Datasets:

DanielPFlorian

/

Transformers-Github-Issues

like 0

{ "login": "Saibo-creator", "id": 53392976, "node_id": "MDQ6VXNlcjUzMzkyOTc2", "avatar_url": "https://avatars.githubusercontent.com/u/53392976?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Saibo-creator", "html_url": "https://github.com/Saibo-creator", "followers_url": "https://api.github.com/users/Saibo-creator/followers", "following_url": "https://api.github.com/users/Saibo-creator/following{/other_user}", "gists_url": "https://api.github.com/users/Saibo-creator/gists{/gist_id}", "starred_url": "https://api.github.com/users/Saibo-creator/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Saibo-creator/subscriptions", "organizations_url": "https://api.github.com/users/Saibo-creator/orgs", "repos_url": "https://api.github.com/users/Saibo-creator/repos", "events_url": "https://api.github.com/users/Saibo-creator/events{/privacy}", "received_events_url": "https://api.github.com/users/Saibo-creator/received_events", "type": "User", "site_admin": false }

[ { "id": 2648621985, "node_id": "MDU6TGFiZWwyNjQ4NjIxOTg1", "url": "https://api.github.com/repos/huggingface/transformers/labels/Feature%20request", "name": "Feature request", "color": "FBCA04", "default": false, "description": "Request for a new feature" } ]

[]

[ "FYI @gante ", "Hi @Saibo-creator 👋 \r\n\r\nI'll do my usual bargain! I haven't seen demand for this feature, so my current decision is negative. However, if this comment reaches 10 reactions within a 3-month time spawn, I'll revisit this decision -- it means the community is interested in the feature 🙌 \r\n\r\n(whoever does the 10th reaction, please ping me)" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28356/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "srikant86panda", "id": 18262494, "node_id": "MDQ6VXNlcjE4MjYyNDk0", "avatar_url": "https://avatars.githubusercontent.com/u/18262494?v=4", "gravatar_id": "", "url": "https://api.github.com/users/srikant86panda", "html_url": "https://github.com/srikant86panda", "followers_url": "https://api.github.com/users/srikant86panda/followers", "following_url": "https://api.github.com/users/srikant86panda/following{/other_user}", "gists_url": "https://api.github.com/users/srikant86panda/gists{/gist_id}", "starred_url": "https://api.github.com/users/srikant86panda/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/srikant86panda/subscriptions", "organizations_url": "https://api.github.com/users/srikant86panda/orgs", "repos_url": "https://api.github.com/users/srikant86panda/repos", "events_url": "https://api.github.com/users/srikant86panda/events{/privacy}", "received_events_url": "https://api.github.com/users/srikant86panda/received_events", "type": "User", "site_admin": false }

[]

[]

[ "cc @Rocketknight1 ", "Hi @srikant86panda, Mistral and Mixtral both use the LLaMA prompt format, which does not include a special token at the start of assistant responses. Instead, they enforce that all conversations must alternate user/assistant/user/assistant, and the assistant response is always written immediately after the user message is finished with the `[/INST]` token.\r\n\r\nAs a result, `add_generation_prompt` does not have any effect for those models, and this is the correct and intended behaviour. I'm going to close this issue for now, but please let me know if you have any further problems or questions!", "@Rocketknight1 thanks for clarifying it. My concern is good to be closed." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28355/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "SangbumChoi", "id": 34004152, "node_id": "MDQ6VXNlcjM0MDA0MTUy", "avatar_url": "https://avatars.githubusercontent.com/u/34004152?v=4", "gravatar_id": "", "url": "https://api.github.com/users/SangbumChoi", "html_url": "https://github.com/SangbumChoi", "followers_url": "https://api.github.com/users/SangbumChoi/followers", "following_url": "https://api.github.com/users/SangbumChoi/following{/other_user}", "gists_url": "https://api.github.com/users/SangbumChoi/gists{/gist_id}", "starred_url": "https://api.github.com/users/SangbumChoi/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/SangbumChoi/subscriptions", "organizations_url": "https://api.github.com/users/SangbumChoi/orgs", "repos_url": "https://api.github.com/users/SangbumChoi/repos", "events_url": "https://api.github.com/users/SangbumChoi/events{/privacy}", "received_events_url": "https://api.github.com/users/SangbumChoi/received_events", "type": "User", "site_admin": false }

[]

[]

[ "@amyeroberts Hi back again with the test!\r\nMainly I added function called test_forward_auxiliary_loss. This will check if the two major components when the config.auxiliary_loss is set to True\r\n1. whether output of auxiliary loss is not None\r\n2. check the output len is equivalent to the num_hidden_layers - 1 (-1 is due to last layers is not 'aux')\r\n\r\nBTW, this test will also check in not only segmentation but also object_detection_model!\r\n\r\nfollowing result is the pytest of detr\r\n```\r\nroot@0a2b4fe54761:/mnt/nas2/users/sbchoi/transformers# RUN_SLOW=1 pytest tests/models/detr\r\n================================================================================ test session starts =================================================================================\r\nplatform linux -- Python 3.10.13, pytest-7.4.4, pluggy-1.0.0\r\nrootdir: /mnt/nas2/users/sbchoi/transformers\r\nconfigfile: pyproject.toml\r\nplugins: hypothesis-6.92.0, hydra-core-1.3.2\r\ncollected 154 items\r\n\r\ntests/models/detr/test_image_processing_detr.py ................ [ 10%]\r\ntests/models/detr/test_modeling_detr.py .......................ssssss..sssssssss......s...............s......s........s....ssssssss.ss.sssssssssssssss.s..s.......s........... [ 97%]\r\n.... [100%]\r\n\r\n================================================================================== warnings summary ==================================================================================\r\n../../../../../opt/conda/lib/python3.10/site-packages/_pytest/config/__init__.py:1373\r\n /opt/conda/lib/python3.10/site-packages/_pytest/config/__init__.py:1373: PytestConfigWarning: Unknown config option: doctest_glob\r\n\r\n self._warn_or_fail_if_strict(f\"Unknown config option: {key}\\n\")\r\n\r\nsrc/transformers/deepspeed.py:23\r\n /mnt/nas2/users/sbchoi/transformers/src/transformers/deepspeed.py:23: FutureWarning: transformers.deepspeed module is deprecated and will be removed in a future version. Please import deepspeed modules directly from transformers.integrations\r\n warnings.warn(\r\n\r\n../../../../../opt/conda/lib/python3.10/site-packages/torch/utils/cpp_extension.py:28\r\n /opt/conda/lib/python3.10/site-packages/torch/utils/cpp_extension.py:28: DeprecationWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html\r\n from pkg_resources import packaging # type: ignore[attr-defined]\r\n\r\n../../../../../opt/conda/lib/python3.10/site-packages/pkg_resources/__init__.py:2871\r\n../../../../../opt/conda/lib/python3.10/site-packages/pkg_resources/__init__.py:2871\r\n /opt/conda/lib/python3.10/site-packages/pkg_resources/__init__.py:2871: DeprecationWarning: Deprecated call to `pkg_resources.declare_namespace('ruamel')`.\r\n Implementing implicit namespace packages (as specified in PEP 420) is preferred to `pkg_resources.declare_namespace`. See https://setuptools.pypa.io/en/latest/references/keywords.html#keyword-namespace-packages\r\n declare_namespace(pkg)\r\n\r\ntests/models/detr/test_modeling_detr.py::DetrModelTest::test_disk_offload_bin\r\n /opt/conda/lib/python3.10/site-packages/torch/_utils.py:831: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly. To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()\r\n return self.fget.__get__(instance, owner)()\r\n\r\ntests/models/detr/test_modeling_detr.py::DetrModelTest::test_fast_init_context_manager\r\n /mnt/nas2/users/sbchoi/transformers/tests/test_modeling_common.py:460: FutureWarning: `torch.testing.assert_allclose()` is deprecated since 1.12 and will be removed in a future release. Please use `torch.testing.assert_close()` instead. You can find detailed upgrade instructions in https://github.com/pytorch/pytorch/issues/61844.\r\n torch.testing.assert_allclose(init_instance.linear.bias, expected_bias, rtol=1e-3, atol=1e-4)\r\n\r\ntests/models/detr/test_modeling_detr.py::DetrModelTest::test_fast_init_context_manager\r\n /mnt/nas2/users/sbchoi/transformers/tests/test_modeling_common.py:463: FutureWarning: `torch.testing.assert_allclose()` is deprecated since 1.12 and will be removed in a future release. Please use `torch.testing.assert_close()` instead. You can find detailed upgrade instructions in https://github.com/pytorch/pytorch/issues/61844.\r\n torch.testing.assert_allclose(\r\n\r\ntests/models/detr/test_modeling_detr.py::DetrModelTest::test_pipeline_image_segmentation\r\ntests/models/detr/test_modeling_detr.py::DetrModelTest::test_pipeline_object_detection\r\n /opt/conda/lib/python3.10/site-packages/huggingface_hub/repocard.py:105: UserWarning: Repo card metadata block was not found. Setting CardData to empty.\r\n warnings.warn(\"Repo card metadata block was not found. Setting CardData to empty.\")\r\n\r\n-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html\r\n============================================================== 107 passed, 47 skipped, 10 warnings in 110.95s (0:01:50) ==============================================================\r\nroot@0a2b4fe54761:/mnt/nas2/users/sbchoi/transformers#\r\n```", "@amyeroberts No problem! However, should I make another PR or just modify in this PR?", "@SangbumChoi Up to you. I'd suggest making a follow-up PR to prevent holding this one up, but happy to go for whatever you'd prefer", "@amyeroberts Yeah, I will make follow-up PR and link this PR also. If there's nothing else to add for this topic, let's MERGE this first :)", "@SangbumChoi Merged! Thanks again for this contribution 💪 " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28354/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28354", "html_url": "https://github.com/huggingface/transformers/pull/28354", "diff_url": "https://github.com/huggingface/transformers/pull/28354.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28354.patch", "merged_at": 1704795428000 }

{ "login": "minmie", "id": 40080081, "node_id": "MDQ6VXNlcjQwMDgwMDgx", "avatar_url": "https://avatars.githubusercontent.com/u/40080081?v=4", "gravatar_id": "", "url": "https://api.github.com/users/minmie", "html_url": "https://github.com/minmie", "followers_url": "https://api.github.com/users/minmie/followers", "following_url": "https://api.github.com/users/minmie/following{/other_user}", "gists_url": "https://api.github.com/users/minmie/gists{/gist_id}", "starred_url": "https://api.github.com/users/minmie/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/minmie/subscriptions", "organizations_url": "https://api.github.com/users/minmie/orgs", "repos_url": "https://api.github.com/users/minmie/repos", "events_url": "https://api.github.com/users/minmie/events{/privacy}", "received_events_url": "https://api.github.com/users/minmie/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hey! You should have a look at https://github.com/huggingface/tokenizers/issues/203 it's expected! 🤗 ", "@ArthurZucker thanks for your help." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28353/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "jp1924", "id": 93233241, "node_id": "U_kgDOBY6gWQ", "avatar_url": "https://avatars.githubusercontent.com/u/93233241?v=4", "gravatar_id": "", "url": "https://api.github.com/users/jp1924", "html_url": "https://github.com/jp1924", "followers_url": "https://api.github.com/users/jp1924/followers", "following_url": "https://api.github.com/users/jp1924/following{/other_user}", "gists_url": "https://api.github.com/users/jp1924/gists{/gist_id}", "starred_url": "https://api.github.com/users/jp1924/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/jp1924/subscriptions", "organizations_url": "https://api.github.com/users/jp1924/orgs", "repos_url": "https://api.github.com/users/jp1924/repos", "events_url": "https://api.github.com/users/jp1924/events{/privacy}", "received_events_url": "https://api.github.com/users/jp1924/received_events", "type": "User", "site_admin": false }

[]

[]

[ "cc @pacman100 ", "# Reproduction Code\r\n```python\r\nfrom transformers import (\r\n TrainingArguments,\r\n Trainer,\r\n BertForSequenceClassification,\r\n BertTokenizer,\r\n)\r\nfrom datasets import load_dataset\r\nfrom transformers.utils import check_min_version\r\nfrom accelerate import __version__ as accelerate_version\r\nfrom packaging import version\r\n\r\nif version.parse(accelerate_version) > version.parse(\"0.24.1\"):\r\n raise RuntimeError(\r\n \"This error occurs in accelerate versions 0.24.1 and earlier. Please lower your accelerate version\"\r\n )\r\n\r\n# This error occurs in transformers 4.36.0 and later\r\ncheck_min_version(\"4.36.0\")\r\n\r\n\r\ndef main(training_args: TrainingArguments) -> None:\r\n model_name_or_path = \"brenomatos/bert-base-uncased\"\r\n tokenizer = BertTokenizer.from_pretrained(model_name_or_path)\r\n model = BertForSequenceClassification.from_pretrained(model_name_or_path)\r\n\r\n trainer = Trainer(\r\n model=model,\r\n tokenizer=tokenizer,\r\n args=training_args,\r\n )\r\n trainer.save_model()\r\n trainer.save_state()\r\n\r\n\r\nif \"__main__\" in __name__:\r\n training_args = TrainingArguments(\r\n output_dir=\"./output_dir\",\r\n run_name=\"FSDP_save_test\",\r\n fsdp=\"full_shard auto_wrap\",\r\n fsdp_transformer_layer_cls_to_wrap=\"BertEncoder\",\r\n )\r\n main(training_args)\r\n```\r\nrun script\r\n>torchrun --nproc_per_node=4 ./fsdp_error_test.py\r\n# Version\r\ntorch 2.0.1+cu118\r\ntransformers 4.36.2\r\naccelerate 0.24.0\r\n# Description\r\nThank you for your interest. @pacman100 \r\n \r\nFirst of all, this is an issue I discovered while trying to further pretrain the LLAMA2 13B model.\r\nMy computer was running low on memory, so if I tried to save a checkpoint in the middle of training, I would get a CPU OOM and the training would end.\r\nHowever, saving the checkpoint after the training was over did not cause the OOM, so I had been using `trainer.save_model()` to save the model after all the training was finished. \r\nHowever, when I saved the model after updating transformers to the latest version, the model was not saved properly.\r\n(No `*.bin` file was created in output_dir.)\r\n\r\nI looked up the problem and realized that it was caused by the `version.parse(accelerate_version) > version.parse(\"0.24.1\")` code added in transformers 4.36.0.\r\nI can understand why they added it, but I thought they should have at least included a warning, so I issued a PR.\r\n" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28352/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28352", "html_url": "https://github.com/huggingface/transformers/pull/28352", "diff_url": "https://github.com/huggingface/transformers/pull/28352.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28352.patch", "merged_at": null }

{ "login": "yuanwu2017", "id": 34643241, "node_id": "MDQ6VXNlcjM0NjQzMjQx", "avatar_url": "https://avatars.githubusercontent.com/u/34643241?v=4", "gravatar_id": "", "url": "https://api.github.com/users/yuanwu2017", "html_url": "https://github.com/yuanwu2017", "followers_url": "https://api.github.com/users/yuanwu2017/followers", "following_url": "https://api.github.com/users/yuanwu2017/following{/other_user}", "gists_url": "https://api.github.com/users/yuanwu2017/gists{/gist_id}", "starred_url": "https://api.github.com/users/yuanwu2017/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/yuanwu2017/subscriptions", "organizations_url": "https://api.github.com/users/yuanwu2017/orgs", "repos_url": "https://api.github.com/users/yuanwu2017/repos", "events_url": "https://api.github.com/users/yuanwu2017/events{/privacy}", "received_events_url": "https://api.github.com/users/yuanwu2017/received_events", "type": "User", "site_admin": false }

[]

[]

[]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28351/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28351", "html_url": "https://github.com/huggingface/transformers/pull/28351", "diff_url": "https://github.com/huggingface/transformers/pull/28351.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28351.patch", "merged_at": 1704453689000 }

{ "login": "yuanwu2017", "id": 34643241, "node_id": "MDQ6VXNlcjM0NjQzMjQx", "avatar_url": "https://avatars.githubusercontent.com/u/34643241?v=4", "gravatar_id": "", "url": "https://api.github.com/users/yuanwu2017", "html_url": "https://github.com/yuanwu2017", "followers_url": "https://api.github.com/users/yuanwu2017/followers", "following_url": "https://api.github.com/users/yuanwu2017/following{/other_user}", "gists_url": "https://api.github.com/users/yuanwu2017/gists{/gist_id}", "starred_url": "https://api.github.com/users/yuanwu2017/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/yuanwu2017/subscriptions", "organizations_url": "https://api.github.com/users/yuanwu2017/orgs", "repos_url": "https://api.github.com/users/yuanwu2017/repos", "events_url": "https://api.github.com/users/yuanwu2017/events{/privacy}", "received_events_url": "https://api.github.com/users/yuanwu2017/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hey! Sure, would you like to open a pr for a fix? " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28350/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "hi-sushanta", "id": 93595990, "node_id": "U_kgDOBZQpVg", "avatar_url": "https://avatars.githubusercontent.com/u/93595990?v=4", "gravatar_id": "", "url": "https://api.github.com/users/hi-sushanta", "html_url": "https://github.com/hi-sushanta", "followers_url": "https://api.github.com/users/hi-sushanta/followers", "following_url": "https://api.github.com/users/hi-sushanta/following{/other_user}", "gists_url": "https://api.github.com/users/hi-sushanta/gists{/gist_id}", "starred_url": "https://api.github.com/users/hi-sushanta/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/hi-sushanta/subscriptions", "organizations_url": "https://api.github.com/users/hi-sushanta/orgs", "repos_url": "https://api.github.com/users/hi-sushanta/repos", "events_url": "https://api.github.com/users/hi-sushanta/events{/privacy}", "received_events_url": "https://api.github.com/users/hi-sushanta/received_events", "type": "User", "site_admin": false }

[]

[]

[]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28349/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28349", "html_url": "https://github.com/huggingface/transformers/pull/28349", "diff_url": "https://github.com/huggingface/transformers/pull/28349.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28349.patch", "merged_at": 1704701946000 }

{ "login": "brresnic", "id": 6865869, "node_id": "MDQ6VXNlcjY4NjU4Njk=", "avatar_url": "https://avatars.githubusercontent.com/u/6865869?v=4", "gravatar_id": "", "url": "https://api.github.com/users/brresnic", "html_url": "https://github.com/brresnic", "followers_url": "https://api.github.com/users/brresnic/followers", "following_url": "https://api.github.com/users/brresnic/following{/other_user}", "gists_url": "https://api.github.com/users/brresnic/gists{/gist_id}", "starred_url": "https://api.github.com/users/brresnic/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/brresnic/subscriptions", "organizations_url": "https://api.github.com/users/brresnic/orgs", "repos_url": "https://api.github.com/users/brresnic/repos", "events_url": "https://api.github.com/users/brresnic/events{/privacy}", "received_events_url": "https://api.github.com/users/brresnic/received_events", "type": "User", "site_admin": false }

[ { "id": 2392046359, "node_id": "MDU6TGFiZWwyMzkyMDQ2MzU5", "url": "https://api.github.com/repos/huggingface/transformers/labels/Good%20Second%20Issue", "name": "Good Second Issue", "color": "dd935a", "default": false, "description": "Issues that are more difficult to do than \"Good First\" issues - give it a try if you want!" }, { "id": 2648621985, "node_id": "MDU6TGFiZWwyNjQ4NjIxOTg1", "url": "https://api.github.com/repos/huggingface/transformers/labels/Feature%20request", "name": "Feature request", "color": "FBCA04", "default": false, "description": "Request for a new feature" } ]

[]

[ "I think there is a draft PR here #26502 ans #27479, maybe working on sdpa might be better" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28348/reactions", "total_count": 1, "+1": 1, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "johnDonor", "id": 70208188, "node_id": "MDQ6VXNlcjcwMjA4MTg4", "avatar_url": "https://avatars.githubusercontent.com/u/70208188?v=4", "gravatar_id": "", "url": "https://api.github.com/users/johnDonor", "html_url": "https://github.com/johnDonor", "followers_url": "https://api.github.com/users/johnDonor/followers", "following_url": "https://api.github.com/users/johnDonor/following{/other_user}", "gists_url": "https://api.github.com/users/johnDonor/gists{/gist_id}", "starred_url": "https://api.github.com/users/johnDonor/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/johnDonor/subscriptions", "organizations_url": "https://api.github.com/users/johnDonor/orgs", "repos_url": "https://api.github.com/users/johnDonor/repos", "events_url": "https://api.github.com/users/johnDonor/events{/privacy}", "received_events_url": "https://api.github.com/users/johnDonor/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hey 🤗 thanks for opening an issue! \r\nYou should check [for OOM ](https://stackoverflow.com/questions/624857/finding-which-process-was-killed-by-linux-oom-killer) issues and whether or not this is what crached your machine. If yes, you should try to use float16 or bfloat16 for training rather than full precision. Also check your memory usage during training. \r\n\r\nAs this is related to custom code, could you ask your question on the [forum](https://discuss.huggingface.co/) instead? I'm sure the community will be of help! We try to keep the github issues for bugs/feature requests 🤗 \r\n\r\nThanks!", "Thanks for the kind comment! For OOM issue, do i have to check hard disk memory? If so, more than 300GB is left, but i guess you got the point. \r\nSince I'm using windows I installed bitsandbytes library compiled for windows, made by individual not official, I thought of some unknown incompatibility like thing happened, so I made new WSL2 Linux environment and run the code, which led me to \"Error out of memory at line 380 in file /mmfs1/gscratch/zlab/timdettmers/git/bitsandbytes/csrc/pythonInterface.c\" Does this indicate something??\r\n\r\n+ You mean deleting this issue and upload at forum instead? If so, I'll delete this and reupload this to forum. Waiting for your answer! Thank you so much.\r\n", "Memory issue is both RAM and DRAM (cpu and GPU) depending on the data processing (CPU) and the model running (GPU?) \r\nNo need to delete the issue. \r\nYes this indicates that you ran out of memory. What is the environnement that you are using? ", "Now I'm using Intel(R) Xeon(R) Gold 6326 CPU, and NVIDIA GeForce RTX 4090 GPU. I'm checking the resource used in the training session, dedicated GPU memory is used about 23.5/24GB and shared GPU memory is used about 5GB/128GB. I didn't check the CPU resource. ", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28347/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "Ayenem", "id": 50707385, "node_id": "MDQ6VXNlcjUwNzA3Mzg1", "avatar_url": "https://avatars.githubusercontent.com/u/50707385?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Ayenem", "html_url": "https://github.com/Ayenem", "followers_url": "https://api.github.com/users/Ayenem/followers", "following_url": "https://api.github.com/users/Ayenem/following{/other_user}", "gists_url": "https://api.github.com/users/Ayenem/gists{/gist_id}", "starred_url": "https://api.github.com/users/Ayenem/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Ayenem/subscriptions", "organizations_url": "https://api.github.com/users/Ayenem/orgs", "repos_url": "https://api.github.com/users/Ayenem/repos", "events_url": "https://api.github.com/users/Ayenem/events{/privacy}", "received_events_url": "https://api.github.com/users/Ayenem/received_events", "type": "User", "site_admin": false }

[ { "id": 2648621985, "node_id": "MDU6TGFiZWwyNjQ4NjIxOTg1", "url": "https://api.github.com/repos/huggingface/transformers/labels/Feature%20request", "name": "Feature request", "color": "FBCA04", "default": false, "description": "Request for a new feature" } ]

[]

[ "FYI @gante ", "@Ayenem that's a really cool use of `sequence_bias`! I had read about token healing in the past, and never thought about it when I added `sequence_bias`.\r\n\r\nWould you be up for adding it to `transformers`? I think we could include it inside `generate` (e.g. [here](https://github.com/huggingface/transformers/blob/6c78bbcb8320d316434262ef003251ca997db0d1/src/transformers/generation/utils.py#L1638)), passing a simple `token_healing=True` flag. What do you think?", "I'm glad you appreciate this use of your feature! Hyrum's law at play :)\r\n\r\nSounds good, I'll get started on a dev branch. Thanks for the pointer!" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28346/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "gary-young", "id": 56245046, "node_id": "MDQ6VXNlcjU2MjQ1MDQ2", "avatar_url": "https://avatars.githubusercontent.com/u/56245046?v=4", "gravatar_id": "", "url": "https://api.github.com/users/gary-young", "html_url": "https://github.com/gary-young", "followers_url": "https://api.github.com/users/gary-young/followers", "following_url": "https://api.github.com/users/gary-young/following{/other_user}", "gists_url": "https://api.github.com/users/gary-young/gists{/gist_id}", "starred_url": "https://api.github.com/users/gary-young/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/gary-young/subscriptions", "organizations_url": "https://api.github.com/users/gary-young/orgs", "repos_url": "https://api.github.com/users/gary-young/repos", "events_url": "https://api.github.com/users/gary-young/events{/privacy}", "received_events_url": "https://api.github.com/users/gary-young/received_events", "type": "User", "site_admin": false }

[ { "id": 1990918270, "node_id": "MDU6TGFiZWwxOTkwOTE4Mjcw", "url": "https://api.github.com/repos/huggingface/transformers/labels/Good%20First%20Issue", "name": "Good First Issue", "color": "bbf794", "default": false, "description": "" } ]

[]

[ "Hey! I think that we used Tuple[X] when there can be N instances of X. I don't mind using ellipsis. \r\ncc @Rocketknight1 who is our type expert! ", "@ArthurZucker Thank you for your reply! I am a new guy for type hint so I am not sure about the situation about the previous version (before Python 3.10) but in the lastest version (Python 3.11 or 3.12), the Tuple[X] just means a tuple with **ONE** element with type X. \r\n\r\nThis is an example provided by the document of Python: https://docs.python.org/3/library/typing.html\r\nAnd my Pylance checker has indeed raise a warning about it.", "And also the type hint `Tuple` and `tuple` are considered as the same:\r\n\r\n", "feel free to open a PR for a fix if you want 🤗 ", "Hi @gary-young, I think you're correct! We realize it would be quite a significant job to update all instances of this across the codebase, but if you want to contribute fixes to one or more files, we'd be happy to accept a PR.", "Ok! I am willing to fix some of them for those I can identify." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28345/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "HelloWorldLTY", "id": 43333475, "node_id": "MDQ6VXNlcjQzMzMzNDc1", "avatar_url": "https://avatars.githubusercontent.com/u/43333475?v=4", "gravatar_id": "", "url": "https://api.github.com/users/HelloWorldLTY", "html_url": "https://github.com/HelloWorldLTY", "followers_url": "https://api.github.com/users/HelloWorldLTY/followers", "following_url": "https://api.github.com/users/HelloWorldLTY/following{/other_user}", "gists_url": "https://api.github.com/users/HelloWorldLTY/gists{/gist_id}", "starred_url": "https://api.github.com/users/HelloWorldLTY/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/HelloWorldLTY/subscriptions", "organizations_url": "https://api.github.com/users/HelloWorldLTY/orgs", "repos_url": "https://api.github.com/users/HelloWorldLTY/repos", "events_url": "https://api.github.com/users/HelloWorldLTY/events{/privacy}", "received_events_url": "https://api.github.com/users/HelloWorldLTY/received_events", "type": "User", "site_admin": false }

