM requirement for this sample after run this myself.", "This worked for me after building from source, and setting num_workers to 0 ", "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "This issue was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" }, { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null } ]

[]

[ "I think `test` should not be in `all`.\r\nFor now, a workaround is:\r\n\r\n```\r\npip install \r\n nemo_toolkit[core]==1.21.0 \\\r\n nemo_toolkit[common]==1.21.0 \\\r\n nemo_toolkit[asr]==1.21.0 \\\r\n nemo_toolkit[nlp]==1.21.0 \\\r\n nemo_toolkit[tts]==1.21.0 \\\r\n nemo_toolkit[slu]==1.21.0 \\\r\n nemo2riva==2.31.1\r\n```", "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "Still not resolved", "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "This issue was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null }, { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "Closing in favor of more-compatible change on the NLPAdapterModelMixin side in #7971." ]

[ { "id": 1923495043, "node_id": "MDU6TGFiZWwxOTIzNDk1MDQz", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/core", "name": "core", "color": "f25e3a", "default": false, "description": "Changes to NeMo Core" }, { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" }, { "id": 4840216855, "node_id": "LA_kwDOC_bI7s8AAAABIH_ZFw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/CI", "name": "CI", "color": "1D76DB", "default": false, "description": "" } ]

[]

[ "I trained the llama model with 2 nodes using model.fp8=True, model.fp8_params=True, model.optim.name=distributed_fused_adam, model.optim.dtype=bf16.\r\nI got this error when saving checkpoint\r\n\r\n<img width=\"1421\" alt=\"Screenshot 2023-11-20 at 17 27 23\" src=\"https://github.com/NVIDIA/NeMo/assets/25007718/5ed193c7-44a4-4387-a2b5-f2cf40aa87af\">\r\n\r\n\r\nHow can i solve this problem\r\n\r\nThank you!", "@chiendb97 Thanks for the report! I reproduce the error and have a fix at https://github.com/NVIDIA/TransformerEngine/pull/529. I haven't tested thoroughly, but I was able to save and load a checkpoint for LLaMa with FP8 params.", "@timmoon10 Is there any blocker for the review? With this PR, the memory allocation per param is: 1(fp8)+1(fp8 transpose)+2(BF16 gradient)+[2(bf16 weight)+2(bf16 momentum)+2(bf16 variance)]/dp=4+6/dp. Is my understanding correct?", "@lhb8125 The only remaining blocker is fixing some convergence issues when running with SFT. I don't fully understand that issue yet, but I don't expect it would require major changes. Starting a review would be great.", "I've found a bug when using this PR with LLaMa SFT. Bugfix: https://github.com/NVIDIA/TransformerEngine/pull/567\r\n\r\nThis does not affect GPT pretraining though. I think this is ready to review and merge.", "jenkins", "The Jenkins failure is because it is using an old version of Apex. The Dockerfile and README have been updated with the required Apex.", "What Base PyTorch version is needed? We can update it in the Jenkinsfile", "jenkins", "jenkins", "jenkins", "This PR requires [Apex commit a2f6683](https://github.com/NVIDIA/apex/commit/a2f6683b10fb4c29ab57c9e3d16957db76a8a5ba) (11/19), which is included in the PyTorch 23.11 container. However, [Apex commit 6ff4548](https://github.com/NVIDIA/apex/commit/6ff45486f432f91eb86937a0def5eb5f2cf792ae) (9/27) breaks the API used by Megatron-LM and causes the Jenkins pipeline to fail.\r\n\r\nProbably the easiest fix is to tweak Apex so that [6ff4548](https://github.com/NVIDIA/apex/commit/6ff45486f432f91eb86937a0def5eb5f2cf792ae) is backward compatible. I'll try to get a PR opened in Apex, although it'll probably be 2024 before we can get it in. Once that issue is debugged, I propose that we build a custom container to use in Jenkins instead of waiting for it to get included in the official PyTorch container.", "jenkins", "jenkins", "jenkins", "@ericharper \r\nCan you do the final check and help merge?", "jenkins", "@athitten \r\nCan you help review this PR? I think @ericharper is away.", "jenkins", "jenkins", "@timmoon10 Can you check why this fails the CI?", "The error message is \"No space left on device\", so I suspect it's related to the recent file system issues on DLCluster. I find I often need to run a couple times to get past these errors, as well as segfaults coming from ASR.", "jenkins", "jenkins", "jenkins", "jenkins", "jenkins", "jenkins" ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" }, { "id": 4840216855, "node_id": "LA_kwDOC_bI7s8AAAABIH_ZFw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/CI", "name": "CI", "color": "1D76DB", "default": false, "description": "" } ]

[]

[ "jenkins" ]

[ { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null } ]