[]

[]

[ "How much memory does the H100 have? Note that you need 70 billion * 2 = 140GB of RAM to run the model in bfloat16.", "Thanks, I have not thought that it is a problem caused by memory. I may seek to the help from other models.", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28344/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "xmy0916", "id": 43675899, "node_id": "MDQ6VXNlcjQzNjc1ODk5", "avatar_url": "https://avatars.githubusercontent.com/u/43675899?v=4", "gravatar_id": "", "url": "https://api.github.com/users/xmy0916", "html_url": "https://github.com/xmy0916", "followers_url": "https://api.github.com/users/xmy0916/followers", "following_url": "https://api.github.com/users/xmy0916/following{/other_user}", "gists_url": "https://api.github.com/users/xmy0916/gists{/gist_id}", "starred_url": "https://api.github.com/users/xmy0916/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/xmy0916/subscriptions", "organizations_url": "https://api.github.com/users/xmy0916/orgs", "repos_url": "https://api.github.com/users/xmy0916/repos", "events_url": "https://api.github.com/users/xmy0916/events{/privacy}", "received_events_url": "https://api.github.com/users/xmy0916/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hey 🤗 thanks for opening an issue! We try to keep the github issues for bugs/feature requests.\r\nCould you ask your question on the [forum](https://discuss.huggingface.co/) instead? I'm sure the community will be of help!\r\n\r\nThanks!" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28343/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "drunkcoding", "id": 14305648, "node_id": "MDQ6VXNlcjE0MzA1NjQ4", "avatar_url": "https://avatars.githubusercontent.com/u/14305648?v=4", "gravatar_id": "", "url": "https://api.github.com/users/drunkcoding", "html_url": "https://github.com/drunkcoding", "followers_url": "https://api.github.com/users/drunkcoding/followers", "following_url": "https://api.github.com/users/drunkcoding/following{/other_user}", "gists_url": "https://api.github.com/users/drunkcoding/gists{/gist_id}", "starred_url": "https://api.github.com/users/drunkcoding/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/drunkcoding/subscriptions", "organizations_url": "https://api.github.com/users/drunkcoding/orgs", "repos_url": "https://api.github.com/users/drunkcoding/repos", "events_url": "https://api.github.com/users/drunkcoding/events{/privacy}", "received_events_url": "https://api.github.com/users/drunkcoding/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hey! Before I can help you, could provide an actual reproducer? \r\nThis one fails with `TypeError: The current model class (SwitchTransformersModel) is not compatible with `.generate()`, as it doesn't have a language model head. Please use one of the following classes instead: {'SwitchTransformersForConditionalGeneration'}`.\r\n", "Moreover when I use the correct class, disabeling `return_dict_in_generate` for convenience, I always get the same output. \r\nhttps://github.com/huggingface/transformers/blob/90224dd59e92e11d99b5b09be84d3fe7794636b9/src/transformers/models/switch_transformers/modeling_switch_transformers.py#L171 \r\n\r\nadds jitter only if training", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28342/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "susnato", "id": 56069179, "node_id": "MDQ6VXNlcjU2MDY5MTc5", "avatar_url": "https://avatars.githubusercontent.com/u/56069179?v=4", "gravatar_id": "", "url": "https://api.github.com/users/susnato", "html_url": "https://github.com/susnato", "followers_url": "https://api.github.com/users/susnato/followers", "following_url": "https://api.github.com/users/susnato/following{/other_user}", "gists_url": "https://api.github.com/users/susnato/gists{/gist_id}", "starred_url": "https://api.github.com/users/susnato/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/susnato/subscriptions", "organizations_url": "https://api.github.com/users/susnato/orgs", "repos_url": "https://api.github.com/users/susnato/repos", "events_url": "https://api.github.com/users/susnato/events{/privacy}", "received_events_url": "https://api.github.com/users/susnato/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Done! @ArthurZucker ", "CI is green now! @ArthurZucker ", "Thanks for the contribution! FYI @pacman100 and @younesbelkada " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28341/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28341", "html_url": "https://github.com/huggingface/transformers/pull/28341", "diff_url": "https://github.com/huggingface/transformers/pull/28341.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28341.patch", "merged_at": 1704469615000 }

{ "login": "ylacombe", "id": 52246514, "node_id": "MDQ6VXNlcjUyMjQ2NTE0", "avatar_url": "https://avatars.githubusercontent.com/u/52246514?v=4", "gravatar_id": "", "url": "https://api.github.com/users/ylacombe", "html_url": "https://github.com/ylacombe", "followers_url": "https://api.github.com/users/ylacombe/followers", "following_url": "https://api.github.com/users/ylacombe/following{/other_user}", "gists_url": "https://api.github.com/users/ylacombe/gists{/gist_id}", "starred_url": "https://api.github.com/users/ylacombe/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/ylacombe/subscriptions", "organizations_url": "https://api.github.com/users/ylacombe/orgs", "repos_url": "https://api.github.com/users/ylacombe/repos", "events_url": "https://api.github.com/users/ylacombe/events{/privacy}", "received_events_url": "https://api.github.com/users/ylacombe/received_events", "type": "User", "site_admin": false }

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/transformers/pr_28340). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "Hey @amyeroberts, thanks for the quick review and the relevant feedback, I've modified the logic a bit, WDYT ? There's also a pending comment on testing for which I'll wait your comment!" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28340/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28340", "html_url": "https://github.com/huggingface/transformers/pull/28340", "diff_url": "https://github.com/huggingface/transformers/pull/28340.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28340.patch", "merged_at": 1704386454000 }

{ "login": "DocShotgun", "id": 126566557, "node_id": "U_kgDOB4tAnQ", "avatar_url": "https://avatars.githubusercontent.com/u/126566557?v=4", "gravatar_id": "", "url": "https://api.github.com/users/DocShotgun", "html_url": "https://github.com/DocShotgun", "followers_url": "https://api.github.com/users/DocShotgun/followers", "following_url": "https://api.github.com/users/DocShotgun/following{/other_user}", "gists_url": "https://api.github.com/users/DocShotgun/gists{/gist_id}", "starred_url": "https://api.github.com/users/DocShotgun/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/DocShotgun/subscriptions", "organizations_url": "https://api.github.com/users/DocShotgun/orgs", "repos_url": "https://api.github.com/users/DocShotgun/repos", "events_url": "https://api.github.com/users/DocShotgun/events{/privacy}", "received_events_url": "https://api.github.com/users/DocShotgun/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hey! Thanks for the report, here are potential PRs that I would suspect: \r\n- #28142 which fixes some of the FA2 bugs\r\n- #28061 wich should fix gradient checkpointing issues\r\nPinging @younesbelkada for when he comes back, would be great if you can isolate the commit that lead to this in the meantime 🤗 ", "This may not be relevant to you but I found [this recent change](https://github.com/OpenAccess-AI-Collective/axolotl/commit/4d2e842e46bf8bd6dd0fda4d2667a7e7d80b4cd4) to Axolotl has made a significant difference to VRAM usage. Previously I could just squeeze in a LoRA on a 34B model on my 3x3090s at batch size 2, seq length 4096, now it OOMs immediately. I undid the change and it fits again. ", "> This may not be relevant to you but I found [this recent change](https://github.com/OpenAccess-AI-Collective/axolotl/commit/4d2e842e46bf8bd6dd0fda4d2667a7e7d80b4cd4) to Axolotl has made a significant difference to VRAM usage. Previously I could just squeeze in a LoRA on a 34B model on my 3x3090s at batch size 2, seq length 4096, now it OOMs immediately. I undid the change and it fits again.\r\n\r\nHmm it's certainly possible since that commit was in between when I did my initial train and the run where I had to drop the batch size. Unfortunately don't have a training instance up right now, so I'd have to test it the next time I try to train.", "I've determined that the cause of the increased VRAM usage was indeed axolotl changing the default for use_reentrant to False for gradient checkpointing. Going to go ahead and close the issue.", "thanks for sharing the solution! 🤗 " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28339/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "jiqing-feng", "id": 107918818, "node_id": "U_kgDOBm614g", "avatar_url": "https://avatars.githubusercontent.com/u/107918818?v=4", "gravatar_id": "", "url": "https://api.github.com/users/jiqing-feng", "html_url": "https://github.com/jiqing-feng", "followers_url": "https://api.github.com/users/jiqing-feng/followers", "following_url": "https://api.github.com/users/jiqing-feng/following{/other_user}", "gists_url": "https://api.github.com/users/jiqing-feng/gists{/gist_id}", "starred_url": "https://api.github.com/users/jiqing-feng/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/jiqing-feng/subscriptions", "organizations_url": "https://api.github.com/users/jiqing-feng/orgs", "repos_url": "https://api.github.com/users/jiqing-feng/repos", "events_url": "https://api.github.com/users/jiqing-feng/events{/privacy}", "received_events_url": "https://api.github.com/users/jiqing-feng/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hi @amyeroberts . Thanks for your review. The default dataclass works well as you said. However, sometimes we would like to enable other optimization like `model = torch.jit.trace(model, inputs)` (which will get a significant acceleration in both GPU and CPU), and it can't recognize the dataclass, so I set `return_dict=False` to return tuple outputs. \r\n\r\nI think considering different types of model outputs would be better since the models support return tuples.", "I agree with @amyeroberts in general with this PR.\r\n\r\nIMO pipeline shouldn't have any excessive amount of code to support use cases like this. If you really want to wrap a jitted model, wrap you own jitted model to return a dict instead of a tuple.\r\n\r\n```python\r\n\r\nclass Wrapper:\r\n def forward(self, ...):\r\n tuple = self.inner.forward(..)\r\n return ModelOutputXXX(tuple)\r\n```\r\n\r\nAlso why using `jit.trace` instead of `torch.compile` ? I though this was the new way to do it.\r\nWhat kind of speedups are you seeing ?", "Hi @amyeroberts @Narsil . Thanks for your review, I propose that since the transformers model supports return tuple, we should support tuple outputs in any other APIs that are related to the model's outputs.\r\n\r\nBut you were right, I can wrap the model in my script.\r\n\r\nThanks again for your review. Feel free to close this PR if you don't have any questions.", "@jiqing-feng Thanks for your time working on this. Following from the discussion, I'm closing this PR." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28338/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28338", "html_url": "https://github.com/huggingface/transformers/pull/28338", "diff_url": "https://github.com/huggingface/transformers/pull/28338.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28338.patch", "merged_at": null }

{ "login": "hxiaoyang", "id": 98200137, "node_id": "U_kgDOBdpqSQ", "avatar_url": "https://avatars.githubusercontent.com/u/98200137?v=4", "gravatar_id": "", "url": "https://api.github.com/users/hxiaoyang", "html_url": "https://github.com/hxiaoyang", "followers_url": "https://api.github.com/users/hxiaoyang/followers", "following_url": "https://api.github.com/users/hxiaoyang/following{/other_user}", "gists_url": "https://api.github.com/users/hxiaoyang/gists{/gist_id}", "starred_url": "https://api.github.com/users/hxiaoyang/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/hxiaoyang/subscriptions", "organizations_url": "https://api.github.com/users/hxiaoyang/orgs", "repos_url": "https://api.github.com/users/hxiaoyang/repos", "events_url": "https://api.github.com/users/hxiaoyang/events{/privacy}", "received_events_url": "https://api.github.com/users/hxiaoyang/received_events", "type": "User", "site_admin": false }

[]

[]

[ "You should use the `generate` function. Otherwise I have no idea what model you are using, but this is expected: the model does not necessarily predict the token you give as an input. Assisted decoding relies on this property. \r\n\r\n```python\r\nprompt = \"Please provide a code sample that reproduces the\"\r\nfrom transformers import AutoModelForCausalLM\r\nmodel = AutoModelForCausalLM.from_pretrained(\"gpt2\")\r\ninput_ids = tokenizer(prompt, return_tensors = \"pt\")\r\nstring = tokenizer.batch_decode(model.generate(**input_ids).tolist(),skip_special_tokens=True)\r\nprint(string)\r\n['Please provide a code sample that reproduces the code.\\n\\nIf you are interested in contributing to']\r\n```\r\nuse `max_new_tokens` if needed. ", "Thank you for the timely response! I've been using `llama-2-13b-chat-hf` FYI.", "A followup question: I'm building a conversational / recursive LM. For my use case, it's important that the previous conversation is not reprocessed. In other words, only previous hidden states and new context should be used. Is this the default behavior of `model.generate()`? If not, what would be the best way to reuse previous hidden states.\r\n\r\nThanks! @ArthurZucker", "By default the `generate` function will use the `past_key_values` instead of re-computing everything which is probably what you are looking for 😉 the past key values still take into account the previous context in an efficient manner. See an answer [here](https://discuss.huggingface.co/t/what-is-the-purpose-of-use-cache-in-decoder/958)", "Thank you!" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28337/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "sfc-gh-zhwang", "id": 135062830, "node_id": "U_kgDOCAzlLg", "avatar_url": "https://avatars.githubusercontent.com/u/135062830?v=4", "gravatar_id": "", "url": "https://api.github.com/users/sfc-gh-zhwang", "html_url": "https://github.com/sfc-gh-zhwang", "followers_url": "https://api.github.com/users/sfc-gh-zhwang/followers", "following_url": "https://api.github.com/users/sfc-gh-zhwang/following{/other_user}", "gists_url": "https://api.github.com/users/sfc-gh-zhwang/gists{/gist_id}", "starred_url": "https://api.github.com/users/sfc-gh-zhwang/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/sfc-gh-zhwang/subscriptions", "organizations_url": "https://api.github.com/users/sfc-gh-zhwang/orgs", "repos_url": "https://api.github.com/users/sfc-gh-zhwang/repos", "events_url": "https://api.github.com/users/sfc-gh-zhwang/events{/privacy}", "received_events_url": "https://api.github.com/users/sfc-gh-zhwang/received_events", "type": "User", "site_admin": false }

[]

[]

[ "You should use `forced_decoder_ids` https://huggingface.co./docs/transformers/main_classes/text_generation#transformers.GenerationConfig.forced_decoder_ids ", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28336/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "snailrowen1337", "id": 45402632, "node_id": "MDQ6VXNlcjQ1NDAyNjMy", "avatar_url": "https://avatars.githubusercontent.com/u/45402632?v=4", "gravatar_id": "", "url": "https://api.github.com/users/snailrowen1337", "html_url": "https://github.com/snailrowen1337", "followers_url": "https://api.github.com/users/snailrowen1337/followers", "following_url": "https://api.github.com/users/snailrowen1337/following{/other_user}", "gists_url": "https://api.github.com/users/snailrowen1337/gists{/gist_id}", "starred_url": "https://api.github.com/users/snailrowen1337/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/snailrowen1337/subscriptions", "organizations_url": "https://api.github.com/users/snailrowen1337/orgs", "repos_url": "https://api.github.com/users/snailrowen1337/repos", "events_url": "https://api.github.com/users/snailrowen1337/events{/privacy}", "received_events_url": "https://api.github.com/users/snailrowen1337/received_events", "type": "User", "site_admin": false }

[]

[]

[ "hey! Could you try upgrading your torch version? 1.13.0 is a bit old. \r\ncc @younesbelkada ", "@ArthurZucker -- I've updated to the latest pytorch but still face the issue", "Alright, now you are using a remote model so would recommend you to open the issue on `https://huggingface.co./NousResearch/Yarn-Mistral-7b-128k/discussions` unless this also fails with `mistralai/Mistral-7B-v0.1` 🤗 ", "@ArthurZucker I still face this issue with the mistral model. Other users have seen similar issues [https://github.com/huggingface/peft/issues/137] but these fixes do not work for me.\r\n\r\nCC @younesbelkada ", "Hi @snailrowen1337 \r\nThe error should disappear if you pass `gradient_checkpointing_kwargs={\"use_reentrant\":False}` in `TrainingArguments`", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28335/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "hadim", "id": 528003, "node_id": "MDQ6VXNlcjUyODAwMw==", "avatar_url": "https://avatars.githubusercontent.com/u/528003?v=4", "gravatar_id": "", "url": "https://api.github.com/users/hadim", "html_url": "https://github.com/hadim", "followers_url": "https://api.github.com/users/hadim/followers", "following_url": "https://api.github.com/users/hadim/following{/other_user}", "gists_url": "https://api.github.com/users/hadim/gists{/gist_id}", "starred_url": "https://api.github.com/users/hadim/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/hadim/subscriptions", "organizations_url": "https://api.github.com/users/hadim/orgs", "repos_url": "https://api.github.com/users/hadim/repos", "events_url": "https://api.github.com/users/hadim/events{/privacy}", "received_events_url": "https://api.github.com/users/hadim/received_events", "type": "User", "site_admin": false }

[]

[]

[ "```diff\r\n+ inputs = processor(raw_image, input_points=input_points, return_tensors=\"pt\").to(torch.float32).to(\"mps\")\r\n- inputs = processor(raw_image, input_points=input_points, return_tensors=\"pt\").to(\"mps\")\r\n```\r\nwill fix this. \r\nI think this is more a mps issue as the conversion works well otherwise. Do you want to open a PR for a fix? \r\n\r\nhttps://github.com/huggingface/transformers/blob/b1292bca6923cfbc9cb3f70cb55df57e4e17e630/src/transformers/models/sam/processing_sam.py#L123\r\n", "Using `.to(torch.float32)` makes it work indeed. Thanks.\r\n\r\nI'm happy to open a PR but then I am not sure what kind of fix it should contain here since it seems to be more of an `mps` issue and your simple fix makes it work.", "Yep let's keep it that way then 😉 " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28334/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "VictorSanh", "id": 16107619, "node_id": "MDQ6VXNlcjE2MTA3NjE5", "avatar_url": "https://avatars.githubusercontent.com/u/16107619?v=4", "gravatar_id": "", "url": "https://api.github.com/users/VictorSanh", "html_url": "https://github.com/VictorSanh", "followers_url": "https://api.github.com/users/VictorSanh/followers", "following_url": "https://api.github.com/users/VictorSanh/following{/other_user}", "gists_url": "https://api.github.com/users/VictorSanh/gists{/gist_id}", "starred_url": "https://api.github.com/users/VictorSanh/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/VictorSanh/subscriptions", "organizations_url": "https://api.github.com/users/VictorSanh/orgs", "repos_url": "https://api.github.com/users/VictorSanh/repos", "events_url": "https://api.github.com/users/VictorSanh/events{/privacy}", "received_events_url": "https://api.github.com/users/VictorSanh/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Adressed the comments and moved the dummy test case into proper tests.\r\nlet me know if you would like something more involved test wise!", "Good to go ! Feel free to merge @VictorSanh 🤗 ", "I am not cool enough to have merge access. The time where i am merging stuff whenever i wanted on hf transformers is well passed haha", "so either you @ArthurZucker or @younesbelkada need to merge lol 😅\r\nbut perhaps i can be promoted to core maintainer with that PR @LysandreJik ?", "Ooops 🤣 ", "Considering it @VictorSanh!", "Hi,\r\nI am still getting the following error when I'm trying to finetune the model for ConditionalGeneration in the forward call:\r\n\r\nfinal_labels[batch_indices, text_to_overwrite] = labels[batch_indices, non_image_indices]\r\nIndexError: index 8 is out of bounds for dimension 1 with size 8\r\n\r\nThe same code works fine if I just change the model to another VLLM like InstructBlip.\r\n\r\nThank you and kind regards,\r\nAlexandros Xenos", "@alexandrosXe do you have a reproduction case we can start debugging from?", "> @alexandrosXe do you have a reproduction case we can start debugging from?\r\n@VictorSanh Thank you for replying so fast! \r\nThis code can reproduce my error: \r\n\r\n```\r\nfrom PIL import Image\r\nimport requests\r\nfrom transformers import AutoProcessor, LlavaForConditionalGeneration\r\n\r\nmodel = LlavaForConditionalGeneration.from_pretrained(\"llava-hf/llava-1.5-7b-hf\")\r\nprocessor = AutoProcessor.from_pretrained(\"llava-hf/llava-1.5-7b-hf\")\r\n\r\nprompt = \"<image>\\nUSER: What's the content of the image?\\nASSISTANT:\"\r\nurl = \"https://www.ilankelman.org/stopsigns/australia.jpg\"\r\nimage = Image.open(requests.get(url, stream=True).raw)\r\nanswer = \"The image has a stop sign in the corner of the road\"\r\n\r\n\r\ninputs = processor(text=prompt, images=image, return_tensors=\"pt\")\r\n\r\nlabels = processor.tokenizer(answer, return_tensors=\"pt\")\r\nlabel_ids = labels[\"input_ids\"]\r\nlabel_mask = labels[\"attention_mask\"].bool()\r\nlabel_ids = label_ids.masked_fill(~label_mask, -100) #We dont count the loss on the padded tokens\r\nloss = model(**inputs, labels = label_ids).loss\r\nprint(\"loss: \", loss)\r\n```", "Can you also make sure you are using the latest version of transformers? ", "> Can you also make sure you are using the latest version of transformers?\r\n\r\n@ArthurZucker I am using the transformers 4.37.2 version. ", "Yes, it seems `oss = model(**inputs, labels = inputs[\"input_ids\"])` works well however. Loss was made to be of the same size as the input ids:\r\n\r\n> labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):\r\n Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,\r\n config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored\r\n (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.\r\n\r\nfrom the doc it should be length sequence length", "@ArthurZucker Thank you, it was my fault using wrong labels. Now everything works fine!" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28333/reactions", "total_count": 1, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 1, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28333", "html_url": "https://github.com/huggingface/transformers/pull/28333", "diff_url": "https://github.com/huggingface/transformers/pull/28333.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28333.patch", "merged_at": 1704872013000 }

{ "login": "weimingzha0", "id": 38259546, "node_id": "MDQ6VXNlcjM4MjU5NTQ2", "avatar_url": "https://avatars.githubusercontent.com/u/38259546?v=4", "gravatar_id": "", "url": "https://api.github.com/users/weimingzha0", "html_url": "https://github.com/weimingzha0", "followers_url": "https://api.github.com/users/weimingzha0/followers", "following_url": "https://api.github.com/users/weimingzha0/following{/other_user}", "gists_url": "https://api.github.com/users/weimingzha0/gists{/gist_id}", "starred_url": "https://api.github.com/users/weimingzha0/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/weimingzha0/subscriptions", "organizations_url": "https://api.github.com/users/weimingzha0/orgs", "repos_url": "https://api.github.com/users/weimingzha0/repos", "events_url": "https://api.github.com/users/weimingzha0/events{/privacy}", "received_events_url": "https://api.github.com/users/weimingzha0/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Closing as the PR was merged" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28332/reactions", "total_count": 1, "+1": 1, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "weimingzha0", "id": 38259546, "node_id": "MDQ6VXNlcjM4MjU5NTQ2", "avatar_url": "https://avatars.githubusercontent.com/u/38259546?v=4", "gravatar_id": "", "url": "https://api.github.com/users/weimingzha0", "html_url": "https://github.com/weimingzha0", "followers_url": "https://api.github.com/users/weimingzha0/followers", "following_url": "https://api.github.com/users/weimingzha0/following{/other_user}", "gists_url": "https://api.github.com/users/weimingzha0/gists{/gist_id}", "starred_url": "https://api.github.com/users/weimingzha0/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/weimingzha0/subscriptions", "organizations_url": "https://api.github.com/users/weimingzha0/orgs", "repos_url": "https://api.github.com/users/weimingzha0/repos", "events_url": "https://api.github.com/users/weimingzha0/events{/privacy}", "received_events_url": "https://api.github.com/users/weimingzha0/received_events", "type": "User", "site_admin": false }

[]

[]

[ "> torch bin files are now deprecated in favor of safetensors but no harm in improving this! Coul you add a test as well in test_modelling_common? 🤗\r\n\r\nSure. Please let me know if my test case is correct.", "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/transformers/pr_28331). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "> Thanks! My last query would be to make sure this works for the deepspeed / sdpa case (device_map = \"meta\")!\r\n\r\nDo you mean add a test case for device_map == \"meta' ?", "That would be a good way of making sure this will be fine with DeepSpeedZero (the if branch) 🤗 ", "> That would be a good way of making sure this will be fine with DeepSpeedZero (the if branch) 🤗\r\n\r\nCurrently none of existing tests covers the branch: \r\nI even added an assert in the \"if branch\" and all tests still passed except for one irrelevant error (see https://github.com/huggingface/transformers/pull/28401)\r\n\r\nAnyway, I added a guard for \"meta\" device.\r\n\r\n" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28331/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28331", "html_url": "https://github.com/huggingface/transformers/pull/28331", "diff_url": "https://github.com/huggingface/transformers/pull/28331.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28331.patch", "merged_at": 1704877051000 }

{ "login": "Taishi-N324", "id": 82321333, "node_id": "MDQ6VXNlcjgyMzIxMzMz", "avatar_url": "https://avatars.githubusercontent.com/u/82321333?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Taishi-N324", "html_url": "https://github.com/Taishi-N324", "followers_url": "https://api.github.com/users/Taishi-N324/followers", "following_url": "https://api.github.com/users/Taishi-N324/following{/other_user}", "gists_url": "https://api.github.com/users/Taishi-N324/gists{/gist_id}", "starred_url": "https://api.github.com/users/Taishi-N324/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Taishi-N324/subscriptions", "organizations_url": "https://api.github.com/users/Taishi-N324/orgs", "repos_url": "https://api.github.com/users/Taishi-N324/repos", "events_url": "https://api.github.com/users/Taishi-N324/events{/privacy}", "received_events_url": "https://api.github.com/users/Taishi-N324/received_events", "type": "User", "site_admin": false }

[]

[]

[ "I think the idea is to remove the call to `model.to_bettertransformer()` and `with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=False, enable_mem_efficient=False):` . The error should be updated to mention this! would you like to open a PR?", "I'm sorry, I don't quite understand. What do you mean?\r\n\r\nAre you saying to remove both `model.to_bettertransformer()` and `with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=False, enable_mem_efficient=False):`?\r\n\r\nOriginally, what I want to do is to wrap GPTBigCode with a bettertransformer and then wrap it with fsdp to train it. However, I encounter this error during the process. I was trying with a minimal configuration, but I still encounter errors.", "It seems that GPTBigCode was supported during https://github.com/huggingface/optimum/pull/1252, but in the latest branch, it appears to no longer be supported according to https://github.com/huggingface/optimum/tree/main/optimum/bettertransformer/models.", "The error means that now transformers supports sdpa natively so you should not need to use better transformer for the usage you are looking for 🤗 ", "Thank you for your kindness!\r\nI understand now.\r\nThank you for your assistance." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28330/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "aureliusnoble", "id": 16746857, "node_id": "MDQ6VXNlcjE2NzQ2ODU3", "avatar_url": "https://avatars.githubusercontent.com/u/16746857?v=4", "gravatar_id": "", "url": "https://api.github.com/users/aureliusnoble", "html_url": "https://github.com/aureliusnoble", "followers_url": "https://api.github.com/users/aureliusnoble/followers", "following_url": "https://api.github.com/users/aureliusnoble/following{/other_user}", "gists_url": "https://api.github.com/users/aureliusnoble/gists{/gist_id}", "starred_url": "https://api.github.com/users/aureliusnoble/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/aureliusnoble/subscriptions", "organizations_url": "https://api.github.com/users/aureliusnoble/orgs", "repos_url": "https://api.github.com/users/aureliusnoble/repos", "events_url": "https://api.github.com/users/aureliusnoble/events{/privacy}", "received_events_url": "https://api.github.com/users/aureliusnoble/received_events", "type": "User", "site_admin": false }

[]

[]

[ "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored.", "Hey 🤗 thanks for opening an issue! We try to keep the github issues for bugs/feature requests.\r\nCould you ask your question on the [forum](https://discuss.huggingface.co/) instead? I'm sure the community will be of help!\r\n\r\nThanks!" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28329/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "michaelfeil", "id": 63565275, "node_id": "MDQ6VXNlcjYzNTY1Mjc1", "avatar_url": "https://avatars.githubusercontent.com/u/63565275?v=4", "gravatar_id": "", "url": "https://api.github.com/users/michaelfeil", "html_url": "https://github.com/michaelfeil", "followers_url": "https://api.github.com/users/michaelfeil/followers", "following_url": "https://api.github.com/users/michaelfeil/following{/other_user}", "gists_url": "https://api.github.com/users/michaelfeil/gists{/gist_id}", "starred_url": "https://api.github.com/users/michaelfeil/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/michaelfeil/subscriptions", "organizations_url": "https://api.github.com/users/michaelfeil/orgs", "repos_url": "https://api.github.com/users/michaelfeil/repos", "events_url": "https://api.github.com/users/michaelfeil/events{/privacy}", "received_events_url": "https://api.github.com/users/michaelfeil/received_events", "type": "User", "site_admin": false }

[ { "id": 2648621985, "node_id": "MDU6TGFiZWwyNjQ4NjIxOTg1", "url": "https://api.github.com/repos/huggingface/transformers/labels/Feature%20request", "name": "Feature request", "color": "FBCA04", "default": false, "description": "Request for a new feature" } ]

[]

[ "cc @younesbelkada ", "This is very cool ! We are definitely interested in adding HQQ inference support in transformers. The cool thing is that indeed it seems you don't need to pre-quantize the weights in order to quantize the models. We'll explore a bit on our side and let you know how it goes\r\ncc @SunMarc @Titus-von-Koeller ", "Hi! I am the maintainer of the HQQ project, happy to assist with anything needed !", "Very glad to e-meet you @mobicham ! do you have an email I can use so that we can contact you through Slack to iterate quickly?", "Glad to e-meet you @younesbelkada as well! Sure: [email protected]", "thanks @mobicham you should have received an invite by now! " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28328/reactions", "total_count": 3, "+1": 3, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "Cazforshort", "id": 6918831, "node_id": "MDQ6VXNlcjY5MTg4MzE=", "avatar_url": "https://avatars.githubusercontent.com/u/6918831?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Cazforshort", "html_url": "https://github.com/Cazforshort", "followers_url": "https://api.github.com/users/Cazforshort/followers", "following_url": "https://api.github.com/users/Cazforshort/following{/other_user}", "gists_url": "https://api.github.com/users/Cazforshort/gists{/gist_id}", "starred_url": "https://api.github.com/users/Cazforshort/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Cazforshort/subscriptions", "organizations_url": "https://api.github.com/users/Cazforshort/orgs", "repos_url": "https://api.github.com/users/Cazforshort/repos", "events_url": "https://api.github.com/users/Cazforshort/events{/privacy}", "received_events_url": "https://api.github.com/users/Cazforshort/received_events", "type": "User", "site_admin": false }

[]

[]

[ "cc @ylacombe as well. Seems like this should be supported from the doc, so might just be the generation config not being initialized properly. \r\nMaybe `model.generation_config = model.config` will fix this in the meantime @Cazforshort ", "> model.generation_config = model.config\r\n\r\nNo luck\r\n\r\n```\r\nMODEL_NAME = \"suno/bark\"\r\nvoice_preset = \"v2/en_speaker_9\"\r\nsampling_rate = 24000\r\ndevice = \"cuda:0\" if torch.cuda.is_available() else \"cpu\"\r\nprocessor = AutoProcessor.from_pretrained(MODEL_NAME)\r\nmin_eos_p=0.5\r\n\r\ntext_prompt = \"[frowns slightly] Hmmph, typical. Ignoring people is quite common among adults, especially when faced with unknown situations. However, we must press forward nonetheless.\"\r\nsentences = nltk.sent_tokenize(text_prompt)\r\n\r\n\r\n# Initializing Bark sub-modules configurations.\r\nsemantic_config = BarkSemanticConfig(min_eos_p=min_eos_p)\r\ncoarse_acoustics_config = BarkCoarseConfig()\r\nfine_acoustics_config = BarkFineConfig()\r\ncodec_config = AutoConfig.from_pretrained(\"facebook/encodec_24khz\")\r\nconfiguration = BarkConfig.from_sub_model_configs(\r\n semantic_config, coarse_acoustics_config, fine_acoustics_config, codec_config\r\n )\r\nmodel = BarkModel(config = configuration).to(device)\r\nmodel.generation_config = model.config\r\n```\r\n\r\nTypeError: transformers.models.bark.generation_configuration_bark.BarkSemanticGenerationConfig() argument after ** must be a mapping, not BarkSemanticConfig", "Hi @Cazforshort, you have to use a `BarkGenerationConfig` instead of a `GenerationConfig`:\r\n```python\r\nfrom transformers.models.bark.generation_configuration_bark import BarkGenerationConfig\r\n\r\nmodel.generation_config = BarkGenerationConfig()\r\n```\r\n\r\nBTW, to modify `min_eos_p`, you don't have to necessarily pass by the generation config, you can simply pass `min_eos_p=XXX` to `BarkModel.generate`. \r\n", "> Hi @Cazforshort, you have to use a `BarkGenerationConfig` instead of a `GenerationConfig`:\r\n> \r\n> ```python\r\n> from transformers.models.bark.generation_configuration_bark import BarkGenerationConfig\r\n> \r\n> model.generation_config = BarkGenerationConfig()\r\n> ```\r\n> \r\n> BTW, to modify `min_eos_p`, you don't have to necessarily pass by the generation config, you can simply pass `min_eos_p=XXX` to `BarkModel.generate`.\r\n\r\nStill seeing the same errors. I tried just using that gen fig and also starting with the other config and overwriting, but I just get the same sort of errors.\r\n\r\n```\r\nfrom transformers import (\r\n BarkSemanticConfig,\r\n BarkCoarseConfig,\r\n BarkFineConfig,\r\n BarkModel,\r\n BarkConfig,\r\n AutoConfig,\r\n )\r\nfrom transformers.models.bark.generation_configuration_bark import BarkGenerationConfig\r\nimport torch\r\nfrom transformers import AutoProcessor, set_seed\r\nfrom optimum.bettertransformer import BetterTransformer\r\nimport scipy\r\nimport numpy as np\r\nimport nltk # we'll use this to split into sentences\r\n\r\n\r\nMODEL_NAME = \"suno/bark\"\r\nvoice_preset = \"v2/en_speaker_9\"\r\nsampling_rate = 24000\r\ndevice = \"cuda:0\" if torch.cuda.is_available() else \"cpu\"\r\nprocessor = AutoProcessor.from_pretrained(MODEL_NAME)\r\nmin_eos_p=0.5\r\n\r\ntext_prompt = \"[frowns slightly] Hmmph, typical. Ignoring people is quite common among adults, especially when faced with unknown situations. However, we must press forward nonetheless.\"\r\nsentences = nltk.sent_tokenize(text_prompt)\r\n\r\n# Initializing Bark sub-modules configurations.\r\nsemantic_config = BarkSemanticConfig(min_eos_p=min_eos_p)\r\ncoarse_acoustics_config = BarkCoarseConfig()\r\nfine_acoustics_config = BarkFineConfig()\r\ncodec_config = AutoConfig.from_pretrained(\"facebook/encodec_24khz\")\r\n\r\nconfiguration = BarkConfig.from_sub_model_configs(\r\n semantic_config, \r\n coarse_acoustics_config, \r\n fine_acoustics_config, \r\n codec_config\r\n )\r\ngen_configuration = BarkGenerationConfig(\r\n semantic_config, \r\n coarse_acoustics_config, \r\n fine_acoustics_config, \r\n codec_config\r\n )\r\n\r\nmodel = BarkModel(config = configuration).to(device)\r\nmodel.generation_config = gen_configuration\r\ninputs = processor(sentences,voice_preset=voice_preset).to(device)\r\n\r\noutput = model.generate(**inputs)\r\n```\r\n**Error:\r\nTypeError: transformers.models.bark.generation_configuration_bark.BarkSemanticGenerationConfig() argument after ** must be a mapping, not BarkSemanticGenerationConfig**\r\n\r\nmodel.generate seems to be ignoring the min_eos_p parameter. I tried passing it there but the results were identical unlike when I tested it without transformers.", "You have to use `BarkGenerationConfig.from_sub_model_configs` !\r\n\r\n> model.generate seems to be ignoring the min_eos_p parameter. I tried passing it there but the results were identical unlike when I tested it without transformers.\r\n\r\nCan you send a snippet for that ?", "> You have to use `BarkGenerationConfig.from_sub_model_configs` !\r\n> \r\n> > model.generate seems to be ignoring the min_eos_p parameter. I tried passing it there but the results were identical unlike when I tested it without transformers.\r\n> \r\n> Can you send a snippet for that ?\r\n\r\nHmm, same error about passing a config instead of a mapping.\r\n\r\n`MODEL_NAME = \"suno/bark\"\r\nvoice_preset = \"v2/en_speaker_9\"\r\nsampling_rate = 24000\r\ndevice = \"cuda:0\" if torch.cuda.is_available() else \"cpu\"\r\nprocessor = AutoProcessor.from_pretrained(MODEL_NAME)\r\nmin_eos_p=0.5\r\n\r\ntext_prompt = \"[frowns slightly] Hmmph, typical. Ignoring people is quite common among adults, especially when faced with unknown situations. However, we must press forward nonetheless.\"\r\nsentences = nltk.sent_tokenize(text_prompt)\r\n\r\n# Initializing Bark sub-modules configurations.\r\nsemantic_config = BarkSemanticConfig(min_eos_p=min_eos_p)\r\ncoarse_acoustics_config = BarkCoarseConfig()\r\nfine_acoustics_config = BarkFineConfig()\r\ncodec_config = AutoConfig.from_pretrained(\"facebook/encodec_24khz\")\r\n\r\nconfiguration = BarkConfig.from_sub_model_configs(\r\n semantic_config, \r\n coarse_acoustics_config, \r\n fine_acoustics_config, \r\n codec_config\r\n )\r\ngen_configuration = BarkGenerationConfig.from_sub_model_configs(\r\n semantic_config, \r\n coarse_acoustics_config, \r\n fine_acoustics_config, \r\n )\r\n# model = BarkModel.from_pretrained(MODEL_NAME, torch_dtype=torch.float16).to(device)\r\nmodel = BarkModel(config = configuration).to(device)\r\nmodel.generation_config = gen_configuration\r\ninputs = processor(sentences,voice_preset=voice_preset).to(device)\r\n\r\noutput = model.generate(**inputs)`\r\n\r\n**Error:\r\nTypeError: transformers.models.bark.generation_configuration_bark.BarkSemanticGenerationConfig() argument after ** must be a mapping, not BarkSemanticGenerationConfig**\r\n\r\nAs for passing it directly to generate, I think I wasn't setting it low enough. It seems to be dropping it, but little static is keeping the clips from shortening. So it probably was working and I just need to play around with temps.", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28327/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "yuanwu2017", "id": 34643241, "node_id": "MDQ6VXNlcjM0NjQzMjQx", "avatar_url": "https://avatars.githubusercontent.com/u/34643241?v=4", "gravatar_id": "", "url": "https://api.github.com/users/yuanwu2017", "html_url": "https://github.com/yuanwu2017", "followers_url": "https://api.github.com/users/yuanwu2017/followers", "following_url": "https://api.github.com/users/yuanwu2017/following{/other_user}", "gists_url": "https://api.github.com/users/yuanwu2017/gists{/gist_id}", "starred_url": "https://api.github.com/users/yuanwu2017/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/yuanwu2017/subscriptions", "organizations_url": "https://api.github.com/users/yuanwu2017/orgs", "repos_url": "https://api.github.com/users/yuanwu2017/repos", "events_url": "https://api.github.com/users/yuanwu2017/events{/privacy}", "received_events_url": "https://api.github.com/users/yuanwu2017/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hey! Feel free to ping me for a review when this is ready! ", "Hi @ArthurZucker,\r\nI tested the failed run_tests case in my local machine, but I got following warning. I think the case failure is not caused by my patch. Do you know how to run the case correctly? Thanks a lot.\r\n\r\n```\r\n(upstream) [yuanwu@skyocean transformers]$ pytest tests/test_modeling_flax_common.py\r\n============================================================================================ test session starts =============================================================================================\r\nplatform linux -- Python 3.9.18, pytest-7.4.0, pluggy-1.0.0\r\nrootdir: /mnt/disk4/yuanwu/workspace/transformers\r\nconfigfile: pyproject.toml\r\nplugins: anyio-3.5.0, xdist-3.5.0, timeout-2.2.0, hypothesis-6.92.2, dash-2.14.2\r\ncollected 0 items\r\n\r\n============================================================================================== warnings summary ==============================================================================================\r\n../../../../../home/yuanwu/.conda/envs/upstream/lib/python3.9/site-packages/_pytest/config/__init__.py:1373\r\n /home/yuanwu/.conda/envs/upstream/lib/python3.9/site-packages/_pytest/config/__init__.py:1373: PytestConfigWarning: Unknown config option: doctest_glob\r\n\r\n self._warn_or_fail_if_strict(f\"Unknown config option: {key}\\n\")\r\n\r\n-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html\r\n============================================================================================= 1 warning in 0.38s =============================================================================================\r\n```", "@ArthurZucker I think the patch is ready. Please review. Thanks.", "@ArthurZucker Please help to review.", "Yes sorry for the delay! ", "Hi @ArthurZucker \r\nDone for raising error when device is not available.", "Thanks @ArthurZucker \r\nDone.", "Thanks, let's make sure the CI is green!", "@yuanwu2017 There was a fix merged into main which resolves the currently failing tests - could you rebase to include these and trigger an new CI run? " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28326/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28326", "html_url": "https://github.com/huggingface/transformers/pull/28326", "diff_url": "https://github.com/huggingface/transformers/pull/28326.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28326.patch", "merged_at": 1705333152000 }

{ "login": "Apsod", "id": 5305850, "node_id": "MDQ6VXNlcjUzMDU4NTA=", "avatar_url": "https://avatars.githubusercontent.com/u/5305850?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Apsod", "html_url": "https://github.com/Apsod", "followers_url": "https://api.github.com/users/Apsod/followers", "following_url": "https://api.github.com/users/Apsod/following{/other_user}", "gists_url": "https://api.github.com/users/Apsod/gists{/gist_id}", "starred_url": "https://api.github.com/users/Apsod/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Apsod/subscriptions", "organizations_url": "https://api.github.com/users/Apsod/orgs", "repos_url": "https://api.github.com/users/Apsod/repos", "events_url": "https://api.github.com/users/Apsod/events{/privacy}", "received_events_url": "https://api.github.com/users/Apsod/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Thanks for the quick reply! \r\nI ran the following: \r\n\r\n```\r\nRUN_SLOW=yes python -m pytest -n auto --dist=loadfile -s -v ./tests/models/gpt_sw3/test_tokenization_gpt_sw3.py\r\n```\r\nIt did fail a test, namely an equality test where the reference output has token_type_ids.\r\n\r\nIt also (catastrophically) failed a test that referenced a closed model. \r\n\r\nI've removed the offending token_type_ids in the equality check and updated the model reference to an openly available model. It now passes all tests. ", "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/transformers/pr_28325). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28325/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28325", "html_url": "https://github.com/huggingface/transformers/pull/28325", "diff_url": "https://github.com/huggingface/transformers/pull/28325.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28325.patch", "merged_at": 1704306367000 }

{ "login": "kitkhai", "id": 71968397, "node_id": "MDQ6VXNlcjcxOTY4Mzk3", "avatar_url": "https://avatars.githubusercontent.com/u/71968397?v=4", "gravatar_id": "", "url": "https://api.github.com/users/kitkhai", "html_url": "https://github.com/kitkhai", "followers_url": "https://api.github.com/users/kitkhai/followers", "following_url": "https://api.github.com/users/kitkhai/following{/other_user}", "gists_url": "https://api.github.com/users/kitkhai/gists{/gist_id}", "starred_url": "https://api.github.com/users/kitkhai/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/kitkhai/subscriptions", "organizations_url": "https://api.github.com/users/kitkhai/orgs", "repos_url": "https://api.github.com/users/kitkhai/repos", "events_url": "https://api.github.com/users/kitkhai/events{/privacy}", "received_events_url": "https://api.github.com/users/kitkhai/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Additionally, is there a way to retrieve (and edit) the merge rules from \"slow\" & \"fast\" tokenizers respectively?", "Hey! Few things here. What you are trying to do is outside the scope of the supported features. Adding a token should be done using `tokenizer.add_tokens` function. \r\nThe fast version is for me more right than what you expect. If there are no `merges`, then there is absolutely no reason for the BPE model to fuse `'▁super', 'long', 'word'` into `superlongword`. Thus the slow version seems more wrong, and specifically because sentencepiece does not really allow adding tokens that way. " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28324/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "andysingal", "id": 20493493, "node_id": "MDQ6VXNlcjIwNDkzNDkz", "avatar_url": "https://avatars.githubusercontent.com/u/20493493?v=4", "gravatar_id": "", "url": "https://api.github.com/users/andysingal", "html_url": "https://github.com/andysingal", "followers_url": "https://api.github.com/users/andysingal/followers", "following_url": "https://api.github.com/users/andysingal/following{/other_user}", "gists_url": "https://api.github.com/users/andysingal/gists{/gist_id}", "starred_url": "https://api.github.com/users/andysingal/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/andysingal/subscriptions", "organizations_url": "https://api.github.com/users/andysingal/orgs", "repos_url": "https://api.github.com/users/andysingal/repos", "events_url": "https://api.github.com/users/andysingal/events{/privacy}", "received_events_url": "https://api.github.com/users/andysingal/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hey 🤗 thanks for opening an issue! We try to keep the github issues for bugs/feature requests.\r\nCould you ask your question on the [dataset's community tab](https://huggingface.co./datasets/mehul7/captioned_military_aircraft/discussions) instead?\r\n\r\nThanks!", "Can you please forward the issue there?\r\n\r\nOn Wed, Jan 3, 2024 at 20:48 Arthur ***@***.***> wrote:\r\n\r\n> Hey 🤗 thanks for opening an issue! We try to keep the github issues for\r\n> bugs/feature requests.\r\n> Could you ask your question on the dataset's community tab\r\n> <https://huggingface.co./datasets/mehul7/captioned_military_aircraft/discussions>\r\n> instead?\r\n>\r\n> Thanks!\r\n>\r\n> —\r\n> Reply to this email directly, view it on GitHub\r\n> <https://github.com/huggingface/transformers/issues/28323#issuecomment-1875535846>,\r\n> or unsubscribe\r\n> <https://github.com/notifications/unsubscribe-auth/AE4LJNKTATTEDLWOUBRDLJ3YMVZDPAVCNFSM6AAAAABBLEWXN6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNZVGUZTKOBUGY>\r\n> .\r\n> You are receiving this because you authored the thread.Message ID:\r\n> ***@***.***>\r\n>\r\n", "You found the bug, feel free to do so! 🤗 ", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28323/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "Facico", "id": 56598258, "node_id": "MDQ6VXNlcjU2NTk4MjU4", "avatar_url": "https://avatars.githubusercontent.com/u/56598258?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Facico", "html_url": "https://github.com/Facico", "followers_url": "https://api.github.com/users/Facico/followers", "following_url": "https://api.github.com/users/Facico/following{/other_user}", "gists_url": "https://api.github.com/users/Facico/gists{/gist_id}", "starred_url": "https://api.github.com/users/Facico/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Facico/subscriptions", "organizations_url": "https://api.github.com/users/Facico/orgs", "repos_url": "https://api.github.com/users/Facico/repos", "events_url": "https://api.github.com/users/Facico/events{/privacy}", "received_events_url": "https://api.github.com/users/Facico/received_events", "type": "User", "site_admin": false }

[]

[]

[ "cc @younesbelkada ", "@younesbelkada Thanks for the suggestion, I've pushed the changes!", "@Facico can you confirm the latest commit on this branch fixes your issue?", "Yes, They all work. @younesbelkada ", "@younesbelkada How do I pass the tests_exotic_models?", "Rebasing on main should help! ", "Yes @Facico please merge your branch with upstream main and the test should be fixed. Afrer that we'll be able to merge your PR", "@younesbelkada @ArthurZucker Thanks! That helped.", "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/transformers/pr_28321). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28321/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28321", "html_url": "https://github.com/huggingface/transformers/pull/28321", "diff_url": "https://github.com/huggingface/transformers/pull/28321.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28321.patch", "merged_at": 1706267101000 }

{ "login": "fabianlim", "id": 8325951, "node_id": "MDQ6VXNlcjgzMjU5NTE=", "avatar_url": "https://avatars.githubusercontent.com/u/8325951?v=4", "gravatar_id": "", "url": "https://api.github.com/users/fabianlim", "html_url": "https://github.com/fabianlim", "followers_url": "https://api.github.com/users/fabianlim/followers", "following_url": "https://api.github.com/users/fabianlim/following{/other_user}", "gists_url": "https://api.github.com/users/fabianlim/gists{/gist_id}", "starred_url": "https://api.github.com/users/fabianlim/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/fabianlim/subscriptions", "organizations_url": "https://api.github.com/users/fabianlim/orgs", "repos_url": "https://api.github.com/users/fabianlim/repos", "events_url": "https://api.github.com/users/fabianlim/events{/privacy}", "received_events_url": "https://api.github.com/users/fabianlim/received_events", "type": "User", "site_admin": false }

[]

[]

[ "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored.", "Hello @fabianlim, I think the PR https://github.com/huggingface/transformers/pull/28297 should resolve this. ", "@pacman100 yes I think so too, closing this issue. " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28320/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "kitkhai", "id": 71968397, "node_id": "MDQ6VXNlcjcxOTY4Mzk3", "avatar_url": "https://avatars.githubusercontent.com/u/71968397?v=4", "gravatar_id": "", "url": "https://api.github.com/users/kitkhai", "html_url": "https://github.com/kitkhai", "followers_url": "https://api.github.com/users/kitkhai/followers", "following_url": "https://api.github.com/users/kitkhai/following{/other_user}", "gists_url": "https://api.github.com/users/kitkhai/gists{/gist_id}", "starred_url": "https://api.github.com/users/kitkhai/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/kitkhai/subscriptions", "organizations_url": "https://api.github.com/users/kitkhai/orgs", "repos_url": "https://api.github.com/users/kitkhai/repos", "events_url": "https://api.github.com/users/kitkhai/events{/privacy}", "received_events_url": "https://api.github.com/users/kitkhai/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hey! Thanks for raising this issue. Do you want to open a PR for a fix? 🤗 ", "Hey! Thanks for the swift response.\r\n\r\nI'm still rather new and am actually confused about what the \"slow\" & \"fast\" Tokenizers are based on... (I don't really understand why one is based on \"sentencepiece\" and the other is based on \"BPE\"? Aren't they supposed to be trained the same way just implemented in different programming languages?)\r\n\r\nHence, I don't think I am in a position to open a PR for a fix? Sorry about that 😬", "Your intuition is correct, the slow version uses the `sentencepiece` backend, so the BPE implementation of `sentencepiece` library. While the fast uses the `tokenizers` backend, with the `BPE` implementation of `tokenizers` that is based on `sentencepiece`! ", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28322/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "whitejeep600", "id": 73194181, "node_id": "MDQ6VXNlcjczMTk0MTgx", "avatar_url": "https://avatars.githubusercontent.com/u/73194181?v=4", "gravatar_id": "", "url": "https://api.github.com/users/whitejeep600", "html_url": "https://github.com/whitejeep600", "followers_url": "https://api.github.com/users/whitejeep600/followers", "following_url": "https://api.github.com/users/whitejeep600/following{/other_user}", "gists_url": "https://api.github.com/users/whitejeep600/gists{/gist_id}", "starred_url": "https://api.github.com/users/whitejeep600/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/whitejeep600/subscriptions", "organizations_url": "https://api.github.com/users/whitejeep600/orgs", "repos_url": "https://api.github.com/users/whitejeep600/repos", "events_url": "https://api.github.com/users/whitejeep600/events{/privacy}", "received_events_url": "https://api.github.com/users/whitejeep600/received_events", "type": "User", "site_admin": false }

[ { "id": 2648621985, "node_id": "MDU6TGFiZWwyNjQ4NjIxOTg1", "url": "https://api.github.com/repos/huggingface/transformers/labels/Feature%20request", "name": "Feature request", "color": "FBCA04", "default": false, "description": "Request for a new feature" } ]

[]