[ { "login": "ericharper", "id": 11999610, "node_id": "MDQ6VXNlcjExOTk5NjEw", "avatar_url": "https://avatars.githubusercontent.com/u/11999610?v=4", "gravatar_id": "", "url": "https://api.github.com/users/ericharper", "html_url": "https://github.com/ericharper", "followers_url": "https://api.github.com/users/ericharper/followers", "following_url": "https://api.github.com/users/ericharper/following{/other_user}", "gists_url": "https://api.github.com/users/ericharper/gists{/gist_id}", "starred_url": "https://api.github.com/users/ericharper/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/ericharper/subscriptions", "organizations_url": "https://api.github.com/users/ericharper/orgs", "repos_url": "https://api.github.com/users/ericharper/repos", "events_url": "https://api.github.com/users/ericharper/events{/privacy}", "received_events_url": "https://api.github.com/users/ericharper/received_events", "type": "User", "user_view_type": "public", "site_admin": false }, { "login": "okuchaiev", "id": 22035961, "node_id": "MDQ6VXNlcjIyMDM1OTYx", "avatar_url": "https://avatars.githubusercontent.com/u/22035961?v=4", "gravatar_id": "", "url": "https://api.github.com/users/okuchaiev", "html_url": "https://github.com/okuchaiev", "followers_url": "https://api.github.com/users/okuchaiev/followers", "following_url": "https://api.github.com/users/okuchaiev/following{/other_user}", "gists_url": "https://api.github.com/users/okuchaiev/gists{/gist_id}", "starred_url": "https://api.github.com/users/okuchaiev/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/okuchaiev/subscriptions", "organizations_url": "https://api.github.com/users/okuchaiev/orgs", "repos_url": "https://api.github.com/users/okuchaiev/repos", "events_url": "https://api.github.com/users/okuchaiev/events{/privacy}", "received_events_url": "https://api.github.com/users/okuchaiev/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "This issue was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null }, { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "This PR was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null }, { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "jenkins", "jenkins", "jenkins", "jenkins", "jenkins", "jenkins", "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "This PR was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "jenkins" ]

[ { "id": 4811884691, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ikw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/ASR", "name": "ASR", "color": "F3B33E", "default": false, "description": "" } ]

[]

[ "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.", "Thanks !" ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "jenkins" ]

[ { "id": 1719393562, "node_id": "MDU6TGFiZWwxNzE5MzkzNTYy", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/feature", "name": "feature", "color": "68F3F0", "default": false, "description": "request/PR for a new feature" }, { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null } ]

[ { "login": "okuchaiev", "id": 22035961, "node_id": "MDQ6VXNlcjIyMDM1OTYx", "avatar_url": "https://avatars.githubusercontent.com/u/22035961?v=4", "gravatar_id": "", "url": "https://api.github.com/users/okuchaiev", "html_url": "https://github.com/okuchaiev", "followers_url": "https://api.github.com/users/okuchaiev/followers", "following_url": "https://api.github.com/users/okuchaiev/following{/other_user}", "gists_url": "https://api.github.com/users/okuchaiev/gists{/gist_id}", "starred_url": "https://api.github.com/users/okuchaiev/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/okuchaiev/subscriptions", "organizations_url": "https://api.github.com/users/okuchaiev/orgs", "repos_url": "https://api.github.com/users/okuchaiev/repos", "events_url": "https://api.github.com/users/okuchaiev/events{/privacy}", "received_events_url": "https://api.github.com/users/okuchaiev/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "This issue was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 4847373924, "node_id": "LA_kwDOC_bI7s8AAAABIO0OZA", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/common", "name": "common", "color": "c5def5", "default": false, "description": "" } ]

[]

[ "jenkins", "> Is there any update on this ? @stevehuang52 @anteju\r\n\r\n@titu1994 Nope, I think it's ready to merge if there're no other comments.", "jenkins" ]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" }, { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null } ]

[]

[ "This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.", "This issue was closed because it has been inactive for 7 days since being marked as stale." ]

[ { "id": 4811886210, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ogg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/NLP", "name": "NLP", "color": "CBF09A", "default": false, "description": "" } ]

[]

[ "jenkins", "jenkins", "jenkins", "@erhoo82 Thanks for hard work. Do you have timeline in mind when the PR will get merged? Thanks!", "@stu1130 Have been caught up by high priority work. Trying to close soon.", "@athitten, @ericharper \r\n- I have removed the torch FSDP patch file as the bug was fixed at PyT23.09 container.\r\n- I have fixed some confusion in the mixed precision definition. Currently, NeMo sets the master param data type as FP32 when `megatron_O2=False`, which is the case of FSDP. I removed explicit model data type casting to FP32 after FSDP because it is already FP32 when `megatron_O2=False`. Also, `param_dtype` in FSDP sets only the compute data type in forward and backward propagation, which matches the original intention of FSDP design.", "jenkins", "@athitten Is the below error relevant?\r\n`ModuleNotFoundError: No module named 'transformer_engine'`", "jenkins", "jenkins", "jenkins", "jenkins" ]

[ { "id": 4585431311, "node_id": "LA_kwDOC_bI7s8AAAABEVAhDw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/stale", "name": "stale", "color": "ededed", "default": false, "description": null }, { "id": 4811884691, "node_id": "LA_kwDOC_bI7s8AAAABHs-Ikw", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/ASR", "name": "ASR", "color": "F3B33E", "default": false, "description": "" } ]

[ { "login": "tango4j", "id": 13211483, "node_id": "MDQ6VXNlcjEzMjExNDgz", "avatar_url": "https://avatars.githubusercontent.com/u/13211483?v=4", "gravatar_id": "", "url": "https://api.github.com/users/tango4j", "html_url": "https://github.com/tango4j", "followers_url": "https://api.github.com/users/tango4j/followers", "following_url": "https://api.github.com/users/tango4j/following{/other_user}", "gists_url": "https://api.github.com/users/tango4j/gists{/gist_id}", "starred_url": "https://api.github.com/users/tango4j/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/tango4j/subscriptions", "organizations_url": "https://api.github.com/users/tango4j/orgs", "repos_url": "https://api.github.com/users/tango4j/repos", "events_url": "https://api.github.com/users/tango4j/events{/privacy}", "received_events_url": "https://api.github.com/users/tango4j/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days." ]