[ "The reason is that generate is for inference only, training requires a custom sampling logic ", "> The reason is that generate is for inference only, training requires a custom sampling logic\r\n\r\nThank you for the quick reply.\r\n\r\nHowever, what if I would just like to use the existing sampling algorithms from the generate function? It's a valid training strategy. It would be convenient to have access to the ready-made code for this purpose." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28319/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "shivance", "id": 51750587, "node_id": "MDQ6VXNlcjUxNzUwNTg3", "avatar_url": "https://avatars.githubusercontent.com/u/51750587?v=4", "gravatar_id": "", "url": "https://api.github.com/users/shivance", "html_url": "https://github.com/shivance", "followers_url": "https://api.github.com/users/shivance/followers", "following_url": "https://api.github.com/users/shivance/following{/other_user}", "gists_url": "https://api.github.com/users/shivance/gists{/gist_id}", "starred_url": "https://api.github.com/users/shivance/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/shivance/subscriptions", "organizations_url": "https://api.github.com/users/shivance/orgs", "repos_url": "https://api.github.com/users/shivance/repos", "events_url": "https://api.github.com/users/shivance/events{/privacy}", "received_events_url": "https://api.github.com/users/shivance/received_events", "type": "User", "site_admin": false }

[]

[]

[ "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28318/reactions", "total_count": 1, "+1": 0, "-1": 0, "laugh": 0, "hooray": 1, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28318", "html_url": "https://github.com/huggingface/transformers/pull/28318", "diff_url": "https://github.com/huggingface/transformers/pull/28318.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28318.patch", "merged_at": null }

{ "login": "Adam1679", "id": 32404962, "node_id": "MDQ6VXNlcjMyNDA0OTYy", "avatar_url": "https://avatars.githubusercontent.com/u/32404962?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Adam1679", "html_url": "https://github.com/Adam1679", "followers_url": "https://api.github.com/users/Adam1679/followers", "following_url": "https://api.github.com/users/Adam1679/following{/other_user}", "gists_url": "https://api.github.com/users/Adam1679/gists{/gist_id}", "starred_url": "https://api.github.com/users/Adam1679/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Adam1679/subscriptions", "organizations_url": "https://api.github.com/users/Adam1679/orgs", "repos_url": "https://api.github.com/users/Adam1679/repos", "events_url": "https://api.github.com/users/Adam1679/events{/privacy}", "received_events_url": "https://api.github.com/users/Adam1679/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Makes sense, would you like to open a PR for the fix? 🤗 ", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28317/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "vwxyzjn", "id": 5555347, "node_id": "MDQ6VXNlcjU1NTUzNDc=", "avatar_url": "https://avatars.githubusercontent.com/u/5555347?v=4", "gravatar_id": "", "url": "https://api.github.com/users/vwxyzjn", "html_url": "https://github.com/vwxyzjn", "followers_url": "https://api.github.com/users/vwxyzjn/followers", "following_url": "https://api.github.com/users/vwxyzjn/following{/other_user}", "gists_url": "https://api.github.com/users/vwxyzjn/gists{/gist_id}", "starred_url": "https://api.github.com/users/vwxyzjn/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/vwxyzjn/subscriptions", "organizations_url": "https://api.github.com/users/vwxyzjn/orgs", "repos_url": "https://api.github.com/users/vwxyzjn/repos", "events_url": "https://api.github.com/users/vwxyzjn/events{/privacy}", "received_events_url": "https://api.github.com/users/vwxyzjn/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Sorry but the source of the regression might be pretty much anything. If the model supports SDPA, it can come from SDPA, if the tokenizer had a bug before, it might be the tokenizer etc etc\r\nI can't debug this as is, would you mind comparing with closer transformers releases? This might help isolating this but otherwise the scope is just way too broad. The modeling code has / might have changed, the caching mechanism has changed, torch operators might have been fix etc etc ", "Thanks for the reply! I compared the following transformers releases and noticed that since 4.36.0, the losses become different. I also validated end-to-end that 4.33.2 is fine. \r\n\r\n<img width=\"1048\" alt=\"image\" src=\"https://github.com/huggingface/transformers/assets/5555347/1bf51581-3f9d-4730-a9a4-3e58a283c9c4\">\r\n\r\n\r\n\r\n```\r\npython repro.py # 4.30.1\r\nepoch: 0\r\nupdate: 9, loss: 0.6904680132865906\r\nupdate: 17, loss: 0.6958459615707397\r\nupdate: 25, loss: 0.6878675818443298\r\nupdate: 33, loss: 0.6945885419845581\r\nupdate: 41, loss: 0.6920362710952759\r\nupdate: 49, loss: 0.6866860389709473\r\nupdate: 57, loss: 0.685932457447052\r\nupdate: 65, loss: 0.6930047273635864\r\nupdate: 73, loss: 0.6854068636894226\r\nupdate: 81, loss: 0.6739884614944458\r\nupdate: 89, loss: 0.6913299560546875\r\nupdate: 97, loss: 0.7025052309036255\r\n\r\n\r\n#4.33.2\r\nepoch: 0\r\nupdate: 9, loss: 0.6904680132865906\r\nupdate: 17, loss: 0.6958459615707397\r\nupdate: 25, loss: 0.6878675818443298\r\nupdate: 33, loss: 0.6945885419845581\r\nupdate: 41, loss: 0.6920362710952759\r\nupdate: 49, loss: 0.6866860389709473\r\nupdate: 57, loss: 0.685932457447052\r\nupdate: 65, loss: 0.6930047273635864\r\nupdate: 73, loss: 0.6854068636894226\r\nupdate: 81, loss: 0.6739884614944458\r\n\r\n# 4.35.1 \r\n===training model===\r\nepoch: 0\r\nupdate: 9, loss: 0.6904680132865906\r\nupdate: 17, loss: 0.6958459615707397\r\nupdate: 25, loss: 0.6878675818443298\r\nupdate: 33, loss: 0.6945885419845581\r\nupdate: 41, loss: 0.6920362710952759\r\nupdate: 49, loss: 0.6866860389709473\r\nupdate: 57, loss: 0.685932457447052\r\nupdate: 65, loss: 0.6930047273635864\r\nupdate: 73, loss: 0.6854068636894226\r\nupdate: 81, loss: 0.6739884614944458\r\nupdate: 89, loss: 0.6913299560546875\r\n\r\n\r\n# 4.36.0\r\n===training model===\r\nepoch: 0\r\nupdate: 9, loss: 0.6855486035346985\r\nupdate: 17, loss: 0.6901922225952148\r\nupdate: 25, loss: 0.6883461475372314\r\nupdate: 33, loss: 0.6975809931755066\r\nupdate: 41, loss: 0.6995139122009277\r\nupdate: 49, loss: 0.6912401914596558\r\nupdate: 57, loss: 0.698995053768158\r\nupdate: 65, loss: 0.7005056142807007\r\nupdate: 73, loss: 0.7048475742340088\r\nupdate: 81, loss: 0.6950501203536987\r\nupdate: 89, loss: 0.7148610949516296\r\n\r\n# 4.36.1\r\nepoch: 0\r\nupdate: 9, loss: 0.6855486035346985\r\nupdate: 17, loss: 0.6901922225952148\r\nupdate: 25, loss: 0.6883461475372314\r\nupdate: 33, loss: 0.6975809931755066\r\nupdate: 41, loss: 0.6995139122009277\r\nupdate: 49, loss: 0.6912401914596558\r\nupdate: 57, loss: 0.698995053768158\r\nupdate: 65, loss: 0.7005056142807007\r\nupdate: 73, loss: 0.7048475742340088\r\nupdate: 81, loss: 0.6950501203536987\r\nupdate: 89, loss: 0.7148610949516296\r\n```", "Could you try using `attn_implementation = \"eager\"` instead of `sdpa` wherever you instantiate a model? One of the biggest changes from 4.36 is this! See [here](https://github.com/huggingface/transformers/issues/28005) ", "Also the number you have don't really seem alarming no? ", "I ran it with \r\n\r\n```\r\n self.lm_backbone = AutoModel.from_pretrained(\r\n config.base_model,\r\n config=self.config.base_config,\r\n trust_remote_code=True,\r\n attn_implementation=\"eager\",\r\n )\r\n```\r\n\r\nand it did not seem to make a difference. \r\n\r\n> Also the number you have don't really seem alarming no?\r\n\r\nYeah, but I guess this is why it's tricky — the numbers do not look that different but it causes a significant regression for reward model training. Maybe the hidden states index are being messed up somehow? It's using `self.scalar_head(output.hidden_states[-1])`.\r\n\r\n", "Oh sorry if using output_hidden_states, `eager` will by default be used. \r\nI have no idea, pinging @pacman100 our training expert for idea and @younesbelkada for SFT training which should be more relevant expertise than me!", "Hi @vwxyzjn !\r\nHappy new year! \r\nHmm this is interesting, I don't have a clear idea either on what could be causing this, but looking at the commit history of GPTNeoX modeling code it could be :\r\n1- Attention dropout support: https://github.com/huggingface/transformers/commit/392740452e86ee6ca523b92bc4ef8527ed4e7a16\r\n2- RoPE scaling: https://github.com/huggingface/transformers/pull/24653\r\n3- Potentially the Gradient checkpointing refactor as well https://github.com/huggingface/transformers/pull/27020\r\nIf the experiments are not too long to be ran, can you try to checkout on each of these commits and see which one might be responsible of the regression?", "I think it's https://github.com/huggingface/transformers/commit/253f9a3f9716d08a81fb305fe71f983122eb608b i'll fix the nans!" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28316/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "sam99dave", "id": 37779169, "node_id": "MDQ6VXNlcjM3Nzc5MTY5", "avatar_url": "https://avatars.githubusercontent.com/u/37779169?v=4", "gravatar_id": "", "url": "https://api.github.com/users/sam99dave", "html_url": "https://github.com/sam99dave", "followers_url": "https://api.github.com/users/sam99dave/followers", "following_url": "https://api.github.com/users/sam99dave/following{/other_user}", "gists_url": "https://api.github.com/users/sam99dave/gists{/gist_id}", "starred_url": "https://api.github.com/users/sam99dave/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/sam99dave/subscriptions", "organizations_url": "https://api.github.com/users/sam99dave/orgs", "repos_url": "https://api.github.com/users/sam99dave/repos", "events_url": "https://api.github.com/users/sam99dave/events{/privacy}", "received_events_url": "https://api.github.com/users/sam99dave/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hey @sam99dave were you waiting for a review? ", "Yeah, I was waiting for one. Not sure if these required changes has already been merged by another PR. There was another PR for this which was being reviewed I think.\n\nPls, let me know on this.", "Seems like #28312 was already approved, but is getting stale. I'll post a message there" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28315/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28315", "html_url": "https://github.com/huggingface/transformers/pull/28315", "diff_url": "https://github.com/huggingface/transformers/pull/28315.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28315.patch", "merged_at": null }

{ "login": "DrChrisLevy", "id": 16509365, "node_id": "MDQ6VXNlcjE2NTA5MzY1", "avatar_url": "https://avatars.githubusercontent.com/u/16509365?v=4", "gravatar_id": "", "url": "https://api.github.com/users/DrChrisLevy", "html_url": "https://github.com/DrChrisLevy", "followers_url": "https://api.github.com/users/DrChrisLevy/followers", "following_url": "https://api.github.com/users/DrChrisLevy/following{/other_user}", "gists_url": "https://api.github.com/users/DrChrisLevy/gists{/gist_id}", "starred_url": "https://api.github.com/users/DrChrisLevy/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/DrChrisLevy/subscriptions", "organizations_url": "https://api.github.com/users/DrChrisLevy/orgs", "repos_url": "https://api.github.com/users/DrChrisLevy/repos", "events_url": "https://api.github.com/users/DrChrisLevy/events{/privacy}", "received_events_url": "https://api.github.com/users/DrChrisLevy/received_events", "type": "User", "site_admin": false }

[]

[]

[ "It seems like OpenBlas is at fault here no? I ran your code and had no warning whatsoever (on a M3). https://github.com/OpenMathLib/OpenBLAS might need to be built with USE_OPENMP=1.\r\n", "Okay let me run it in a python virtual env (not docker) on the same machine and see if the issue goes away. You ran it in the above docker container ? Or in a slightly different env? Thanks for the quick reply and checking. I just thought it was something to do with transformers since I only saw that warning when upgrading recently. I'll post back later today. \r\n\r\n**update** (I also updated some of the findings in the description at the top).\r\n- The issue **is** present in the sample docker container on macm1 with transformers versions higher than 4.35.2. It does not happen in 4.35.2 though in the same Docker container.\r\n- The issue is **not** present when running within python venv (not Docker) on mac m1, even for transformers versions higher than 4.35.2.\r\n\r\nIm not sure if you have any other thoughts on that @ArthurZucker . I would be interested in knowing if you ran my reproducible example in the same docker container environment. \r\n\r\nIf you think this is just some environment thing then feel free to close the issue.\r\n", "Sorry for the confusion I simply ran it on pyenv locally as well 😉 \r\nGlad that the findings show this might be more a docker issue than a transformers issue! ", "It should be possible to disable multithreading in openblas without recompiling it, by `export OPENBLAS_NUM_THREADS=1` before running the python program in the terminal https://pythonspeed.com/articles/concurrency-control/", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28314/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "kevherro", "id": 10460086, "node_id": "MDQ6VXNlcjEwNDYwMDg2", "avatar_url": "https://avatars.githubusercontent.com/u/10460086?v=4", "gravatar_id": "", "url": "https://api.github.com/users/kevherro", "html_url": "https://github.com/kevherro", "followers_url": "https://api.github.com/users/kevherro/followers", "following_url": "https://api.github.com/users/kevherro/following{/other_user}", "gists_url": "https://api.github.com/users/kevherro/gists{/gist_id}", "starred_url": "https://api.github.com/users/kevherro/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/kevherro/subscriptions", "organizations_url": "https://api.github.com/users/kevherro/orgs", "repos_url": "https://api.github.com/users/kevherro/repos", "events_url": "https://api.github.com/users/kevherro/events{/privacy}", "received_events_url": "https://api.github.com/users/kevherro/received_events", "type": "User", "site_admin": false }

[]

[]

[ "@ArthurZucker I appreciate your review!\r\n\r\n> we should probably link @LysandreJik comment from the issue\r\n\r\n\"In the future we'll likely remove the huggingface channel support for conda unless we see explicit demand here; if that's the case, please put a thumbs up here (we'll monitor), otherwise, the conda-forge channel is usually very up to date.\" ([here](https://github.com/huggingface/transformers/issues/28248#issuecomment-1869765891))\r\n\r\n> mention that the huggingface channel is deprecated for transformers\r\n\r\nGood callout. Should this go in the README as well?\r\n", "If you're up for it, would you also mind updating this in the [Installation](https://huggingface.co./docs/transformers/installation#install-with-conda) docs? 🙂 ", "@stevhliu I think I did this correctly! Note that this also cleans up some whitespaces and newlines in the changed files.", "My pleasure! ", "Just seeing this now, but thanks for the PR @kevherro!" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28313/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28313", "html_url": "https://github.com/huggingface/transformers/pull/28313", "diff_url": "https://github.com/huggingface/transformers/pull/28313.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28313.patch", "merged_at": 1704389776000 }

{ "login": "Tanmaypatil123", "id": 77950208, "node_id": "MDQ6VXNlcjc3OTUwMjA4", "avatar_url": "https://avatars.githubusercontent.com/u/77950208?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Tanmaypatil123", "html_url": "https://github.com/Tanmaypatil123", "followers_url": "https://api.github.com/users/Tanmaypatil123/followers", "following_url": "https://api.github.com/users/Tanmaypatil123/following{/other_user}", "gists_url": "https://api.github.com/users/Tanmaypatil123/gists{/gist_id}", "starred_url": "https://api.github.com/users/Tanmaypatil123/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Tanmaypatil123/subscriptions", "organizations_url": "https://api.github.com/users/Tanmaypatil123/orgs", "repos_url": "https://api.github.com/users/Tanmaypatil123/repos", "events_url": "https://api.github.com/users/Tanmaypatil123/events{/privacy}", "received_events_url": "https://api.github.com/users/Tanmaypatil123/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hi, thanks for your PR. It looks like there's an issue with the segmentation models.", "@NielsRogge could you take a look at this PR again . I have made required changes for segmentation models.", "Thanks, LGTM.", "hey @amyeroberts removed conflicts .but whenever I tried to run test cases by `RUN_SLOW=1 pytest tests/models/conditional_detr/` I am getting following error : \r\n\r\n", "@Tanmaypatil123 - Not sure exactly what's causing this. The function `is_g2p_en_available` was added in #23439. However, we haven't seen this import error in our CI runs which validate slow tests and I'm able to run `RUN_SLOW=1 pytest tests/models/conditional_detr/` locally on `main`.\r\n\r\nIn a python session, are you about to run: \r\n```\r\nfrom transformers.utils import is_g2p_en_available\r\n```\r\n? \r\n\r\nCould you try rebasing on main to make sure all updates/fixes are included in this branch? ", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored.", "The issue is related to evaluate I think: https://github.com/huggingface/peft/issues/1351 ", "cc @Tanmaypatil123 let's rebase this and I think we could merge this! ", "@ArthurZucker Changes are approved but there is issue with segmentation model; test cases fails for it and i could not figure out what the problem is. ", "@Tanmaypatil123 you just need to fix the upstream conflicts, like so:\r\n```\r\ngit remote add upstream https://github.com/huggingface/transformers.git\r\ngit fetch upstream\r\ngit merge upstream/main\r\n```", "@ArthurZucker I have resolved the conflicts ", "@Tanmaypatil123 Great! \r\n\r\nFor the failing tests, there's two which look like they might be related to this PR which will needed to be addressed before merge: \r\n\r\n```\r\nFAILED tests/models/yolos/test_modeling_yolos.py::YolosModelTest::test_pipeline_image_feature_extraction - TypeError: forward() got an unexpected keyword argument 'pixel_mask'\r\nFAILED tests/models/yolos/test_modeling_yolos.py::YolosModelTest::test_pipeline_object_detection - TypeError: forward() got an unexpected keyword argument 'pixel_mask'\r\n```\r\n\r\nFor the build documentation tests, there was a recent fix merged into main. Rebasing to include this should resolve. ", "> @Tanmaypatil123 Great!\r\n> \r\n> For the failing tests, there's two which look like they might be related to this PR which will needed to be addressed before merge:\r\n> \r\n> ```\r\n> FAILED tests/models/yolos/test_modeling_yolos.py::YolosModelTest::test_pipeline_image_feature_extraction - TypeError: forward() got an unexpected keyword argument 'pixel_mask'\r\n> FAILED tests/models/yolos/test_modeling_yolos.py::YolosModelTest::test_pipeline_object_detection - TypeError: forward() got an unexpected keyword argument 'pixel_mask'\r\n> ```\r\n> \r\n> For the build documentation tests, there was a recent fix merged into main. Rebasing to include this should resolve.\r\n\r\nDon't know how that test case is failing. I didn't make any changes to the parameters that are provided to `YolosForObjectDetection`. should we add parameters or is there any change in the test case?", "@Tanmaypatil123 Indeed. It's quite weird these tests are failing as they don't seem related to this PR. They aren't however failing on any other PRs. What I think is happening is that some tests weren't collected by the test fetched (cc @ydshieh for reference) when the image feature extraction pipeline was added. And now, because this PR touches yolos they're being run now. I'm going to look into it and let you know asap. ", "@amyeroberts Done 😀 Thanks for the help. Huggingface has the best open source maintainers.", "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/transformers/pr_28312). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "@Tanmaypatil123 Thanks for this great contribution and patience with our failing CI tests! " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28312/reactions", "total_count": 2, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 2, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28312", "html_url": "https://github.com/huggingface/transformers/pull/28312", "diff_url": "https://github.com/huggingface/transformers/pull/28312.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28312.patch", "merged_at": 1708083539000 }

{ "login": "dependabot[bot]", "id": 49699333, "node_id": "MDM6Qm90NDk2OTkzMzM=", "avatar_url": "https://avatars.githubusercontent.com/in/29110?v=4", "gravatar_id": "", "url": "https://api.github.com/users/dependabot%5Bbot%5D", "html_url": "https://github.com/apps/dependabot", "followers_url": "https://api.github.com/users/dependabot%5Bbot%5D/followers", "following_url": "https://api.github.com/users/dependabot%5Bbot%5D/following{/other_user}", "gists_url": "https://api.github.com/users/dependabot%5Bbot%5D/gists{/gist_id}", "starred_url": "https://api.github.com/users/dependabot%5Bbot%5D/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/dependabot%5Bbot%5D/subscriptions", "organizations_url": "https://api.github.com/users/dependabot%5Bbot%5D/orgs", "repos_url": "https://api.github.com/users/dependabot%5Bbot%5D/repos", "events_url": "https://api.github.com/users/dependabot%5Bbot%5D/events{/privacy}", "received_events_url": "https://api.github.com/users/dependabot%5Bbot%5D/received_events", "type": "Bot", "site_admin": false }

[ { "id": 1905493434, "node_id": "MDU6TGFiZWwxOTA1NDkzNDM0", "url": "https://api.github.com/repos/huggingface/transformers/labels/dependencies", "name": "dependencies", "color": "0366d6", "default": false, "description": "Pull requests that update a dependency file" }, { "id": 6384750322, "node_id": "LA_kwDOCUB6oc8AAAABfI-O8g", "url": "https://api.github.com/repos/huggingface/transformers/labels/github_actions", "name": "github_actions", "color": "000000", "default": false, "description": "Pull requests that update GitHub Actions code" } ]

[]

[ "fyi @ydshieh " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28311/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28311", "html_url": "https://github.com/huggingface/transformers/pull/28311", "diff_url": "https://github.com/huggingface/transformers/pull/28311.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28311.patch", "merged_at": 1704269574000 }

{ "login": "isRambler", "id": 118053582, "node_id": "U_kgDOBwlazg", "avatar_url": "https://avatars.githubusercontent.com/u/118053582?v=4", "gravatar_id": "", "url": "https://api.github.com/users/isRambler", "html_url": "https://github.com/isRambler", "followers_url": "https://api.github.com/users/isRambler/followers", "following_url": "https://api.github.com/users/isRambler/following{/other_user}", "gists_url": "https://api.github.com/users/isRambler/gists{/gist_id}", "starred_url": "https://api.github.com/users/isRambler/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/isRambler/subscriptions", "organizations_url": "https://api.github.com/users/isRambler/orgs", "repos_url": "https://api.github.com/users/isRambler/repos", "events_url": "https://api.github.com/users/isRambler/events{/privacy}", "received_events_url": "https://api.github.com/users/isRambler/received_events", "type": "User", "site_admin": false }

[]

[]

[ "me too", "Hey! I think that a conversion is attempted between pt and tf, but transformers does not support a TF implementation of this model. Could you rather open an issue on the `TensorRT-LLM` library with a full reproducer of what you are trying to do and the full stacktrace ", "> Hey! I think that a conversion is attempted between pt and tf, but transformers does not support a TF implementation of this model. Could you rather open an issue on the `TensorRT-LLM` library with a full reproducer of what you are trying to do and the full stacktrace\r\n\r\nOkay, I'm going to ask the relevant questions" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28310/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "NielsRogge", "id": 48327001, "node_id": "MDQ6VXNlcjQ4MzI3MDAx", "avatar_url": "https://avatars.githubusercontent.com/u/48327001?v=4", "gravatar_id": "", "url": "https://api.github.com/users/NielsRogge", "html_url": "https://github.com/NielsRogge", "followers_url": "https://api.github.com/users/NielsRogge/followers", "following_url": "https://api.github.com/users/NielsRogge/following{/other_user}", "gists_url": "https://api.github.com/users/NielsRogge/gists{/gist_id}", "starred_url": "https://api.github.com/users/NielsRogge/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/NielsRogge/subscriptions", "organizations_url": "https://api.github.com/users/NielsRogge/orgs", "repos_url": "https://api.github.com/users/NielsRogge/repos", "events_url": "https://api.github.com/users/NielsRogge/events{/privacy}", "received_events_url": "https://api.github.com/users/NielsRogge/received_events", "type": "User", "site_admin": false }

[ { "id": 1990918270, "node_id": "MDU6TGFiZWwxOTkwOTE4Mjcw", "url": "https://api.github.com/repos/huggingface/transformers/labels/Good%20First%20Issue", "name": "Good First Issue", "color": "bbf794", "default": false, "description": "" } ]

[]

[ "Hi @NielsRogge, I would like to take this up", "Hi @NielsRogge is this open for just 1 person taking it up as @sam99dave already indicated or multiple contribution? If it is the latter, I can take one of this up.", "Hi, it looks like the PRs above addressed all models at once, which is fine for me.", "Closing as the PR above was merged." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28309/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "Hambaobao", "id": 48345096, "node_id": "MDQ6VXNlcjQ4MzQ1MDk2", "avatar_url": "https://avatars.githubusercontent.com/u/48345096?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Hambaobao", "html_url": "https://github.com/Hambaobao", "followers_url": "https://api.github.com/users/Hambaobao/followers", "following_url": "https://api.github.com/users/Hambaobao/following{/other_user}", "gists_url": "https://api.github.com/users/Hambaobao/gists{/gist_id}", "starred_url": "https://api.github.com/users/Hambaobao/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Hambaobao/subscriptions", "organizations_url": "https://api.github.com/users/Hambaobao/orgs", "repos_url": "https://api.github.com/users/Hambaobao/repos", "events_url": "https://api.github.com/users/Hambaobao/events{/privacy}", "received_events_url": "https://api.github.com/users/Hambaobao/received_events", "type": "User", "site_admin": false }

[ { "id": 2648621985, "node_id": "MDU6TGFiZWwyNjQ4NjIxOTg1", "url": "https://api.github.com/repos/huggingface/transformers/labels/Feature%20request", "name": "Feature request", "color": "FBCA04", "default": false, "description": "Request for a new feature" } ]

{ "login": "muellerzr", "id": 7831895, "node_id": "MDQ6VXNlcjc4MzE4OTU=", "avatar_url": "https://avatars.githubusercontent.com/u/7831895?v=4", "gravatar_id": "", "url": "https://api.github.com/users/muellerzr", "html_url": "https://github.com/muellerzr", "followers_url": "https://api.github.com/users/muellerzr/followers", "following_url": "https://api.github.com/users/muellerzr/following{/other_user}", "gists_url": "https://api.github.com/users/muellerzr/gists{/gist_id}", "starred_url": "https://api.github.com/users/muellerzr/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/muellerzr/subscriptions", "organizations_url": "https://api.github.com/users/muellerzr/orgs", "repos_url": "https://api.github.com/users/muellerzr/repos", "events_url": "https://api.github.com/users/muellerzr/events{/privacy}", "received_events_url": "https://api.github.com/users/muellerzr/received_events", "type": "User", "site_admin": false }

[ { "login": "muellerzr", "id": 7831895, "node_id": "MDQ6VXNlcjc4MzE4OTU=", "avatar_url": "https://avatars.githubusercontent.com/u/7831895?v=4", "gravatar_id": "", "url": "https://api.github.com/users/muellerzr", "html_url": "https://github.com/muellerzr", "followers_url": "https://api.github.com/users/muellerzr/followers", "following_url": "https://api.github.com/users/muellerzr/following{/other_user}", "gists_url": "https://api.github.com/users/muellerzr/gists{/gist_id}", "starred_url": "https://api.github.com/users/muellerzr/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/muellerzr/subscriptions", "organizations_url": "https://api.github.com/users/muellerzr/orgs", "repos_url": "https://api.github.com/users/muellerzr/repos", "events_url": "https://api.github.com/users/muellerzr/events{/privacy}", "received_events_url": "https://api.github.com/users/muellerzr/received_events", "type": "User", "site_admin": false } ]

[ "fyi @pacman100 and @muellerzr 🤗 " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28308/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "ILG2021", "id": 93691919, "node_id": "U_kgDOBZWgDw", "avatar_url": "https://avatars.githubusercontent.com/u/93691919?v=4", "gravatar_id": "", "url": "https://api.github.com/users/ILG2021", "html_url": "https://github.com/ILG2021", "followers_url": "https://api.github.com/users/ILG2021/followers", "following_url": "https://api.github.com/users/ILG2021/following{/other_user}", "gists_url": "https://api.github.com/users/ILG2021/gists{/gist_id}", "starred_url": "https://api.github.com/users/ILG2021/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/ILG2021/subscriptions", "organizations_url": "https://api.github.com/users/ILG2021/orgs", "repos_url": "https://api.github.com/users/ILG2021/repos", "events_url": "https://api.github.com/users/ILG2021/events{/privacy}", "received_events_url": "https://api.github.com/users/ILG2021/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hey, could you share the training command you used? (a reproducer for training) as well as the full traceback? 🤗 \r\n", "> `Could not locate the best model at whisper-large-v2-fil/checkpoint-2000/pytorch_model.bin, if you are running a distributed training on multiple nodes, you should activate `--save_on_each_node`.`\r\n\r\nI use the finetune whisper notebook in this blog:\r\nhttps://huggingface.co./blog/fine-tune-whisper", "We need the call you used, and the full traceback 😉 ", "I had the same issue, i fixed with instead of this\r\n\r\nmodel.save_pretrained(output_dir)\r\n\r\ni've use\r\n\r\nmodel.save_pretrained(output_dir, safe_serialization=False)\r\n\r\nAnd also on the training arguments, i 've added save_safetensors=False which by default is True\r\n\r\ntraining_args = TrainingArguments(\r\n output_dir=output_dir,\r\n overwrite_output_dir=True,\r\n num_train_epochs=epochs, \r\n per_device_train_batch_size=batch_size,\r\n learning_rate=2e-5, \r\n weight_decay=0.02, \r\n max_grad_norm=1.0,\r\n save_steps=30,\r\n save_total_limit=2,\r\n save_strategy=\"steps\",\r\n warmup_steps=1000, \r\n gradient_accumulation_steps=1,\r\n logging_dir = output_dir + '/logs',\r\n logging_steps=30,\r\n save_safetensors=False\r\n )\r\n\r\n\r\n", "\r\n\r\n\r\n> I had the same issue, i fixed with instead of this\r\n> \r\n> model.save_pretrained(output_dir)\r\n> \r\n> i've use\r\n> \r\n> model.save_pretrained(output_dir, safe_serialization=False)\r\n> \r\n> And also on the training arguments, i 've added save_safetensors=False which by default is True\r\n> \r\n> training_args = TrainingArguments( output_dir=output_dir, overwrite_output_dir=True, num_train_epochs=epochs, per_device_train_batch_size=batch_size, learning_rate=2e-5, weight_decay=0.02, max_grad_norm=1.0, save_steps=30, save_total_limit=2, save_strategy=\"steps\", warmup_steps=1000, gradient_accumulation_steps=1, logging_dir = output_dir + '/logs', logging_steps=30, save_safetensors=False )\r\n\r\nThanks for you sharing. That's the problem." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28307/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "Spico197", "id": 22840952, "node_id": "MDQ6VXNlcjIyODQwOTUy", "avatar_url": "https://avatars.githubusercontent.com/u/22840952?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Spico197", "html_url": "https://github.com/Spico197", "followers_url": "https://api.github.com/users/Spico197/followers", "following_url": "https://api.github.com/users/Spico197/following{/other_user}", "gists_url": "https://api.github.com/users/Spico197/gists{/gist_id}", "starred_url": "https://api.github.com/users/Spico197/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Spico197/subscriptions", "organizations_url": "https://api.github.com/users/Spico197/orgs", "repos_url": "https://api.github.com/users/Spico197/repos", "events_url": "https://api.github.com/users/Spico197/events{/privacy}", "received_events_url": "https://api.github.com/users/Spico197/received_events", "type": "User", "site_admin": false }

[ { "id": 1843244711, "node_id": "MDU6TGFiZWwxODQzMjQ0NzEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/New%20model", "name": "New model", "color": "fbca04", "default": false, "description": "" } ]

[]

[]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28306/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "bhosalems", "id": 10846405, "node_id": "MDQ6VXNlcjEwODQ2NDA1", "avatar_url": "https://avatars.githubusercontent.com/u/10846405?v=4", "gravatar_id": "", "url": "https://api.github.com/users/bhosalems", "html_url": "https://github.com/bhosalems", "followers_url": "https://api.github.com/users/bhosalems/followers", "following_url": "https://api.github.com/users/bhosalems/following{/other_user}", "gists_url": "https://api.github.com/users/bhosalems/gists{/gist_id}", "starred_url": "https://api.github.com/users/bhosalems/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/bhosalems/subscriptions", "organizations_url": "https://api.github.com/users/bhosalems/orgs", "repos_url": "https://api.github.com/users/bhosalems/repos", "events_url": "https://api.github.com/users/bhosalems/events{/privacy}", "received_events_url": "https://api.github.com/users/bhosalems/received_events", "type": "User", "site_admin": false }

[]

[]

[ "@bhosalems you will find them [here](https://github.com/huggingface/transformers/blob/6ea3ee3cd215dfe0b32034299da3f876af0e7c4e/src/transformers/models/clip/modeling_clip.py) so you will probably need this version of the [transformers library](https://github.com/huggingface/transformers/releases/tag/v4.32.0) : v4.32.0\r\njust use \r\n```\r\npip install transformers==4.32.0\r\n```\r\nand everything should go back to normal", "` _make_causal_mask, _expand_mask` are private functions. There is no guaranty that they will be properly imported and are not part of the public api. They were removed as such! \r\nI do not recommend you to go back to previous versions but rather use \r\n\r\nhttps://github.com/huggingface/transformers/blob/b1292bca6923cfbc9cb3f70cb55df57e4e17e630/src/transformers/modeling_attn_mask_utils.py#L21\r\n\r\nand it's utilities ", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28305/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "cnut1648", "id": 37067883, "node_id": "MDQ6VXNlcjM3MDY3ODgz", "avatar_url": "https://avatars.githubusercontent.com/u/37067883?v=4", "gravatar_id": "", "url": "https://api.github.com/users/cnut1648", "html_url": "https://github.com/cnut1648", "followers_url": "https://api.github.com/users/cnut1648/followers", "following_url": "https://api.github.com/users/cnut1648/following{/other_user}", "gists_url": "https://api.github.com/users/cnut1648/gists{/gist_id}", "starred_url": "https://api.github.com/users/cnut1648/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/cnut1648/subscriptions", "organizations_url": "https://api.github.com/users/cnut1648/orgs", "repos_url": "https://api.github.com/users/cnut1648/repos", "events_url": "https://api.github.com/users/cnut1648/events{/privacy}", "received_events_url": "https://api.github.com/users/cnut1648/received_events", "type": "User", "site_admin": false }

[]

[]

[ "BTW similar issue seems to happen on mistral as well. Training on the same data, non-fast tokenizer gives ~1 loss while fast tokenizer gives > 10 loss.", "There are know issues with LlamaFast tokenizer, could you try with this: #26678 \r\n\r\n```python\r\nfrom tokenizers import pre_tokenizers, normalizers\r\nfrom transformers import AutoTokenizer\r\nold_tokenizer = AutoTokenizer.from_pretrained(\"meta-llama/Llama-2-7b-hf\")\r\nold_tokenizer._tokenizer.normalizer = normalizers.Sequence([])\r\nold_tokenizer._tokenizer.pre_tokenizer = pre_tokenizers.Metaspace(\"▁\", True, prepend_scheme = \"first\")\r\n```", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28304/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "MidAtBest", "id": 154687332, "node_id": "U_kgDOCThXZA", "avatar_url": "https://avatars.githubusercontent.com/u/154687332?v=4", "gravatar_id": "", "url": "https://api.github.com/users/MidAtBest", "html_url": "https://github.com/MidAtBest", "followers_url": "https://api.github.com/users/MidAtBest/followers", "following_url": "https://api.github.com/users/MidAtBest/following{/other_user}", "gists_url": "https://api.github.com/users/MidAtBest/gists{/gist_id}", "starred_url": "https://api.github.com/users/MidAtBest/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/MidAtBest/subscriptions", "organizations_url": "https://api.github.com/users/MidAtBest/orgs", "repos_url": "https://api.github.com/users/MidAtBest/repos", "events_url": "https://api.github.com/users/MidAtBest/events{/privacy}", "received_events_url": "https://api.github.com/users/MidAtBest/received_events", "type": "User", "site_admin": false }

[]

[]

[ "> Hi @MidAtBest, thanks for taking care of this, this is a great PR! The logic is quite difficult so congrats on handling it so nicely !\r\n> \r\n> Two general comments:\r\n> \r\n> 1. you should add test for bark modeling code as well !\r\n> 2. it's best to stay with `voice_preset` instead of `voice_presets`.\r\n> \r\n> I'm responding to your question below, I hope that it helps!\r\n> \r\n> Thanks again!\r\n> \r\n> In terms of your question, I believe that we should refactor the logic of the coarse generation quite a bit to make it work.\r\n> \r\n> The key to that is IMO in the following lines:\r\n> \r\n> These ones creates `semantic_output`:\r\n> \r\n> https://github.com/huggingface/transformers/blob/87ae2a4632de9c3090272d5cd37db86c0e03a0a9/src/transformers/models/bark/modeling_bark.py#L1175-L1185\r\n> \r\n> Those 3 lines are located in the generation loop and creates `input_coarse`:\r\n> \r\n> https://github.com/huggingface/transformers/blob/87ae2a4632de9c3090272d5cd37db86c0e03a0a9/src/transformers/models/bark/modeling_bark.py#L1196-L1204\r\n> \r\n> As you can see, the only thing that changes here is `semantic_idx`, the rest is quite static.\r\n> \r\n> So in my opinion, you keep track of the length of each `semantic_history` and merge each `semantic_history` with the corresponding `semantic_output`.\r\n> \r\n> Once arriving in the generation loop, you can do the 2 first lines for each sample (taking into account than each sample has its own `semantic_idx`) then use `torch.rnn.pad_sequence` to pad it to the right length.\r\n\r\nThanks a lot for the reply, yes it took a bit of time for me to understand both the model and the shape manipulations going on. \r\nRegarding the tests and other changes I had omitted them as I was a bit puzzled regarding the question I mentioned in the body of the PR. \r\nSetting the PR as WIP and will try to have something by Wednesday evening.\r\n", "cc @ylacombe would be nice if you can review here!\r\n" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28303/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28303", "html_url": "https://github.com/huggingface/transformers/pull/28303", "diff_url": "https://github.com/huggingface/transformers/pull/28303.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28303.patch", "merged_at": null }

{ "login": "rajveer43", "id": 64583161, "node_id": "MDQ6VXNlcjY0NTgzMTYx", "avatar_url": "https://avatars.githubusercontent.com/u/64583161?v=4", "gravatar_id": "", "url": "https://api.github.com/users/rajveer43", "html_url": "https://github.com/rajveer43", "followers_url": "https://api.github.com/users/rajveer43/followers", "following_url": "https://api.github.com/users/rajveer43/following{/other_user}", "gists_url": "https://api.github.com/users/rajveer43/gists{/gist_id}", "starred_url": "https://api.github.com/users/rajveer43/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/rajveer43/subscriptions", "organizations_url": "https://api.github.com/users/rajveer43/orgs", "repos_url": "https://api.github.com/users/rajveer43/repos", "events_url": "https://api.github.com/users/rajveer43/events{/privacy}", "received_events_url": "https://api.github.com/users/rajveer43/received_events", "type": "User", "site_admin": false }

[]

[]

[ "@stevhliu", "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/transformers/pr_28302). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28302/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28302", "html_url": "https://github.com/huggingface/transformers/pull/28302", "diff_url": "https://github.com/huggingface/transformers/pull/28302.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28302.patch", "merged_at": 1706807755000 }

{ "login": "rajveer43", "id": 64583161, "node_id": "MDQ6VXNlcjY0NTgzMTYx", "avatar_url": "https://avatars.githubusercontent.com/u/64583161?v=4", "gravatar_id": "", "url": "https://api.github.com/users/rajveer43", "html_url": "https://github.com/rajveer43", "followers_url": "https://api.github.com/users/rajveer43/followers", "following_url": "https://api.github.com/users/rajveer43/following{/other_user}", "gists_url": "https://api.github.com/users/rajveer43/gists{/gist_id}", "starred_url": "https://api.github.com/users/rajveer43/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/rajveer43/subscriptions", "organizations_url": "https://api.github.com/users/rajveer43/orgs", "repos_url": "https://api.github.com/users/rajveer43/repos", "events_url": "https://api.github.com/users/rajveer43/events{/privacy}", "received_events_url": "https://api.github.com/users/rajveer43/received_events", "type": "User", "site_admin": false }

[ { "id": 2796628563, "node_id": "MDU6TGFiZWwyNzk2NjI4NTYz", "url": "https://api.github.com/repos/huggingface/transformers/labels/WIP", "name": "WIP", "color": "234C99", "default": false, "description": "Label your PR/Issue with WIP for some long outstanding Issues/PRs that are work in progress" } ]

[]

[ "@stevhliu would you add the more models here in the list so other can look at this issue and collaborate?" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28301/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "daehuikim", "id": 40377750, "node_id": "MDQ6VXNlcjQwMzc3NzUw", "avatar_url": "https://avatars.githubusercontent.com/u/40377750?v=4", "gravatar_id": "", "url": "https://api.github.com/users/daehuikim", "html_url": "https://github.com/daehuikim", "followers_url": "https://api.github.com/users/daehuikim/followers", "following_url": "https://api.github.com/users/daehuikim/following{/other_user}", "gists_url": "https://api.github.com/users/daehuikim/gists{/gist_id}", "starred_url": "https://api.github.com/users/daehuikim/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/daehuikim/subscriptions", "organizations_url": "https://api.github.com/users/daehuikim/orgs", "repos_url": "https://api.github.com/users/daehuikim/repos", "events_url": "https://api.github.com/users/daehuikim/events{/privacy}", "received_events_url": "https://api.github.com/users/daehuikim/received_events", "type": "User", "site_admin": false }

[]

[]

[ "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored.", "Hey 🤗 thanks for opening an issue! We try to keep the github issues for bugs/feature requests.\r\nCould you ask your question on the [forum](https://discuss.huggingface.co/) instead? I'm sure the community will be of help!\r\n\r\nThanks!" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28300/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "avimanyu786", "id": 28894462, "node_id": "MDQ6VXNlcjI4ODk0NDYy", "avatar_url": "https://avatars.githubusercontent.com/u/28894462?v=4", "gravatar_id": "", "url": "https://api.github.com/users/avimanyu786", "html_url": "https://github.com/avimanyu786", "followers_url": "https://api.github.com/users/avimanyu786/followers", "following_url": "https://api.github.com/users/avimanyu786/following{/other_user}", "gists_url": "https://api.github.com/users/avimanyu786/gists{/gist_id}", "starred_url": "https://api.github.com/users/avimanyu786/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/avimanyu786/subscriptions", "organizations_url": "https://api.github.com/users/avimanyu786/orgs", "repos_url": "https://api.github.com/users/avimanyu786/repos", "events_url": "https://api.github.com/users/avimanyu786/events{/privacy}", "received_events_url": "https://api.github.com/users/avimanyu786/received_events", "type": "User", "site_admin": false }

[]

[]

[ "I do agree this is better to remove shell access to this command line. I don't think however this is critical by any means since this is already a CLI utility which we do not leverage directly.", "@michaelbenayoun @amyeroberts \r\n", "@Narsil @michaelbenayoun @amyeroberts Good to know it's not critical. Thanks a lot for your collective feedback and approval!", "Thanks so much for the merge! I must admit, security isn't my primary domain 😅, but this experience piqued my curiosity about the potential security implications in different use cases of the `transformers` library.\r\n\r\nI believe this information could be beneficial for our community, especially for those implementing `transformers` in varied environments. Therefore, I'd like to share a summary of these insights for knowledge sharing purposes only:\r\n\r\nThrough my additional research, including a discussion with GPT4, I've gained a deeper understanding of the potential security implications of using `shell=True` in `subprocess.Popen` within the transformers library. It appears that while this issue is generally not critical for the library's primary use cases, it can become more significant in specific production environments. Particularly in scenarios where the library interacts with user-generated input — such as in web application backends, desktop applications, and cloud-based ML services — the risk of arbitrary code execution increases. This insight highlights the importance of context in evaluating security risks and underscores the need for careful consideration of security best practices, especially in diverse deployment scenarios.\r\n\r\nI'm looking forward to any further insights or thoughts from the community on this matter. It's good to have such conversations going to ensure the security and robustness of the library!\r\n\r\nSincere regards,\r\nAvi" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28299/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28299", "html_url": "https://github.com/huggingface/transformers/pull/28299", "diff_url": "https://github.com/huggingface/transformers/pull/28299.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28299.patch", "merged_at": 1704724408000 }

{ "login": "Hannibal046", "id": 38466901, "node_id": "MDQ6VXNlcjM4NDY2OTAx", "avatar_url": "https://avatars.githubusercontent.com/u/38466901?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Hannibal046", "html_url": "https://github.com/Hannibal046", "followers_url": "https://api.github.com/users/Hannibal046/followers", "following_url": "https://api.github.com/users/Hannibal046/following{/other_user}", "gists_url": "https://api.github.com/users/Hannibal046/gists{/gist_id}", "starred_url": "https://api.github.com/users/Hannibal046/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Hannibal046/subscriptions", "organizations_url": "https://api.github.com/users/Hannibal046/orgs", "repos_url": "https://api.github.com/users/Hannibal046/repos", "events_url": "https://api.github.com/users/Hannibal046/events{/privacy}", "received_events_url": "https://api.github.com/users/Hannibal046/received_events", "type": "User", "site_admin": false }

[]

[]

[ "It seems that the problem was in `from_pretrained`. When I init the model with config, it works fine.\r\n\r\n\r\n", "OK, I see. It is a feature for LLM loading. Using the following would solve the problem:\r\n```python\r\nMyBERT.from_pretrained(\"bert-base-uncased\",_fast_init=False)\r\n```", "Hey! It's more like[ the init scheme](https://github.com/huggingface/transformers/blob/b1292bca6923cfbc9cb3f70cb55df57e4e17e630/src/transformers/models/bert/modeling_bert.py#L744)\r\n\r\nneeds to be updated to your needs! `_fast_init` will skip normal inits and `_init_weights` is the only function that will be called for this. ", "Got it! Appreciate" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28298/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "AjayP13", "id": 5404177, "node_id": "MDQ6VXNlcjU0MDQxNzc=", "avatar_url": "https://avatars.githubusercontent.com/u/5404177?v=4", "gravatar_id": "", "url": "https://api.github.com/users/AjayP13", "html_url": "https://github.com/AjayP13", "followers_url": "https://api.github.com/users/AjayP13/followers", "following_url": "https://api.github.com/users/AjayP13/following{/other_user}", "gists_url": "https://api.github.com/users/AjayP13/gists{/gist_id}", "starred_url": "https://api.github.com/users/AjayP13/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/AjayP13/subscriptions", "organizations_url": "https://api.github.com/users/AjayP13/orgs", "repos_url": "https://api.github.com/users/AjayP13/repos", "events_url": "https://api.github.com/users/AjayP13/events{/privacy}", "received_events_url": "https://api.github.com/users/AjayP13/received_events", "type": "User", "site_admin": false }

[]

[]

[ "cc @younesbelkada ", "@pacman100 Thanks for the insight on this.\r\n\r\nEdit: @pacman100 's right, I just tested this PR again:\r\n\r\n- PEFT + FSDP + Load Best Checkpoint at End: does work\r\n- PEFT + FSDP + Resume From Checkpoint: does not work", "@pacman100 @younesbelkada - The PR is ready for review again and this time it supports FSDP + PEFT + Resuming. It's now reliant on a PR I made over in `accelerate`: https://github.com/huggingface/accelerate/pull/2321.\r\n\r\n@younesbelkada I was able to rent a multi-GPU machine for development and testing, I've attached a test [branch_test.zip](https://github.com/huggingface/transformers/files/13876998/branch_test.zip) file, that may help create an actual test case for the CI system. You can see the `pytorch_model_fsdp.bin` is only 12.15MB now in the test output log (vs. the size of the full model weights).\r\n\r\nHere is the test information:\r\n\r\n**Summary of test:**\r\n\r\n0. Pull in changes from `transformers` PR and `accelerate` PR via `requirements.txt`\r\n1. Train for 99 epochs\r\n2. Resume from Checkpoint\r\n3. Train for 1 Additional Epoch \r\n4. Load Best Checkpoint\r\n5. Test Final Trained Adapter Weights\r\n\r\n<details>\r\n <summary>Test Dataset</summary>\r\n \r\n ```python\r\ndataset = Dataset.from_dict({\r\n \"text\": [\r\n \"Input: A Output: 14\",\r\n \"Input: B Output: 52\",\r\n \"Input: C Output: 83\",\r\n \"Input: D Output: 57\",\r\n \"Input: E Output: 38\",\r\n \"Input: F Output: 54\",\r\n ]\r\n})\r\n ```\r\n</details>\r\n\r\n<details>\r\n <summary>Output Log of Training / Test</summary>\r\n \r\n ```\r\nInstalling requirements.txt...\r\n========= preparing dataset ============\r\n========= loading models + tokenizer ============\r\n========= preparing for training ============\r\n========= training ============\r\n{'loss': 6.8567, 'learning_rate': 0.000494949494949495, 'epoch': 1.0}\r\n{'eval_loss': 5.881069183349609, 'eval_runtime': 0.06, 'eval_samples_per_second': 99.934, 'eval_steps_per_second': 16.656, 'epoch': 1.0}\r\n{'loss': 5.986, 'learning_rate': 0.0004898989898989899, 'epoch': 2.0}\r\n{'eval_loss': 5.117019176483154, 'eval_runtime': 0.0795, 'eval_samples_per_second': 75.427, 'eval_steps_per_second': 12.571, 'epoch': 2.0}\r\n{'loss': 5.2125, 'learning_rate': 0.0004848484848484849, 'epoch': 3.0}\r\n{'eval_loss': 4.231423854827881, 'eval_runtime': 0.0595, 'eval_samples_per_second': 100.84, 'eval_steps_per_second': 16.807, 'epoch': 3.0}\r\n{'loss': 4.6937, 'learning_rate': 0.0004797979797979798, 'epoch': 4.0}\r\n{'eval_loss': 3.4078128337860107, 'eval_runtime': 0.0651, 'eval_samples_per_second': 92.211, 'eval_steps_per_second': 15.369, 'epoch': 4.0}\r\n{'loss': 4.2886, 'learning_rate': 0.00047474747474747476, 'epoch': 5.0}\r\n{'eval_loss': 2.664665937423706, 'eval_runtime': 0.0593, 'eval_samples_per_second': 101.123, 'eval_steps_per_second': 16.854, 'epoch': 5.0}\r\n{'loss': 2.9141, 'learning_rate': 0.0004696969696969697, 'epoch': 6.0}\r\n{'eval_loss': 2.032987117767334, 'eval_runtime': 0.0612, 'eval_samples_per_second': 97.965, 'eval_steps_per_second': 16.328, 'epoch': 6.0}\r\n{'loss': 2.2988, 'learning_rate': 0.0004646464646464646, 'epoch': 7.0}\r\n{'eval_loss': 1.591854453086853, 'eval_runtime': 0.0731, 'eval_samples_per_second': 82.099, 'eval_steps_per_second': 13.683, 'epoch': 7.0}\r\n{'loss': 1.8455, 'learning_rate': 0.00045959595959595964, 'epoch': 8.0}\r\n{'eval_loss': 1.2650986909866333, 'eval_runtime': 0.0669, 'eval_samples_per_second': 89.719, 'eval_steps_per_second': 14.953, 'epoch': 8.0}\r\n{'loss': 1.7772, 'learning_rate': 0.00045454545454545455, 'epoch': 9.0}\r\n{'eval_loss': 1.072060227394104, 'eval_runtime': 0.0671, 'eval_samples_per_second': 89.479, 'eval_steps_per_second': 14.913, 'epoch': 9.0}\r\n{'loss': 1.4851, 'learning_rate': 0.0004494949494949495, 'epoch': 10.0}\r\n{'eval_loss': 0.9297663569450378, 'eval_runtime': 0.0744, 'eval_samples_per_second': 80.613, 'eval_steps_per_second': 13.436, 'epoch': 10.0}\r\n{'loss': 1.4598, 'learning_rate': 0.0004444444444444444, 'epoch': 11.0}\r\n{'eval_loss': 0.8234553933143616, 'eval_runtime': 0.0691, 'eval_samples_per_second': 86.777, 'eval_steps_per_second': 14.463, 'epoch': 11.0}\r\n{'loss': 1.3115, 'learning_rate': 0.0004393939393939394, 'epoch': 12.0}\r\n{'eval_loss': 0.7762313485145569, 'eval_runtime': 0.0601, 'eval_samples_per_second': 99.775, 'eval_steps_per_second': 16.629, 'epoch': 12.0}\r\n{'loss': 0.8095, 'learning_rate': 0.0004343434343434344, 'epoch': 13.0}\r\n{'eval_loss': 0.7559001445770264, 'eval_runtime': 0.0611, 'eval_samples_per_second': 98.177, 'eval_steps_per_second': 16.363, 'epoch': 13.0}\r\n{'loss': 0.8534, 'learning_rate': 0.0004292929292929293, 'epoch': 14.0}\r\n{'eval_loss': 0.749554455280304, 'eval_runtime': 0.0649, 'eval_samples_per_second': 92.517, 'eval_steps_per_second': 15.42, 'epoch': 14.0}\r\n{'loss': 0.8644, 'learning_rate': 0.00042424242424242425, 'epoch': 15.0}\r\n{'eval_loss': 0.7470491528511047, 'eval_runtime': 0.0728, 'eval_samples_per_second': 82.379, 'eval_steps_per_second': 13.73, 'epoch': 15.0}\r\n{'loss': 1.017, 'learning_rate': 0.00041919191919191916, 'epoch': 16.0}\r\n{'eval_loss': 0.7447641491889954, 'eval_runtime': 0.0675, 'eval_samples_per_second': 88.867, 'eval_steps_per_second': 14.811, 'epoch': 16.0}\r\n{'loss': 1.0205, 'learning_rate': 0.0004141414141414142, 'epoch': 17.0}\r\n{'eval_loss': 0.7410345673561096, 'eval_runtime': 0.0605, 'eval_samples_per_second': 99.164, 'eval_steps_per_second': 16.527, 'epoch': 17.0}\r\n{'loss': 1.0018, 'learning_rate': 0.00040909090909090913, 'epoch': 18.0}\r\n{'eval_loss': 0.7377716898918152, 'eval_runtime': 0.0716, 'eval_samples_per_second': 83.775, 'eval_steps_per_second': 13.962, 'epoch': 18.0}\r\n{'loss': 0.7863, 'learning_rate': 0.00040404040404040404, 'epoch': 19.0}\r\n{'eval_loss': 0.7327172756195068, 'eval_runtime': 0.0611, 'eval_samples_per_second': 98.14, 'eval_steps_per_second': 16.357, 'epoch': 19.0}\r\n{'loss': 0.7273, 'learning_rate': 0.000398989898989899, 'epoch': 20.0}\r\n{'eval_loss': 0.7267041206359863, 'eval_runtime': 0.067, 'eval_samples_per_second': 89.487, 'eval_steps_per_second': 14.915, 'epoch': 20.0}\r\n{'loss': 1.0861, 'learning_rate': 0.0003939393939393939, 'epoch': 21.0}\r\n{'eval_loss': 0.7202563285827637, 'eval_runtime': 0.0623, 'eval_samples_per_second': 96.246, 'eval_steps_per_second': 16.041, 'epoch': 21.0}\r\n{'loss': 1.0336, 'learning_rate': 0.0003888888888888889, 'epoch': 22.0}\r\n{'eval_loss': 0.7143564820289612, 'eval_runtime': 0.0599, 'eval_samples_per_second': 100.172, 'eval_steps_per_second': 16.695, 'epoch': 22.0}\r\n{'loss': 0.7619, 'learning_rate': 0.00038383838383838383, 'epoch': 23.0}\r\n{'eval_loss': 0.7085524201393127, 'eval_runtime': 0.0694, 'eval_samples_per_second': 86.504, 'eval_steps_per_second': 14.417, 'epoch': 23.0}\r\n{'loss': 1.0004, 'learning_rate': 0.0003787878787878788, 'epoch': 24.0}\r\n{'eval_loss': 0.704052746295929, 'eval_runtime': 0.0658, 'eval_samples_per_second': 91.239, 'eval_steps_per_second': 15.206, 'epoch': 24.0}\r\n{'loss': 0.6674, 'learning_rate': 0.00037373737373737375, 'epoch': 25.0}\r\n{'eval_loss': 0.7021067142486572, 'eval_runtime': 0.0721, 'eval_samples_per_second': 83.256, 'eval_steps_per_second': 13.876, 'epoch': 25.0}\r\n{'loss': 1.0418, 'learning_rate': 0.0003686868686868687, 'epoch': 26.0}\r\n{'eval_loss': 0.7003771662712097, 'eval_runtime': 0.0624, 'eval_samples_per_second': 96.231, 'eval_steps_per_second': 16.038, 'epoch': 26.0}\r\n{'loss': 0.7463, 'learning_rate': 0.00036363636363636367, 'epoch': 27.0}\r\n{'eval_loss': 0.6978425979614258, 'eval_runtime': 0.076, 'eval_samples_per_second': 78.91, 'eval_steps_per_second': 13.152, 'epoch': 27.0}\r\n{'loss': 0.9374, 'learning_rate': 0.0003585858585858586, 'epoch': 28.0}\r\n{'eval_loss': 0.6954618096351624, 'eval_runtime': 0.0707, 'eval_samples_per_second': 84.814, 'eval_steps_per_second': 14.136, 'epoch': 28.0}\r\n{'loss': 0.6975, 'learning_rate': 0.00035353535353535354, 'epoch': 29.0}\r\n{'eval_loss': 0.6914337277412415, 'eval_runtime': 0.0755, 'eval_samples_per_second': 79.493, 'eval_steps_per_second': 13.249, 'epoch': 29.0}\r\n{'loss': 0.7536, 'learning_rate': 0.0003484848484848485, 'epoch': 30.0}\r\n{'eval_loss': 0.6854903697967529, 'eval_runtime': 0.0596, 'eval_samples_per_second': 100.615, 'eval_steps_per_second': 16.769, 'epoch': 30.0}\r\n{'loss': 0.758, 'learning_rate': 0.00034343434343434346, 'epoch': 31.0}\r\n{'eval_loss': 0.6792461276054382, 'eval_runtime': 0.0597, 'eval_samples_per_second': 100.564, 'eval_steps_per_second': 16.761, 'epoch': 31.0}\r\n{'loss': 0.8828, 'learning_rate': 0.0003383838383838384, 'epoch': 32.0}\r\n{'eval_loss': 0.6729412078857422, 'eval_runtime': 0.0691, 'eval_samples_per_second': 86.891, 'eval_steps_per_second': 14.482, 'epoch': 32.0}\r\n{'loss': 0.9425, 'learning_rate': 0.0003333333333333333, 'epoch': 33.0}\r\n{'eval_loss': 0.6672318577766418, 'eval_runtime': 0.0715, 'eval_samples_per_second': 83.967, 'eval_steps_per_second': 13.994, 'epoch': 33.0}\r\n{'loss': 0.8739, 'learning_rate': 0.0003282828282828283, 'epoch': 34.0}\r\n{'eval_loss': 0.6612274050712585, 'eval_runtime': 0.0691, 'eval_samples_per_second': 86.887, 'eval_steps_per_second': 14.481, 'epoch': 34.0}\r\n{'loss': 0.6459, 'learning_rate': 0.00032323232323232324, 'epoch': 35.0}\r\n{'eval_loss': 0.6533860564231873, 'eval_runtime': 0.069, 'eval_samples_per_second': 86.928, 'eval_steps_per_second': 14.488, 'epoch': 35.0}\r\n{'loss': 0.71, 'learning_rate': 0.0003181818181818182, 'epoch': 36.0}\r\n{'eval_loss': 0.6435515284538269, 'eval_runtime': 0.0622, 'eval_samples_per_second': 96.408, 'eval_steps_per_second': 16.068, 'epoch': 36.0}\r\n{'loss': 0.6217, 'learning_rate': 0.00031313131313131316, 'epoch': 37.0}\r\n{'eval_loss': 0.6283969283103943, 'eval_runtime': 0.0641, 'eval_samples_per_second': 93.673, 'eval_steps_per_second': 15.612, 'epoch': 37.0}\r\n{'loss': 0.8286, 'learning_rate': 0.00030808080808080807, 'epoch': 38.0}\r\n{'eval_loss': 0.6098210215568542, 'eval_runtime': 0.0664, 'eval_samples_per_second': 90.421, 'eval_steps_per_second': 15.07, 'epoch': 38.0}\r\n{'loss': 0.8709, 'learning_rate': 0.00030303030303030303, 'epoch': 39.0}\r\n{'eval_loss': 0.5880406498908997, 'eval_runtime': 0.0605, 'eval_samples_per_second': 99.113, 'eval_steps_per_second': 16.519, 'epoch': 39.0}\r\n{'loss': 0.6882, 'learning_rate': 0.00029797979797979794, 'epoch': 40.0}\r\n{'eval_loss': 0.5603983402252197, 'eval_runtime': 0.0663, 'eval_samples_per_second': 90.523, 'eval_steps_per_second': 15.087, 'epoch': 40.0}\r\n{'loss': 0.8101, 'learning_rate': 0.00029292929292929295, 'epoch': 41.0}\r\n{'eval_loss': 0.535556972026825, 'eval_runtime': 0.0738, 'eval_samples_per_second': 81.351, 'eval_steps_per_second': 13.558, 'epoch': 41.0}\r\n{'loss': 0.988, 'learning_rate': 0.0002878787878787879, 'epoch': 42.0}\r\n{'eval_loss': 0.5133206248283386, 'eval_runtime': 0.0652, 'eval_samples_per_second': 91.956, 'eval_steps_per_second': 15.326, 'epoch': 42.0}\r\n{'loss': 0.7791, 'learning_rate': 0.0002828282828282828, 'epoch': 43.0}\r\n{'eval_loss': 0.49268361926078796, 'eval_runtime': 0.0677, 'eval_samples_per_second': 88.604, 'eval_steps_per_second': 14.767, 'epoch': 43.0}\r\n{'loss': 0.5676, 'learning_rate': 0.0002777777777777778, 'epoch': 44.0}\r\n{'eval_loss': 0.466182678937912, 'eval_runtime': 0.0608, 'eval_samples_per_second': 98.746, 'eval_steps_per_second': 16.458, 'epoch': 44.0}\r\n{'loss': 0.616, 'learning_rate': 0.00027272727272727274, 'epoch': 45.0}\r\n{'eval_loss': 0.4505433738231659, 'eval_runtime': 0.0801, 'eval_samples_per_second': 74.905, 'eval_steps_per_second': 12.484, 'epoch': 45.0}\r\n{'loss': 0.5265, 'learning_rate': 0.0002676767676767677, 'epoch': 46.0}\r\n{'eval_loss': 0.43705224990844727, 'eval_runtime': 0.0742, 'eval_samples_per_second': 80.819, 'eval_steps_per_second': 13.47, 'epoch': 46.0}\r\n{'loss': 0.5378, 'learning_rate': 0.00026262626262626266, 'epoch': 47.0}\r\n{'eval_loss': 0.42776837944984436, 'eval_runtime': 0.0802, 'eval_samples_per_second': 74.852, 'eval_steps_per_second': 12.475, 'epoch': 47.0}\r\n{'loss': 0.5289, 'learning_rate': 0.00025757575757575756, 'epoch': 48.0}\r\n{'eval_loss': 0.41791656613349915, 'eval_runtime': 0.0617, 'eval_samples_per_second': 97.202, 'eval_steps_per_second': 16.2, 'epoch': 48.0}\r\n{'loss': 0.6216, 'learning_rate': 0.0002525252525252525, 'epoch': 49.0}\r\n{'eval_loss': 0.4046964645385742, 'eval_runtime': 0.0597, 'eval_samples_per_second': 100.523, 'eval_steps_per_second': 16.754, 'epoch': 49.0}\r\n{'loss': 0.4525, 'learning_rate': 0.0002474747474747475, 'epoch': 50.0}\r\n{'eval_loss': 0.3916676938533783, 'eval_runtime': 0.0599, 'eval_samples_per_second': 100.174, 'eval_steps_per_second': 16.696, 'epoch': 50.0}\r\n{'loss': 0.4181, 'learning_rate': 0.00024242424242424245, 'epoch': 51.0}\r\n{'eval_loss': 0.38267311453819275, 'eval_runtime': 0.0742, 'eval_samples_per_second': 80.905, 'eval_steps_per_second': 13.484, 'epoch': 51.0}\r\n{'loss': 0.3541, 'learning_rate': 0.00023737373737373738, 'epoch': 52.0}\r\n{'eval_loss': 0.3767699897289276, 'eval_runtime': 0.0595, 'eval_samples_per_second': 100.789, 'eval_steps_per_second': 16.798, 'epoch': 52.0}\r\n{'loss': 0.5992, 'learning_rate': 0.0002323232323232323, 'epoch': 53.0}\r\n{'eval_loss': 0.3728148639202118, 'eval_runtime': 0.0597, 'eval_samples_per_second': 100.515, 'eval_steps_per_second': 16.752, 'epoch': 53.0}\r\n{'loss': 0.614, 'learning_rate': 0.00022727272727272727, 'epoch': 54.0}\r\n{'eval_loss': 0.3699725568294525, 'eval_runtime': 0.0667, 'eval_samples_per_second': 90.017, 'eval_steps_per_second': 15.003, 'epoch': 54.0}\r\n{'loss': 0.4125, 'learning_rate': 0.0002222222222222222, 'epoch': 55.0}\r\n{'eval_loss': 0.3682810962200165, 'eval_runtime': 0.0692, 'eval_samples_per_second': 86.711, 'eval_steps_per_second': 14.452, 'epoch': 55.0}\r\n{'loss': 0.4785, 'learning_rate': 0.0002171717171717172, 'epoch': 56.0}\r\n{'eval_loss': 0.3675954341888428, 'eval_runtime': 0.0728, 'eval_samples_per_second': 82.372, 'eval_steps_per_second': 13.729, 'epoch': 56.0}\r\n{'loss': 0.5642, 'learning_rate': 0.00021212121212121213, 'epoch': 57.0}\r\n{'eval_loss': 0.3668169677257538, 'eval_runtime': 0.0664, 'eval_samples_per_second': 90.378, 'eval_steps_per_second': 15.063, 'epoch': 57.0}\r\n{'loss': 0.3966, 'learning_rate': 0.0002070707070707071, 'epoch': 58.0}\r\n{'eval_loss': 0.3652132451534271, 'eval_runtime': 0.0612, 'eval_samples_per_second': 98.002, 'eval_steps_per_second': 16.334, 'epoch': 58.0}\r\n{'loss': 0.539, 'learning_rate': 0.00020202020202020202, 'epoch': 59.0}\r\n{'eval_loss': 0.36377087235450745, 'eval_runtime': 0.0622, 'eval_samples_per_second': 96.52, 'eval_steps_per_second': 16.087, 'epoch': 59.0}\r\n{'loss': 0.5168, 'learning_rate': 0.00019696969696969695, 'epoch': 60.0}\r\n{'eval_loss': 0.3619898557662964, 'eval_runtime': 0.062, 'eval_samples_per_second': 96.847, 'eval_steps_per_second': 16.141, 'epoch': 60.0}\r\n{'loss': 0.655, 'learning_rate': 0.00019191919191919191, 'epoch': 61.0}\r\n{'eval_loss': 0.36089444160461426, 'eval_runtime': 0.0609, 'eval_samples_per_second': 98.531, 'eval_steps_per_second': 16.422, 'epoch': 61.0}\r\n{'loss': 0.528, 'learning_rate': 0.00018686868686868687, 'epoch': 62.0}\r\n{'eval_loss': 0.36035969853401184, 'eval_runtime': 0.0601, 'eval_samples_per_second': 99.776, 'eval_steps_per_second': 16.629, 'epoch': 62.0}\r\n{'loss': 0.3566, 'learning_rate': 0.00018181818181818183, 'epoch': 63.0}\r\n{'eval_loss': 0.3601444661617279, 'eval_runtime': 0.0758, 'eval_samples_per_second': 79.129, 'eval_steps_per_second': 13.188, 'epoch': 63.0}\r\n{'loss': 0.3974, 'learning_rate': 0.00017676767676767677, 'epoch': 64.0}\r\n{'eval_loss': 0.36019062995910645, 'eval_runtime': 0.0608, 'eval_samples_per_second': 98.757, 'eval_steps_per_second': 16.459, 'epoch': 64.0}\r\n{'loss': 0.3913, 'learning_rate': 0.00017171717171717173, 'epoch': 65.0}\r\n{'eval_loss': 0.36041781306266785, 'eval_runtime': 0.0647, 'eval_samples_per_second': 92.715, 'eval_steps_per_second': 15.453, 'epoch': 65.0}\r\n{'loss': 0.4687, 'learning_rate': 0.00016666666666666666, 'epoch': 66.0}\r\n{'eval_loss': 0.36066195368766785, 'eval_runtime': 0.0616, 'eval_samples_per_second': 97.351, 'eval_steps_per_second': 16.225, 'epoch': 66.0}\r\n{'loss': 0.4079, 'learning_rate': 0.00016161616161616162, 'epoch': 67.0}\r\n{'eval_loss': 0.36099013686180115, 'eval_runtime': 0.0655, 'eval_samples_per_second': 91.608, 'eval_steps_per_second': 15.268, 'epoch': 67.0}\r\n{'loss': 0.3759, 'learning_rate': 0.00015656565656565658, 'epoch': 68.0}\r\n{'eval_loss': 0.3616194427013397, 'eval_runtime': 0.0769, 'eval_samples_per_second': 78.04, 'eval_steps_per_second': 13.007, 'epoch': 68.0}\r\n{'loss': 0.4366, 'learning_rate': 0.00015151515151515152, 'epoch': 69.0}\r\n{'eval_loss': 0.3618600070476532, 'eval_runtime': 0.0631, 'eval_samples_per_second': 95.107, 'eval_steps_per_second': 15.851, 'epoch': 69.0}\r\n{'loss': 0.4562, 'learning_rate': 0.00014646464646464648, 'epoch': 70.0}\r\n{'eval_loss': 0.36202383041381836, 'eval_runtime': 0.0693, 'eval_samples_per_second': 86.527, 'eval_steps_per_second': 14.421, 'epoch': 70.0}\r\n{'loss': 0.4306, 'learning_rate': 0.0001414141414141414, 'epoch': 71.0}\r\n{'eval_loss': 0.3623146116733551, 'eval_runtime': 0.0651, 'eval_samples_per_second': 92.107, 'eval_steps_per_second': 15.351, 'epoch': 71.0}\r\n{'loss': 0.4461, 'learning_rate': 0.00013636363636363637, 'epoch': 72.0}\r\n{'eval_loss': 0.36247682571411133, 'eval_runtime': 0.0602, 'eval_samples_per_second': 99.713, 'eval_steps_per_second': 16.619, 'epoch': 72.0}\r\n{'loss': 0.5163, 'learning_rate': 0.00013131313131313133, 'epoch': 73.0}\r\n{'eval_loss': 0.3626745045185089, 'eval_runtime': 0.0714, 'eval_samples_per_second': 83.979, 'eval_steps_per_second': 13.996, 'epoch': 73.0}\r\n{'loss': 0.4319, 'learning_rate': 0.00012626262626262626, 'epoch': 74.0}\r\n{'eval_loss': 0.36249294877052307, 'eval_runtime': 0.0643, 'eval_samples_per_second': 93.381, 'eval_steps_per_second': 15.563, 'epoch': 74.0}\r\n{'loss': 0.3901, 'learning_rate': 0.00012121212121212122, 'epoch': 75.0}\r\n{'eval_loss': 0.3623819351196289, 'eval_runtime': 0.082, 'eval_samples_per_second': 73.159, 'eval_steps_per_second': 12.193, 'epoch': 75.0}\r\n{'loss': 0.3908, 'learning_rate': 0.00011616161616161616, 'epoch': 76.0}\r\n{'eval_loss': 0.3622519075870514, 'eval_runtime': 0.0729, 'eval_samples_per_second': 82.254, 'eval_steps_per_second': 13.709, 'epoch': 76.0}\r\n{'loss': 0.3609, 'learning_rate': 0.0001111111111111111, 'epoch': 77.0}\r\n{'eval_loss': 0.3618755042552948, 'eval_runtime': 0.0659, 'eval_samples_per_second': 91.017, 'eval_steps_per_second': 15.169, 'epoch': 77.0}\r\n{'loss': 0.3959, 'learning_rate': 0.00010606060606060606, 'epoch': 78.0}\r\n{'eval_loss': 0.3615656793117523, 'eval_runtime': 0.0675, 'eval_samples_per_second': 88.871, 'eval_steps_per_second': 14.812, 'epoch': 78.0}\r\n{'loss': 0.4189, 'learning_rate': 0.00010101010101010101, 'epoch': 79.0}\r\n{'eval_loss': 0.3611675798892975, 'eval_runtime': 0.0701, 'eval_samples_per_second': 85.573, 'eval_steps_per_second': 14.262, 'epoch': 79.0}\r\n{'loss': 0.4515, 'learning_rate': 9.595959595959596e-05, 'epoch': 80.0}\r\n{'eval_loss': 0.36078181862831116, 'eval_runtime': 0.0745, 'eval_samples_per_second': 80.54, 'eval_steps_per_second': 13.423, 'epoch': 80.0}\r\n{'loss': 0.3864, 'learning_rate': 9.090909090909092e-05, 'epoch': 81.0}\r\n{'eval_loss': 0.3604678213596344, 'eval_runtime': 0.0645, 'eval_samples_per_second': 93.02, 'eval_steps_per_second': 15.503, 'epoch': 81.0}\r\n{'loss': 0.3414, 'learning_rate': 8.585858585858586e-05, 'epoch': 82.0}\r\n{'eval_loss': 0.36031660437583923, 'eval_runtime': 0.0598, 'eval_samples_per_second': 100.369, 'eval_steps_per_second': 16.728, 'epoch': 82.0}\r\n{'loss': 0.4023, 'learning_rate': 8.080808080808081e-05, 'epoch': 83.0}\r\n{'eval_loss': 0.3602430522441864, 'eval_runtime': 0.0759, 'eval_samples_per_second': 79.061, 'eval_steps_per_second': 13.177, 'epoch': 83.0}\r\n{'loss': 0.3382, 'learning_rate': 7.575757575757576e-05, 'epoch': 84.0}\r\n{'eval_loss': 0.36019110679626465, 'eval_runtime': 0.0668, 'eval_samples_per_second': 89.801, 'eval_steps_per_second': 14.967, 'epoch': 84.0}\r\n{'loss': 0.392, 'learning_rate': 7.07070707070707e-05, 'epoch': 85.0}\r\n{'eval_loss': 0.360186904668808, 'eval_runtime': 0.0853, 'eval_samples_per_second': 70.381, 'eval_steps_per_second': 11.73, 'epoch': 85.0}\r\n{'loss': 0.5104, 'learning_rate': 6.565656565656566e-05, 'epoch': 86.0}\r\n{'eval_loss': 0.36022713780403137, 'eval_runtime': 0.0673, 'eval_samples_per_second': 89.159, 'eval_steps_per_second': 14.86, 'epoch': 86.0}\r\n{'loss': 0.3643, 'learning_rate': 6.060606060606061e-05, 'epoch': 87.0}\r\n{'eval_loss': 0.3602248728275299, 'eval_runtime': 0.0629, 'eval_samples_per_second': 95.465, 'eval_steps_per_second': 15.911, 'epoch': 87.0}\r\n{'loss': 0.3615, 'learning_rate': 5.555555555555555e-05, 'epoch': 88.0}\r\n{'eval_loss': 0.3601932227611542, 'eval_runtime': 0.066, 'eval_samples_per_second': 90.975, 'eval_steps_per_second': 15.163, 'epoch': 88.0}\r\n{'loss': 0.396, 'learning_rate': 5.0505050505050505e-05, 'epoch': 89.0}\r\n{'eval_loss': 0.3601706922054291, 'eval_runtime': 0.064, 'eval_samples_per_second': 93.73, 'eval_steps_per_second': 15.622, 'epoch': 89.0}\r\n{'loss': 0.3589, 'learning_rate': 4.545454545454546e-05, 'epoch': 90.0}\r\n{'eval_loss': 0.3601481020450592, 'eval_runtime': 0.0596, 'eval_samples_per_second': 100.659, 'eval_steps_per_second': 16.777, 'epoch': 90.0}\r\n{'loss': 0.4028, 'learning_rate': 4.0404040404040405e-05, 'epoch': 91.0}\r\n{'eval_loss': 0.3601456880569458, 'eval_runtime': 0.0711, 'eval_samples_per_second': 84.444, 'eval_steps_per_second': 14.074, 'epoch': 91.0}\r\n{'loss': 0.3747, 'learning_rate': 3.535353535353535e-05, 'epoch': 92.0}\r\n{'eval_loss': 0.3601391017436981, 'eval_runtime': 0.0611, 'eval_samples_per_second': 98.232, 'eval_steps_per_second': 16.372, 'epoch': 92.0}\r\n{'loss': 0.3753, 'learning_rate': 3.0303030303030306e-05, 'epoch': 93.0}\r\n{'eval_loss': 0.3601249158382416, 'eval_runtime': 0.0682, 'eval_samples_per_second': 87.988, 'eval_steps_per_second': 14.665, 'epoch': 93.0}\r\n{'loss': 0.397, 'learning_rate': 2.5252525252525253e-05, 'epoch': 94.0}\r\n{'eval_loss': 0.36011412739753723, 'eval_runtime': 0.0629, 'eval_samples_per_second': 95.374, 'eval_steps_per_second': 15.896, 'epoch': 94.0}\r\n{'loss': 0.399, 'learning_rate': 2.0202020202020203e-05, 'epoch': 95.0}\r\n{'eval_loss': 0.36010661721229553, 'eval_runtime': 0.0609, 'eval_samples_per_second': 98.569, 'eval_steps_per_second': 16.428, 'epoch': 95.0}\r\n{'loss': 0.4198, 'learning_rate': 1.5151515151515153e-05, 'epoch': 96.0}\r\n{'eval_loss': 0.3600958287715912, 'eval_runtime': 0.0704, 'eval_samples_per_second': 85.187, 'eval_steps_per_second': 14.198, 'epoch': 96.0}\r\n{'loss': 0.3618, 'learning_rate': 1.0101010101010101e-05, 'epoch': 97.0}\r\n{'eval_loss': 0.3600878417491913, 'eval_runtime': 0.0691, 'eval_samples_per_second': 86.839, 'eval_steps_per_second': 14.473, 'epoch': 97.0}\r\n{'loss': 0.4199, 'learning_rate': 5.050505050505051e-06, 'epoch': 98.0}\r\n{'eval_loss': 0.3600861728191376, 'eval_runtime': 0.0622, 'eval_samples_per_second': 96.43, 'eval_steps_per_second': 16.072, 'epoch': 98.0}\r\n{'loss': 0.3681, 'learning_rate': 0.0, 'epoch': 99.0}\r\n{'eval_loss': 0.3600858747959137, 'eval_runtime': 0.0644, 'eval_samples_per_second': 93.18, 'eval_steps_per_second': 15.53, 'epoch': 99.0}\r\n{'train_runtime': 201.2614, 'train_samples_per_second': 2.951, 'train_steps_per_second': 0.492, 'train_loss': 0.9207682233266156, 'epoch': 99.0}\r\n========= preparing dataset ============\r\n========= loading models + tokenizer ============\r\n========= preparing for training ============\r\n========= resuming ============\r\n{'loss': 0.3515, 'learning_rate': 0.0, 'epoch': 100.0}\r\n{'eval_loss': 0.3600858747959137, 'eval_runtime': 0.0703, 'eval_samples_per_second': 85.379, 'eval_steps_per_second': 14.23, 'epoch': 100.0}\r\n{'train_runtime': 3.0627, 'train_samples_per_second': 195.905, 'train_steps_per_second': 32.651, 'train_loss': 0.0035148251056671144, 'epoch': 100.0}\r\n========= load best adapter at: ./output/checkpoint-99 ============\r\npytorch_model_fsdp.bin size: ./output/checkpoint-99: 12.15 MB\r\n[{'generated_text': 'Input: A Output: 14'}]\r\n[{'generated_text': 'Input: B Output: Output'}]\r\n[{'generated_text': 'Input: C Output: 83'}]\r\n[{'generated_text': 'Input: D Output: 57'}]\r\n[{'generated_text': 'Input: E Output: 38'}]\r\n[{'generated_text': 'Input: F Output: 54'}]\r\n ```\r\n</details>\r\n\r\n", "Hello, Thank you @AjayP13 for reiterating. Do you observe decrease in GPU memory usage with PEFT + FSDP? I had to do the following wherein `use_orig_params` needed to be False and the custom auto wrap policy to account for the different modules having trianable and non-trainable parameters\r\n\r\nhttps://github.com/pacman100/DHS-LLM-Workshop/blob/08b3bd5e618c9e258bf06390c78e28764daff273/chat_assistant/training/train.py#L167-L190\r\n\r\n\r\n\r\n", "> Hello, Thank you @AjayP13 for reiterating. Do you observe decrease in GPU memory usage with PEFT + FSDP? I had to do the following wherein `use_orig_params` needed to be False and the custom auto wrap policy to account for the different modules having trianable and non-trainable parameters\r\n> \r\n> https://github.com/pacman100/DHS-LLM-Workshop/blob/08b3bd5e618c9e258bf06390c78e28764daff273/chat_assistant/training/train.py#L167-L190 \r\n\r\nThis does indeed save RAM in my tests:\r\n\r\n**FSDP + PEFT:**\r\n\r\nDevice Memory Usage 0 -- 534.88 MB\r\nDevice Memory Usage 1 -- 524.16 MB\r\n`pytorch_model_fsdp.bin` Size: 12.15 MB\r\n\r\n**FSDP + No PEFT:**\r\nDevice Memory Usage 0 -- 992.93 MB\r\nDevice Memory Usage 1 -- 992.75 MB\r\n`pytorch_model_fsdp.bin` Size: 622.00 MB\r\n\r\nI believe the issue you are referring to was in PyTorch and was possibly fixed in the latest versions according to the discussion here: https://github.com/pytorch/pytorch/issues/91165#issuecomment-1869255170", "> Thank you @AjayP13 for these changes. As replied on the other PR, if this is not meant when using `FULL_STATE_DICT`, the argument `adapter_only` should be set accordingly.\r\n\r\nReplied in the other PR @pacman100, but it should work with `SHARDED_STATE_DICT` as well.", "@AjayP13 There was a recent fix push to main, which resolves the failing tests with `natten` currently on the CI. Rebasing and pushing the changes to trigger a new CI run should resolve this.", "@amyeroberts @younesbelkada Done, I've rebased the changes upon the latest `transformers` main and tests are passing now.", "Thanks @AjayP13 ! Let's merge this PR once https://github.com/huggingface/accelerate/pull/2321 gets merged!", "@amyeroberts @younesbelkada Could one of you please merge this PR now, the `accelerate` PR is now merged. :) " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28297/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28297", "html_url": "https://github.com/huggingface/transformers/pull/28297", "diff_url": "https://github.com/huggingface/transformers/pull/28297.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28297.patch", "merged_at": 1706548215000 }

{ "login": "ekuznetsov139", "id": 12205429, "node_id": "MDQ6VXNlcjEyMjA1NDI5", "avatar_url": "https://avatars.githubusercontent.com/u/12205429?v=4", "gravatar_id": "", "url": "https://api.github.com/users/ekuznetsov139", "html_url": "https://github.com/ekuznetsov139", "followers_url": "https://api.github.com/users/ekuznetsov139/followers", "following_url": "https://api.github.com/users/ekuznetsov139/following{/other_user}", "gists_url": "https://api.github.com/users/ekuznetsov139/gists{/gist_id}", "starred_url": "https://api.github.com/users/ekuznetsov139/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/ekuznetsov139/subscriptions", "organizations_url": "https://api.github.com/users/ekuznetsov139/orgs", "repos_url": "https://api.github.com/users/ekuznetsov139/repos", "events_url": "https://api.github.com/users/ekuznetsov139/events{/privacy}", "received_events_url": "https://api.github.com/users/ekuznetsov139/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Here's what needs to be done:\r\n\r\nhttps://github.com/ROCmSoftwarePlatform/transformers/commit/0b3d80de089bde6ad163042916b35bd26d38b434\r\n\r\nSome of this may break keras2 operation, I began adding version checks but have not had time to do it properly. I had to disable jit_compile, because I was getting XLA-related errors and this was an easy way out; I need to investigate and fix that problem as well.\r\n\r\nThis gets me as far as being able to train for at least one epoch. Loss values seem to be off but the model loads and trains and loss goes down with time.", "I'll just keep talking to myself here, nevermind me.\r\n\r\nhttps://github.com/ROCmSoftwarePlatform/transformers/commit/a488708118999418d6fc3640ea2680aaa97ed21d\r\n\r\nIt trains, apparently correctly, on multiple GPUs (using tf.distribute.MirroredStrategy) and with XLA enabled.\r\n\r\nReported loss is multiplied by the number of GPUs, and I can't quite work out why.\r\n\r\nThe bigger issue, however, is that mixed precision is broken:\r\n\r\n```\r\n File \"/usr/local/lib/python3.9/dist-packages/keras/src/backend/tensorflow/trainer.py\", line 105, in one_step_on_data **\r\n return self.train_step(data)\r\n File \"/usr/local/lib/python3.9/dist-packages/transformers/modeling_tf_utils.py\", line 1703, in train_step\r\n self.optimizer.apply_gradients(zip(grads, self.trainable_variables))\r\n File \"/usr/local/lib/python3.9/dist-packages/keras/src/optimizers/base_optimizer.py\", line 206, in apply_gradients\r\n self.apply(grads, trainable_variables)\r\n File \"/usr/local/lib/python3.9/dist-packages/keras/src/optimizers/loss_scale_optimizer.py\", line 183, in apply\r\n ops.cond(finite, handle_finite_grads, handle_non_finite_grads)\r\n File \"/usr/local/lib/python3.9/dist-packages/keras/src/ops/core.py\", line 594, in cond\r\n return Cond()(pred, true_fn, false_fn)\r\n File \"/usr/local/lib/python3.9/dist-packages/keras/src/utils/traceback_utils.py\", line 123, in error_handler\r\n raise e.with_traceback(filtered_tb) from None\r\n File \"/usr/local/lib/python3.9/dist-packages/keras/src/backend/tensorflow/optimizer.py\", line 82, in _internal_apply_gradients\r\n tf.__internal__.distribute.interim.maybe_merge_call(\r\n\r\n RuntimeError: Exception encountered when calling Cond.call().\r\n \r\n `merge_call` called while defining a new graph or a tf.function. This can often happen if the function `fn` passed to `strategy.run()` contains a nested `@tf.function`, and the nested `@tf.function` contains a synchronization point, such as aggregating gradients (e.g, optimizer.apply_gradients), or if the function `fn` uses a control flow statement which contains a synchronization point in the body. Such behaviors are not yet supported. Instead, please avoid nested `tf.function`s or control flow statements that may potentially cross a synchronization boundary, for example, wrap the `fn` passed to `strategy.run` or the entire `strategy.run` inside a `tf.function` or move the control flow out of `fn`. If you are subclassing a `tf.keras.Model`, please avoid decorating overridden methods `test_step` and `train_step` in `tf.function`.\r\n```\r\nGoing to tackle this one today.", "thanks! cc @Rocketknight1 ", "Ok, the mixed precision issue from my last post was actually fixed in Keras, and I was only seeing it because I had a somewhat outdated version of Keras3 in the system (2023-10-17 instead of 2023-12-31.)\r\n\r\nThere was another issue with mixed precision which only affected testing (I had it fixed in the GPT-2 pathway, it may affect other models.) Saving was also broken. Here's a patch that fixes everything I found, except loss being multiplied by #GPUs:\r\n\r\nhttps://github.com/huggingface/transformers/commit/4fa126006a2db315cc4c9d8c4606b329292d0b95", "You should open a PR with the patch! 🤗 (linking this issue) ", "Will do once I'm satisfied that I've resolved all the issues.", "Hi @ekuznetsov139, thanks for the investigation here - this looks really good! Just to give you some context, the reason the errors change in the latest `main` version of `transformers` is that I've been working on Keras 3 PRs behind the scenes as well. The biggest one is that we now use proper `build()` methods for all our TF models instead of building them with dummy inputs - this avoids lots of issues related to name hierarchies that changed in Keras 3. You can see some of the PRs here:\r\n\r\n- #27794 \r\n- #28046\r\n- #28081\r\n- #28146\r\n\r\nI think the plan from here is that in our TensorFlow code, we're going to **completely remove all direct imports of `keras`, and only use `from tensorflow import keras`.** This ties the Keras version to the TF version, although we will still need to support Keras 3 as we understand that the built-in version of Keras is going to be Keras 3 starting from TF 2.16.\r\n\r\nOur primary goal is to ensure that Keras 3 doesn't break backward compatibility for TF code, even if we don't fully support other frameworks with Keras 3. Once backward compatibility is secure, we have plans to fully support Keras 3, which will probably require a community push to make full Keras ports of all of our models that don't use any TensorFlow ops - there's a partial PR at #26224 but it's on hold because of the number of other backward compatibility issues that need to be resolved first.", "Hi @ekuznetsov139 I also meet the same problems when I used tensorflow & keras 3 to load transformers models. Do you fix it?", "Hi @lingluodlut @ekuznetsov139, I believe this is the last PR we need https://github.com/huggingface/transformers/pull/28588\r\n\r\nNote that we still won't have full Keras 3 support, but at least Transformers will continue working when Keras 3 is installed after this PR is merged." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28296/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "bytebarde", "id": 154845754, "node_id": "U_kgDOCTrCOg", "avatar_url": "https://avatars.githubusercontent.com/u/154845754?v=4", "gravatar_id": "", "url": "https://api.github.com/users/bytebarde", "html_url": "https://github.com/bytebarde", "followers_url": "https://api.github.com/users/bytebarde/followers", "following_url": "https://api.github.com/users/bytebarde/following{/other_user}", "gists_url": "https://api.github.com/users/bytebarde/gists{/gist_id}", "starred_url": "https://api.github.com/users/bytebarde/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/bytebarde/subscriptions", "organizations_url": "https://api.github.com/users/bytebarde/orgs", "repos_url": "https://api.github.com/users/bytebarde/repos", "events_url": "https://api.github.com/users/bytebarde/events{/privacy}", "received_events_url": "https://api.github.com/users/bytebarde/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Current progress with running `flash_attn_test`. Will dive deeper to fix the error.\r\n\r\n\r\n\r\n\r\n\r\n", "Hi @bytebarde, what is the error message? \r\nIf it is something like - \"IndexError: tensors used as ...\", then updating CUDA could solve the error (At least it was for my case in OPT).\r\n \r\nBTW run `make fixup` to make the CI green!", "Hi @susnato, thank you so much for for your attention to this PR!\r\n\r\nI believe the error originates from two factors: (1) my preliminary implementation of `GPTJFlashAttention2`, which aimed to eliminate \"redundant\" transposing of the key and query, and (2) the execution of `test_flash_attn_2_generate_padding_right` using the testing configuration.\r\n\r\nTo address these issues, I have reinstated the original transposing operations and reverted the QKV cache concatenation. Additionally, I overwrote `test_flash_attn_2_generate_padding_right` by using the actual checkpoint and passed all eight tests, similar to what @younesbelkada and you did for llama2 and phi2.\r\n\r\nCurrently, the code has some problems with `make fixup`. Will work on this for the next step.\r\n\r\n\r\n\r\n\r\n\r\n\r\n", "Hi @younesbelkada,\r\n\r\nI believe this pull request is now ready for your review.\r\n\r\nI'd like to highlight a few changes, especially regarding `check_copies.py`, that I'm not entirely confident about. To ensure the branch passes the `make fixup` check, I removed the \"copies\" lines before both `modeling_codegen.CodeGenBlock` and `test_modeling_gptj.test_flash_attn_2_generate_padding_right`. This was done because the changes involved are somehow complex.\r\n\r\nI would really appreciate your guidance on this. If there's a more standard or preferable way to handle such intricate changes, please let me know so I can make the necessary adjustments.\r\n\r\nThank you for your time on this!", "Hi @younesbelkada, thank you very much for your valuable input and guidance! I apologize for the delayed response.\r\n\r\nI've addressed the comment regarding the copy mechanism, and the branch successfully passed the `make fixup` test.\r\n\r\nAdditionally, I've conducted the speed test. However, the observed speedup was not as significant as what we noted with OPT. The test was performed on an Nvidia RTX 4090, utilizing `max-batch-size=8` and `max-seqlen=32` to conserve memory. The model checkpoint used was `EleutherAI/gpt-j-6b` with the revision set to \"float16\". I've attached the speedup graph below for your review.\r\n\r\n\r\n\r\n\r\n\r\n\r\nCould you also perform the test on an A100 GPU for comparison?\r\n\r\nThank you once again for your time. I look forward to hearing your thoughts on this!\r\n\r\n\r\n\r\n", "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/transformers/pr_28295). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "Hi @ArthurZucker and @younesbelkada ,\r\n\r\nThank you so much for your additional suggestions!\r\n\r\nI am sorry. I had assumed that `GPTJ_ATTENTION_CLASSES` had already been introduced by @ArthurZucker previously...\r\n\r\nI have now added GPTJ_ATTENTION_CLASSES and made the necessary code modifications. \r\nFurthermore, I re-ran the test suite and successfully passed all the tests.\r\n\r\nPlease let me know if there's anything more I can do!\r\nThank you so much! ", "Good for me merging! 🤗 " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28295/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28295", "html_url": "https://github.com/huggingface/transformers/pull/28295", "diff_url": "https://github.com/huggingface/transformers/pull/28295.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28295.patch", "merged_at": null }

{ "login": "surapuramakhil", "id": 9161543, "node_id": "MDQ6VXNlcjkxNjE1NDM=", "avatar_url": "https://avatars.githubusercontent.com/u/9161543?v=4", "gravatar_id": "", "url": "https://api.github.com/users/surapuramakhil", "html_url": "https://github.com/surapuramakhil", "followers_url": "https://api.github.com/users/surapuramakhil/followers", "following_url": "https://api.github.com/users/surapuramakhil/following{/other_user}", "gists_url": "https://api.github.com/users/surapuramakhil/gists{/gist_id}", "starred_url": "https://api.github.com/users/surapuramakhil/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/surapuramakhil/subscriptions", "organizations_url": "https://api.github.com/users/surapuramakhil/orgs", "repos_url": "https://api.github.com/users/surapuramakhil/repos", "events_url": "https://api.github.com/users/surapuramakhil/events{/privacy}", "received_events_url": "https://api.github.com/users/surapuramakhil/received_events", "type": "User", "site_admin": false }

[]

[]

[ "After installing Accelerate please try restarting the runtime, this is a common issue :) ", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28294/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "siebeniris", "id": 1593540, "node_id": "MDQ6VXNlcjE1OTM1NDA=", "avatar_url": "https://avatars.githubusercontent.com/u/1593540?v=4", "gravatar_id": "", "url": "https://api.github.com/users/siebeniris", "html_url": "https://github.com/siebeniris", "followers_url": "https://api.github.com/users/siebeniris/followers", "following_url": "https://api.github.com/users/siebeniris/following{/other_user}", "gists_url": "https://api.github.com/users/siebeniris/gists{/gist_id}", "starred_url": "https://api.github.com/users/siebeniris/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/siebeniris/subscriptions", "organizations_url": "https://api.github.com/users/siebeniris/orgs", "repos_url": "https://api.github.com/users/siebeniris/repos", "events_url": "https://api.github.com/users/siebeniris/events{/privacy}", "received_events_url": "https://api.github.com/users/siebeniris/received_events", "type": "User", "site_admin": false }

[]

[]

[ "hmmm do you know what might be happening here @Narsil ? With mt5", "Might be fixed by #28414 ? ", "Happy to take a look if I can have acces either to the finetune (even dummy I just need to look at those tensors) or a reproducer.\r\n\r\nI have no idea what makes some tensors non contiguous and what kind of non contiguous those are\r\n", "Non-contiguous parameters/buffers can be saved with `safe_serialization=False` but not with `safe_serialization=True`.", "I was try to ask more, what lib is actually creating non contiguous tensors ? Seems odd to me that we need to create non contiguous tensors for training.\r\n\r\nDeepspeed for isntant it's not non contiguous it' s more that they abuse the storage system to force several matmul locality (which I think it to optimize network transport), therefore it was easy to fix once identified (because that's a condition where it's easy to rework the tensors on behalf of users since the non contiguity is not really important for the model).", "I ran into this issue due to a custom weight tying scheme (output layer is a transpose of the vocabulary embedding, so the former is not contiguous). I got around the error by turning off safe serialization as noted above. ", "Hi, thanks all for the comments. I have no idea why there are even non-contiguous tensors. I think make them contiguous makes more sense? And it solves the problem and the model training seems to be well. I found it odd that the error doesn't occur for trains T5 models, only for MT5 models, since MT5 is built upon T5 in transformers scripts.\r\n" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28293/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "CyrusVorwald", "id": 90732384, "node_id": "MDQ6VXNlcjkwNzMyMzg0", "avatar_url": "https://avatars.githubusercontent.com/u/90732384?v=4", "gravatar_id": "", "url": "https://api.github.com/users/CyrusVorwald", "html_url": "https://github.com/CyrusVorwald", "followers_url": "https://api.github.com/users/CyrusVorwald/followers", "following_url": "https://api.github.com/users/CyrusVorwald/following{/other_user}", "gists_url": "https://api.github.com/users/CyrusVorwald/gists{/gist_id}", "starred_url": "https://api.github.com/users/CyrusVorwald/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/CyrusVorwald/subscriptions", "organizations_url": "https://api.github.com/users/CyrusVorwald/orgs", "repos_url": "https://api.github.com/users/CyrusVorwald/repos", "events_url": "https://api.github.com/users/CyrusVorwald/events{/privacy}", "received_events_url": "https://api.github.com/users/CyrusVorwald/received_events", "type": "User", "site_admin": false }

[]

[]

[ "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored.", "Hi @CyrusVorwald, thanks for opening this issue! \r\n\r\n`get_depth_map` isn't defined in the transformers library, and so it's not something we can work on. I'd suggest opening a discussion on the model page and sharing these results. \r\n\r\n@NielsRogge Could you look into some weights being randomly initialized when loading from this checkpoint? ", "I think the warning being shown of some weights not being initialized happened after @younesbelkada added support for DPT-hybrid in the `modeling_dpt.py` code. This hybrid version of DPT introduced some other parameters, which aren't used by the default DPT model." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28292/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "zhangnju", "id": 8900927, "node_id": "MDQ6VXNlcjg5MDA5Mjc=", "avatar_url": "https://avatars.githubusercontent.com/u/8900927?v=4", "gravatar_id": "", "url": "https://api.github.com/users/zhangnju", "html_url": "https://github.com/zhangnju", "followers_url": "https://api.github.com/users/zhangnju/followers", "following_url": "https://api.github.com/users/zhangnju/following{/other_user}", "gists_url": "https://api.github.com/users/zhangnju/gists{/gist_id}", "starred_url": "https://api.github.com/users/zhangnju/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/zhangnju/subscriptions", "organizations_url": "https://api.github.com/users/zhangnju/orgs", "repos_url": "https://api.github.com/users/zhangnju/repos", "events_url": "https://api.github.com/users/zhangnju/events{/privacy}", "received_events_url": "https://api.github.com/users/zhangnju/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Maybe cc @ydshieh as you have played with RAG recently. ", "Hi @zhangnju \r\n\r\nWhat is your `datasets` version?\r\n\r\nOn [this google colab](https://colab.research.google.com/drive/1pCLInwexxrwg2WGIxgjmFMs6YeHZsCOm?usp=sharing), I can't reproduce this issue, but it uses `datasets==2.16.1`.", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28291/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "chedatomasz", "id": 47541285, "node_id": "MDQ6VXNlcjQ3NTQxMjg1", "avatar_url": "https://avatars.githubusercontent.com/u/47541285?v=4", "gravatar_id": "", "url": "https://api.github.com/users/chedatomasz", "html_url": "https://github.com/chedatomasz", "followers_url": "https://api.github.com/users/chedatomasz/followers", "following_url": "https://api.github.com/users/chedatomasz/following{/other_user}", "gists_url": "https://api.github.com/users/chedatomasz/gists{/gist_id}", "starred_url": "https://api.github.com/users/chedatomasz/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/chedatomasz/subscriptions", "organizations_url": "https://api.github.com/users/chedatomasz/orgs", "repos_url": "https://api.github.com/users/chedatomasz/repos", "events_url": "https://api.github.com/users/chedatomasz/events{/privacy}", "received_events_url": "https://api.github.com/users/chedatomasz/received_events", "type": "User", "site_admin": false }

[]

[]

[]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28290/reactions", "total_count": 1, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 1 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28290", "html_url": "https://github.com/huggingface/transformers/pull/28290", "diff_url": "https://github.com/huggingface/transformers/pull/28290.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28290.patch", "merged_at": null }

{ "login": "kedarthakkar", "id": 22385733, "node_id": "MDQ6VXNlcjIyMzg1NzMz", "avatar_url": "https://avatars.githubusercontent.com/u/22385733?v=4", "gravatar_id": "", "url": "https://api.github.com/users/kedarthakkar", "html_url": "https://github.com/kedarthakkar", "followers_url": "https://api.github.com/users/kedarthakkar/followers", "following_url": "https://api.github.com/users/kedarthakkar/following{/other_user}", "gists_url": "https://api.github.com/users/kedarthakkar/gists{/gist_id}", "starred_url": "https://api.github.com/users/kedarthakkar/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/kedarthakkar/subscriptions", "organizations_url": "https://api.github.com/users/kedarthakkar/orgs", "repos_url": "https://api.github.com/users/kedarthakkar/repos", "events_url": "https://api.github.com/users/kedarthakkar/events{/privacy}", "received_events_url": "https://api.github.com/users/kedarthakkar/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Another option could be to instead define a function `model.generate_streaming` which takes in `batch_size` and yields generated sequences in `batch_size` increments. This would then allow the user to determine what they want to do with the yielded sequences (i.e. write them to a file or compute some metrics).", "cc @gante ", "Hi @kedarthakkar 👋 \r\n\r\nYou can easily control the batch size with a few extra lines before calling `model.generate(), e.g. looping over your batch using smaller batches as input to the function. That is a much simpler solution than adding tens of lines of additional complexity to an already complex function :)", "Ok, makes sense, I'll close this issue." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28289/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "ercaronte", "id": 25960640, "node_id": "MDQ6VXNlcjI1OTYwNjQw", "avatar_url": "https://avatars.githubusercontent.com/u/25960640?v=4", "gravatar_id": "", "url": "https://api.github.com/users/ercaronte", "html_url": "https://github.com/ercaronte", "followers_url": "https://api.github.com/users/ercaronte/followers", "following_url": "https://api.github.com/users/ercaronte/following{/other_user}", "gists_url": "https://api.github.com/users/ercaronte/gists{/gist_id}", "starred_url": "https://api.github.com/users/ercaronte/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/ercaronte/subscriptions", "organizations_url": "https://api.github.com/users/ercaronte/orgs", "repos_url": "https://api.github.com/users/ercaronte/repos", "events_url": "https://api.github.com/users/ercaronte/events{/privacy}", "received_events_url": "https://api.github.com/users/ercaronte/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Sorry @ylacombe for a last minute (small) code change.\r\n\r\nI was running a load test, where I transcribed a 2 hours audio file and I stumbled into another `torch.std_mean` instruction (at line 2602) that I did not fix in the initial commit.\r\nAfter this second fix I got no more issues and could complete the full transcription. \r\nI also reviewed all the other changes in your initial PR https://github.com/huggingface/transformers/pull/28114 and I could not find any other apparent incompatibility with the current status of the MPS backend.\r\n\r\nPlease have a quick check :-)", "No worries, could you actually use the `is_torch_mps_available` method available in `utils` to check if `torch.std_mean` is available instead of the error catching mechanism ?\r\n", "Sure, I am totally fine with it and I agree it is more readable.\r\n\r\nIn theory, the `is_torch_mps_available` check verifies if mps is available, but the user can also set the torch device to be the CPU or other external CUDA cards. In any case, the code is exactly equivalent and it is much more readable. So that when in the future the MPS backend will support the std_mean this (hopefully temporary) fix will be immediately spotted and removed.\r\n\r\nI have just committed the changes. Tested with the short and long audio files.\r\n\r\nPlease have a check :-)", "LGTM ! thanks for the quick iteration !\r\n", "Thank you @amyeroberts, the proposed changes make the code much cleaner.\r\nI didn't dare to be so drastic as the initial commit, but I am quite happy this way. ", "Thanks again @ercaronte, merging!" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28288/reactions", "total_count": 3, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 3, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28288", "html_url": "https://github.com/huggingface/transformers/pull/28288", "diff_url": "https://github.com/huggingface/transformers/pull/28288.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28288.patch", "merged_at": 1704212549000 }

{ "login": "fleonce", "id": 8986525, "node_id": "MDQ6VXNlcjg5ODY1MjU=", "avatar_url": "https://avatars.githubusercontent.com/u/8986525?v=4", "gravatar_id": "", "url": "https://api.github.com/users/fleonce", "html_url": "https://github.com/fleonce", "followers_url": "https://api.github.com/users/fleonce/followers", "following_url": "https://api.github.com/users/fleonce/following{/other_user}", "gists_url": "https://api.github.com/users/fleonce/gists{/gist_id}", "starred_url": "https://api.github.com/users/fleonce/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/fleonce/subscriptions", "organizations_url": "https://api.github.com/users/fleonce/orgs", "repos_url": "https://api.github.com/users/fleonce/repos", "events_url": "https://api.github.com/users/fleonce/events{/privacy}", "received_events_url": "https://api.github.com/users/fleonce/received_events", "type": "User", "site_admin": false }

[]

[]

[ "When confirmed this is unexpected behaviour I would be willing to research behaviour of other tokenizers in the same setting and submit a pull request addressing the issue :-)", "Thanks @fleonce for your issue! @ArthurZucker will take a look at your issue as soon as he's back from leave, in about a week. Thanks a lot!", "Hey @fleonce the issue is rather with https://github.com/huggingface/transformers/blob/b1292bca6923cfbc9cb3f70cb55df57e4e17e630/src/transformers/models/mbart50/tokenization_mbart50.py#L249\r\n\r\nthe `legacy_added_tokens` was added in #23909 to be BC. \r\nI am almost certain that if you checkout an earlier commit the same issue would be present since the `convert_tokens_to_string` method is what is wrong for me here since it takes : `['en_XX', '▁This', '▁is', '▁my', '▁example', '▁sentence', '▁with', '▁a', '▁special']` and outputs `'en_XXThis is my example sentence with a special'`. If you want to open a PR for a fix feel free to do so! \r\nThis might affect tokenizer with the exact same `convert_tokens_to_string`", "In this case, the `_decode` might need a tiny rework ! ", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28287/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "AmitMY", "id": 5757359, "node_id": "MDQ6VXNlcjU3NTczNTk=", "avatar_url": "https://avatars.githubusercontent.com/u/5757359?v=4", "gravatar_id": "", "url": "https://api.github.com/users/AmitMY", "html_url": "https://github.com/AmitMY", "followers_url": "https://api.github.com/users/AmitMY/followers", "following_url": "https://api.github.com/users/AmitMY/following{/other_user}", "gists_url": "https://api.github.com/users/AmitMY/gists{/gist_id}", "starred_url": "https://api.github.com/users/AmitMY/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/AmitMY/subscriptions", "organizations_url": "https://api.github.com/users/AmitMY/orgs", "repos_url": "https://api.github.com/users/AmitMY/repos", "events_url": "https://api.github.com/users/AmitMY/events{/privacy}", "received_events_url": "https://api.github.com/users/AmitMY/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hi @AmitMY, thanks for opening an issue! \r\n\r\nThere's a few different points to address here. \r\n\r\n> When using --train_file dataset.csv, the tokenizer fails if the caption is \"None\", \"null\" or \"NA\"\r\n\r\nIs this the same dataset being prepared in the `README.md` example or a different one? \r\n\r\nAs a general comment, the scripts are meant to provide a simple script to show how to train a model for a specific task, but it isn't written to cover all use cases. You may need to adapt processing code to your use case.\r\n\r\n> There seems to be no parameter to specify the hub repository to push to.\r\n\r\nThe model will be pushed to the hub under `{YOUR_USERNAME}/output_dir`. If you wish to change this, you can use `--push_to_hub_model_id` to control this\r\n\r\n> Also, there seems to be no place to track the experiment (like wandb)\r\n\r\nThere are many different integrations available with Trainer. For wandb: \r\nhttps://docs.wandb.ai/guides/integrations/huggingface\r\n\r\n> Actual issue\r\n\r\nWithout a copy of the data in `train.csv` it's not possible for us to help here. Could you share an example sample or publicly available dataset that triggers the same error? ", "Thanks for your response. \r\n\r\n---\r\n\r\nRegarding the `None` issue - \r\nthat's just the default behavior when using a `train_file` that is a `csv`\r\n\r\n---\r\n\r\nAbout the actual issue, we can replicate it with the original example and dataset, but just changing \r\n```\r\n--model_name_or_path ./clip-roberta\r\n```\r\nfrom the example to \r\n```\r\n--model_name_or_path \"openai/clip-vit-base-patch32\"\r\n```\r\n(and data dir to point to where we store the dataset)\r\n\r\nFull command:\r\n```py\r\npython examples/pytorch/contrastive-image-text/run_clip.py \\\r\n --output_dir ./clip-finetuned \\\r\n --model_name_or_path \"openai/clip-vit-base-patch32\" \\\r\n --data_dir /scratch/amoryo/tmp/coco/data \\\r\n --dataset_name ydshieh/coco_dataset_script \\\r\n --dataset_config_name=2017 \\\r\n --image_column image_path \\\r\n --caption_column caption \\\r\n --remove_unused_columns=False \\\r\n --do_train --do_eval \\\r\n --per_device_train_batch_size=\"64\" \\\r\n --per_device_eval_batch_size=\"64\" \\\r\n --learning_rate=\"5e-5\" --warmup_steps=\"0\" --weight_decay 0.1 \\\r\n --overwrite_output_dir \\\r\n --push_to_hub\r\n```\r\n\r\nError:\r\n\r\n```bash\r\nTraceback (most recent call last):\r\n File \"/home/amoryo/sign-language/signwriting-clip/signwriting_clip/transformers/examples/pytorch/contrastive-image-text/run_clip.py\", line 590, in <module>\r\n main()\r\n File \"/home/amoryo/sign-language/signwriting-clip/signwriting_clip/transformers/examples/pytorch/contrastive-image-text/run_clip.py\", line 559, in main\r\n train_result = trainer.train(resume_from_checkpoint=checkpoint)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/transformers/trainer.py\", line 1534, in train\r\n return inner_training_loop(\r\n ^^^^^^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/transformers/trainer.py\", line 1860, in _inner_training_loop\r\n tr_loss_step = self.training_step(model, inputs)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/transformers/trainer.py\", line 2737, in training_step\r\n loss = self.compute_loss(model, inputs)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/transformers/trainer.py\", line 2760, in compute_loss\r\n outputs = model(**inputs)\r\n ^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1518, in _wrapped_call_impl\r\n return self._call_impl(*args, **kwargs)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1527, in _call_impl\r\n return forward_call(*args, **kwargs)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/transformers/models/clip/modeling_clip.py\", line 1108, in forward\r\n text_outputs = self.text_model(\r\n ^^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1518, in _wrapped_call_impl\r\n return self._call_impl(*args, **kwargs)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1527, in _call_impl\r\n return forward_call(*args, **kwargs)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/transformers/models/clip/modeling_clip.py\", line 691, in forward\r\n hidden_states = self.embeddings(input_ids=input_ids, position_ids=position_ids)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1518, in _wrapped_call_impl\r\n return self._call_impl(*args, **kwargs)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1527, in _call_impl\r\n return forward_call(*args, **kwargs)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File \"/data/amoryo/conda/envs/clip/lib/python3.11/site-packages/transformers/models/clip/modeling_clip.py\", line 219, in forward\r\n embeddings = inputs_embeds + position_embeddings\r\n ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~\r\nRuntimeError: The size of tensor a (128) must match the size of tensor b (77) at non-singleton dimension 1\r\n```", "Another issue, is that if following the guide to the letter, the README.md is generated with a local path as the basemodel\r\n\r\n```md\r\n---\r\nbase_model: /scratch/amoryo/models/signwriting-clip/reoberta-clip\r\n---\r\n```\r\nwhich is then rejected by the hub `\"base_model\" with value \"/scratch/amoryo/models/signwriting-clip/roberta-clip\" is not valid. Use a model id from https://hf.co/models.`\r\n", "cc @ydshieh ", "I try to change this, and somehow it works.\r\n\r\n\r\n```\r\ntext_inputs = tokenizer(\r\n captions, max_length=data_args.max_seq_length, padding=\"max_length\", truncation= True)\r\n```\r\n\r\nto\r\n\r\n```\r\ntext_inputs = tokenizer(\r\n captions, max_length=data_args.max_seq_length, padding=True, truncation=True)\r\n```", "> /scratch/amoryo/models/signwriting-clip/reoberta-clip\r\n\r\nI don't have full context. But the README use `clip-roberta`, so make sure you created the local directory and using the corresponding name in the command line to launch the training (in your case, `roberta-clip`)", "You are loading `openai/clip-vit-base-patch32`, so you are likely using a `CLIPTokenizer` which has `max_position_embeddings=77`. The issue could be also resolved if you specify `max_seq_length=77` in the commandline when launching the training.", "Thank you for the `max_seq_length` note.\r\n\r\nWhat about https://github.com/huggingface/transformers/issues/28286#issuecomment-1881147267 ? This is indeed the correct local path for the base model, but when pushed to the hub, it gives an error (if I remove this line from the README, no error)", "Could you provide the full command line you used that will fail when push to hub?", "Full command here: https://github.com/huggingface/transformers/issues/28286#issuecomment-1874400979\r\nAt the end of training, it creates a `README.md` file, as described here https://github.com/huggingface/transformers/issues/28286#issuecomment-1881147267 which fails to be pushed to the hub (alongside everything else)\r\nYou can easily replicate it by running the model for 10 steps, or creating a directory with `README.md` as described, and `huggingface-cli upload me/my-model README.md`", "Another issue: it seems like the data does not shuffle, but shuffling is very important for a clip-like model\r\n\r\n<img width=\"100%\" alt=\"image\" src=\"https://github.com/huggingface/transformers/assets/5757359/fab86795-5062-46e1-93f9-92db25859ff7\">\r\n", "I confirmed that the issue of `base_model` is reproducible. Thanks a lot - will fix it!", "BTW, I get\r\n\r\n> \"base_model\" with value \"./clip-roberta\" is not valid. Use a model id from https://hf.co/models.\r\n\r\nBut you showed\r\n\r\n> \"/scratch/amoryo/models/signwriting-clip/roberta-clip\" is not valid. Use a model id from https://hf.co/models.\r\n\r\nJust to make sure, are you specifying, in the command line, the absolute path to the local model directory `roberta-clip`?", "yes, i am specifying the absolute path", "The `base_model` issue is addressed in #28482", "Regarding shuffling, it's a bit hidden, but if you set a breakpoint at\r\n\r\nhttps://github.com/huggingface/transformers/blob/edb314ae2ba4ac0e89d6a31d48037b8943978bff/src/transformers/trainer.py#L776\r\n\r\nyou will see `RandomSampler` is used, so we are fine.", "I guess that's everything. Thanks so much! Feel free to close once #28482 is one\r\n\r\nI still find the training loss periodicity puzzling, but i have no idea. it also happens with a different base model \r\n\r\n", "Fixed in #28482 as mentioned earlier" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28286/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "chedatomasz", "id": 47541285, "node_id": "MDQ6VXNlcjQ3NTQxMjg1", "avatar_url": "https://avatars.githubusercontent.com/u/47541285?v=4", "gravatar_id": "", "url": "https://api.github.com/users/chedatomasz", "html_url": "https://github.com/chedatomasz", "followers_url": "https://api.github.com/users/chedatomasz/followers", "following_url": "https://api.github.com/users/chedatomasz/following{/other_user}", "gists_url": "https://api.github.com/users/chedatomasz/gists{/gist_id}", "starred_url": "https://api.github.com/users/chedatomasz/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/chedatomasz/subscriptions", "organizations_url": "https://api.github.com/users/chedatomasz/orgs", "repos_url": "https://api.github.com/users/chedatomasz/repos", "events_url": "https://api.github.com/users/chedatomasz/events{/privacy}", "received_events_url": "https://api.github.com/users/chedatomasz/received_events", "type": "User", "site_admin": false }

[ { "id": 1843244711, "node_id": "MDU6TGFiZWwxODQzMjQ0NzEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/New%20model", "name": "New model", "color": "fbca04", "default": false, "description": "" } ]

[]

[]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28285/reactions", "total_count": 2, "+1": 2, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "levidehaan", "id": 304932, "node_id": "MDQ6VXNlcjMwNDkzMg==", "avatar_url": "https://avatars.githubusercontent.com/u/304932?v=4", "gravatar_id": "", "url": "https://api.github.com/users/levidehaan", "html_url": "https://github.com/levidehaan", "followers_url": "https://api.github.com/users/levidehaan/followers", "following_url": "https://api.github.com/users/levidehaan/following{/other_user}", "gists_url": "https://api.github.com/users/levidehaan/gists{/gist_id}", "starred_url": "https://api.github.com/users/levidehaan/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/levidehaan/subscriptions", "organizations_url": "https://api.github.com/users/levidehaan/orgs", "repos_url": "https://api.github.com/users/levidehaan/repos", "events_url": "https://api.github.com/users/levidehaan/events{/privacy}", "received_events_url": "https://api.github.com/users/levidehaan/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hi @levidehaan, thanks for raising this issue! \r\n\r\nCould you provide a checkpoint for a public model which replicates this issue? Does this happen for any llama model? \r\nDo you observe this issue when not running through a flask app? i.e. does this work: \r\n\r\n```py\r\nimport torch\r\nfrom transformers import AutoModelForCausalLM, AutoTokenizer\r\nimport time\r\nimport logging\r\nimport os\r\nprint(torch.version.cuda)\r\n\r\nmodellocation = \"/home/levi/projects/text-generation-webui/models/Upstage_SOLAR-10.7B-Instruct-v1.0\"\r\n\r\ntokenizer = AutoTokenizer.from_pretrained(\"/home/levi/projects/text-generation-webui/models/Upstage_SOLAR-10.7B-Instruct-v1.0\")\r\n\r\nmodel = AutoModelForCausalLM.from_pretrained(\r\n modellocation,\r\n device_map=\"auto\"\r\n)\r\n\r\nstart_time = time.time()\r\ndata = # fill in data here\r\nconversation = data['messages']\r\nprompt = tokenizer.apply_chat_template(\r\n\tconversation, \r\n\ttokenize=False, \r\n\tadd_generation_prompt=True\r\n)\r\ninputs = tokenizer(prompt, return_tensors=\"pt\").to(model.device)\r\noutputs = model.generate(**inputs, use_cache=True, max_length=4096)\r\noutputs = model.generate(**inputs, use_cache=True, max_length=4096)\r\noutput_text = tokenizer.decode(outputs[0])\r\njson_output = jsonify(\r\n\t{'choices': [{'message': {'role': 'assistant', 'content': output_text}}]}\r\n)\r\nrequest_duration = time.time() - request.start_time\r\nprint(f\"Request took {request_duration} seconds\")\r\nprint(json_output)\r\n\r\n```\r\nBased on the error message, I'd be willing to bet it's an indexing issue with `position_ids`. \r\n\r\ncc @gante as this covers rope, cuda and generate :) ", "@levidehaan Is there any follow-up on this issue? I'm encountering exactly the same issue happened at the same line of the LLaMA-2 inferrence.", "Hi @BiEchi 👋 we can look further into the issue as soon as we have a short stand-alone reproducible script. Otherwise, it will next to impossible to reproduce the bug and, consequently, find out what's wrong :)", "Hi @gante, my code is\r\n```\r\nfrom transformers import AutoModelForCausalLM, AutoTokenizer\r\nfrom transformers import BitsAndBytesConfig\r\nimport torch\r\nprint(torch.version.cuda)\r\n\r\nllama_path = \"llama-2-7b-hf\"\r\ntokenizer = AutoTokenizer.from_pretrained(llama_path)\r\nmodel = AutoModelForCausalLM.from_pretrained(llama_path, device_map='auto')\r\n\r\nprompt = \"Hello World!\"\r\ninputs = tokenizer(prompt, return_tensors=\"pt\").to(model.device)\r\nprint(inputs)\r\nwith torch.inference_mode():\r\n logits = model(**inputs).logits[0]\r\nprint(logits)\r\n```\r\n\r\nand the error is\r\n\r\n```\r\n$ NCCL_P2P_DISABLE=1 CUDA_VISIBLE_DEVICES=0,1 CUDA_LAUNCH_BLOCKING=1 python test.py\r\n11.3\r\nLoading checkpoint shards: 100%|█████████████████████████████████████████████████████████████████████████████| 3/3 [00:08<00:00, 2.71s/it]\r\n{'input_ids': tensor([[ 1, 15043, 2787, 29991]], device='cuda:0'), 'attention_mask': tensor([[1, 1, 1, 1]], device='cuda:0')}\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [0,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [1,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [2,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [3,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [4,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [5,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [6,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [7,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [8,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [9,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [10,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [11,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [12,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [13,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [14,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [15,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [16,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [17,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [18,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [19,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [20,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [21,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [22,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [23,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [24,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [25,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [26,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [27,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [28,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [29,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [30,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [31,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [32,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [33,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [34,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [35,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [36,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [37,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [38,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [39,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [40,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [41,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [42,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [43,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [44,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [45,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [46,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [47,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [48,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [49,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [50,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [51,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [52,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [53,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [54,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [55,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [56,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [57,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [58,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [59,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [60,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [61,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [62,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [63,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [64,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [65,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [66,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [67,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [68,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [69,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [70,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [71,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [72,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [73,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [74,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [75,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [76,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [77,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [78,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [79,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [80,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [81,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [82,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [83,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [84,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [85,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [86,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [87,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [88,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [89,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [90,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [91,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [92,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [93,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [94,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [95,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [96,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [97,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [98,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [99,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [100,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [101,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [102,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [103,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [104,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [105,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [106,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [107,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [108,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [109,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [110,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [111,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [112,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [113,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [114,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [115,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [116,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [117,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [118,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [119,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [120,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [121,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [122,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [123,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [124,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [125,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [126,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\n../aten/src/ATen/native/cuda/IndexKernel.cu:91: operator(): block: [0,0,0], thread: [127,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && \"index out of bounds\"` failed.\r\nTraceback (most recent call last):\r\nsite-packages/transformers/models/llama/modeling_llama.py\", line 232, in apply_rotary_pos_emb\r\n cos = cos[position_ids].unsqueeze(unsqueeze_dim)\r\nRuntimeError: CUDA error: device-side assert triggered\r\n```", "1. It works on a single GPU, but triggered the error above on multi-GPUs (i.e. we pass multiple visible devices **and** use `device_map=\"auto\"`).\r\n3. It works on another of our machine.\r\n4. We have to set `NCCL_P2P_DISABLE=1` to make multi-GPU work, due to some underlying machine issues.", "@muellerzr this seems accelerate-related (multi-device) -- do you know what might be wrong?" ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28284/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "sr5434", "id": 118690585, "node_id": "U_kgDOBxMTGQ", "avatar_url": "https://avatars.githubusercontent.com/u/118690585?v=4", "gravatar_id": "", "url": "https://api.github.com/users/sr5434", "html_url": "https://github.com/sr5434", "followers_url": "https://api.github.com/users/sr5434/followers", "following_url": "https://api.github.com/users/sr5434/following{/other_user}", "gists_url": "https://api.github.com/users/sr5434/gists{/gist_id}", "starred_url": "https://api.github.com/users/sr5434/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/sr5434/subscriptions", "organizations_url": "https://api.github.com/users/sr5434/orgs", "repos_url": "https://api.github.com/users/sr5434/repos", "events_url": "https://api.github.com/users/sr5434/events{/privacy}", "received_events_url": "https://api.github.com/users/sr5434/received_events", "type": "User", "site_admin": false }

[]

[]

[ "It turns out the kernel was busy, but checkpoints were still being saved." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28283/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "Marwen94", "id": 36446303, "node_id": "MDQ6VXNlcjM2NDQ2MzAz", "avatar_url": "https://avatars.githubusercontent.com/u/36446303?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Marwen94", "html_url": "https://github.com/Marwen94", "followers_url": "https://api.github.com/users/Marwen94/followers", "following_url": "https://api.github.com/users/Marwen94/following{/other_user}", "gists_url": "https://api.github.com/users/Marwen94/gists{/gist_id}", "starred_url": "https://api.github.com/users/Marwen94/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Marwen94/subscriptions", "organizations_url": "https://api.github.com/users/Marwen94/orgs", "repos_url": "https://api.github.com/users/Marwen94/repos", "events_url": "https://api.github.com/users/Marwen94/events{/privacy}", "received_events_url": "https://api.github.com/users/Marwen94/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hi @marwen94, I'm struggling to reproduce this here. I'm not familiar with the Poetry dependency manager, but I tried reproducing those package versions in a Python 3.9 env with `pip` and the model loaded fine - the issue seems to be very specific to the environment you're using.\r\n\r\nCan you figure out a list of package versions that I can install with `conda` or `pip` that reproduces the issue? Once I can reproduce it here I can work on diagnosing and fixing it!", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28282/reactions", "total_count": 1, "+1": 1, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "unography", "id": 5240449, "node_id": "MDQ6VXNlcjUyNDA0NDk=", "avatar_url": "https://avatars.githubusercontent.com/u/5240449?v=4", "gravatar_id": "", "url": "https://api.github.com/users/unography", "html_url": "https://github.com/unography", "followers_url": "https://api.github.com/users/unography/followers", "following_url": "https://api.github.com/users/unography/following{/other_user}", "gists_url": "https://api.github.com/users/unography/gists{/gist_id}", "starred_url": "https://api.github.com/users/unography/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/unography/subscriptions", "organizations_url": "https://api.github.com/users/unography/orgs", "repos_url": "https://api.github.com/users/unography/repos", "events_url": "https://api.github.com/users/unography/events{/privacy}", "received_events_url": "https://api.github.com/users/unography/received_events", "type": "User", "site_admin": false }

[]

[]

[ "Hi @unography, thanks for raising this issue. \r\n\r\nThis appears to be a `poetry` releated issue - `transformers` can be successfully installed with `pip`. Did you run `pip wheel --use-pep517 \"flash-attn (==2.1.0)\"` as recommended by the error message? \r\n\r\nNote, flash-attn is needed if using flash attention for a model. However, it's not required, you can use 'regular' attention for our models, and as such it's not listed as a requirement in our `setup.py`. " ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28281/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "loveunk", "id": 1678567, "node_id": "MDQ6VXNlcjE2Nzg1Njc=", "avatar_url": "https://avatars.githubusercontent.com/u/1678567?v=4", "gravatar_id": "", "url": "https://api.github.com/users/loveunk", "html_url": "https://github.com/loveunk", "followers_url": "https://api.github.com/users/loveunk/followers", "following_url": "https://api.github.com/users/loveunk/following{/other_user}", "gists_url": "https://api.github.com/users/loveunk/gists{/gist_id}", "starred_url": "https://api.github.com/users/loveunk/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/loveunk/subscriptions", "organizations_url": "https://api.github.com/users/loveunk/orgs", "repos_url": "https://api.github.com/users/loveunk/repos", "events_url": "https://api.github.com/users/loveunk/events{/privacy}", "received_events_url": "https://api.github.com/users/loveunk/received_events", "type": "User", "site_admin": false }

[]

[]

[ "tried to run the following code, stuck with backend `nccl`, but works with `gloo`:\r\n\r\n```python\r\nimport torch.distributed as dist\r\nimport argparse\r\nimport torch\r\n\r\ntorch.cuda.set_device(int(os.environ['LOCAL_RANK']))\r\ndevice = torch.device(\"cuda\", int(os.environ['LOCAL_RANK']))\r\n\r\ndist.init_process_group(\"nccl\") \r\ndist.all_reduce(torch.ones(1).to(device), op=dist.ReduceOp.SUM)\r\n```\r\n\r\nrun command: \r\n```shell\r\npython -m torch.distributed.launch --nproc_per_node=2 test.py\r\n```", "It looks like there are some NCCL issues with my GPU server.\r\nThe training works after turning off the GPU P2P communication with `export NCCL_P2P_DISABLE=1`.", "FYI: https://blog.csdn.net/qq_40947610/article/details/128118180", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md) are likely to be ignored." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28280/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "login": "Titus-von-Koeller", "id": 9048635, "node_id": "MDQ6VXNlcjkwNDg2MzU=", "avatar_url": "https://avatars.githubusercontent.com/u/9048635?v=4", "gravatar_id": "", "url": "https://api.github.com/users/Titus-von-Koeller", "html_url": "https://github.com/Titus-von-Koeller", "followers_url": "https://api.github.com/users/Titus-von-Koeller/followers", "following_url": "https://api.github.com/users/Titus-von-Koeller/following{/other_user}", "gists_url": "https://api.github.com/users/Titus-von-Koeller/gists{/gist_id}", "starred_url": "https://api.github.com/users/Titus-von-Koeller/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/Titus-von-Koeller/subscriptions", "organizations_url": "https://api.github.com/users/Titus-von-Koeller/orgs", "repos_url": "https://api.github.com/users/Titus-von-Koeller/repos", "events_url": "https://api.github.com/users/Titus-von-Koeller/events{/privacy}", "received_events_url": "https://api.github.com/users/Titus-von-Koeller/received_events", "type": "User", "site_admin": false }

[]

{ "login": "younesbelkada", "id": 49240599, "node_id": "MDQ6VXNlcjQ5MjQwNTk5", "avatar_url": "https://avatars.githubusercontent.com/u/49240599?v=4", "gravatar_id": "", "url": "https://api.github.com/users/younesbelkada", "html_url": "https://github.com/younesbelkada", "followers_url": "https://api.github.com/users/younesbelkada/followers", "following_url": "https://api.github.com/users/younesbelkada/following{/other_user}", "gists_url": "https://api.github.com/users/younesbelkada/gists{/gist_id}", "starred_url": "https://api.github.com/users/younesbelkada/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/younesbelkada/subscriptions", "organizations_url": "https://api.github.com/users/younesbelkada/orgs", "repos_url": "https://api.github.com/users/younesbelkada/repos", "events_url": "https://api.github.com/users/younesbelkada/events{/privacy}", "received_events_url": "https://api.github.com/users/younesbelkada/received_events", "type": "User", "site_admin": false }

[ { "login": "younesbelkada", "id": 49240599, "node_id": "MDQ6VXNlcjQ5MjQwNTk5", "avatar_url": "https://avatars.githubusercontent.com/u/49240599?v=4", "gravatar_id": "", "url": "https://api.github.com/users/younesbelkada", "html_url": "https://github.com/younesbelkada", "followers_url": "https://api.github.com/users/younesbelkada/followers", "following_url": "https://api.github.com/users/younesbelkada/following{/other_user}", "gists_url": "https://api.github.com/users/younesbelkada/gists{/gist_id}", "starred_url": "https://api.github.com/users/younesbelkada/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/younesbelkada/subscriptions", "organizations_url": "https://api.github.com/users/younesbelkada/orgs", "repos_url": "https://api.github.com/users/younesbelkada/repos", "events_url": "https://api.github.com/users/younesbelkada/events{/privacy}", "received_events_url": "https://api.github.com/users/younesbelkada/received_events", "type": "User", "site_admin": false } ]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/transformers/pr_28279). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

{ "url": "https://api.github.com/repos/huggingface/transformers/issues/28279/reactions", "total_count": 0, "+1": 0, "-1": 0, "laugh": 0, "hooray": 0, "confused": 0, "heart": 0, "rocket": 0, "eyes": 0 }

{ "url": "https://api.github.com/repos/huggingface/transformers/pulls/28279", "html_url": "https://github.com/huggingface/transformers/pull/28279", "diff_url": "https://github.com/huggingface/transformers/pull/28279.diff", "patch_url": "https://github.com/huggingface/transformers/pull/28279.patch", "merged_at": null }