Datasets:

---

DanielPFlorian

/

Transformers-Github-Issues

like 0

[ { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "Hello! Thank you for your proposal. This, however, sounds like a colossal amount of work for limited gain - we try to keep the breaking changes across versions to a minimum. I understand that these do happen from time to time, but only for very good reasons that are thoroughly discussed beforehand.\r\n\r\nWould you be able to share what breaking changes have impacted you, and have been a bit too hard to overcome/not documented enough, preventing you from upgrading? Understanding this will help us do better in the future.", "This issue has been automatically marked as stale and been closed because it has not had recent activity. Thank you for your contributions.\n\nIf you think this still needs to be addressed please comment on this thread." ]

[]

[]

[ "Hey @Dmitry-Sn,\r\n\r\nIt would be great if you could ask these kind of questions on the forum: https://discuss.huggingface.co/ . We try to keep github for issues and less for user-specific use cases. Thanks!", "> \r\n\r\nHi, have you found some way to solve this problem? I meet the same situation, some proper nouns can' t translated properly, and I want to keep it as its native format.", "Hi @vpegasus! There were not great ideas, generally.\r\nAs far as I remember, I decided to try to finetune the model for a special token. In my task, it was simple - since I wanted to keep geographical names in the original language, I replaced only them in the training data with a special token (I didn't have time to check the effectiveness, since I left that company).\r\nColleagues also suggested a solution with special characters (it seems to separate the word with <> characters), but it worked poorly.\r\nAll this applies to the models on MarianMT." ]

[ { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "This issue has been automatically marked as stale and been closed because it has not had recent activity. Thank you for your contributions.\n\nIf you think this still needs to be addressed please comment on this thread." ]

[ { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "Hi, could you please provide your environment info as asked in the template? `transformers-cli env`", "hey, any solution yet?", "@LysandreJik I am having the same issue and this is my env:\r\n- `transformers` version: 4.1.0.dev0\r\n- Platform: Linux-5.4.0-1029-gcp-x86_64-with-debian-buster-sid\r\n- Python version: 3.7.6\r\n- PyTorch version (GPU?): not installed (NA)\r\n- Tensorflow version (GPU?): 2.3.1 (False)\r\n- Using GPU in script?: NO (using TPU)\r\n- Using distributed or parallel set-up in script?: ", "I got a similar error trying to run `run_clm.py` on a TPU.\r\n` File \"/kaggle/working/transformers/examples/language-modeling/run_clm.py\", line 33, in <module>\r\n from transformers import (\r\nImportError: cannot import name 'MODEL_FOR_CAUSAL_LM_MAPPING' from 'transformers' (/opt/conda/lib/python3.7/site-packages/transformers/__init__.py)`", "@Clickative any solution?\r\n", "It seems you do not have PyTorch installed? `run_clm.py` is a PyTorch script.", "I encountered this same error and followed the advice from @LysandreJik. I installed PyTorch using `pip3 install torch torchvision` and this resolved the issue.", "On Kaggle TPU, the current docker seems to have old version of transformers and it reads from the conda environment, so new installs are not taken into account (try `pip show`). Even if docker is set to latest. The GPU docker seem to run a version with 0.9.3 of tokenizers and latest transformers need 0.9.4 which is another issue.", "This issue has been automatically marked as stale and been closed because it has not had recent activity. Thank you for your contributions.\n\nIf you think this still needs to be addressed please comment on this thread." ]

[]

[]

[ "Hey @BeerTai, \r\n\r\nIt would be great if you could ask these kind of questions on the forum: https://discuss.huggingface.co/ . We try to keep github for issues and less for user-specific use cases. Thanks!" ]

[ { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "Hey @trungtruc123, \r\n\r\ncan you try upgrading your transformer version?\r\nThe following code snippet (as stated on the model card: https://github.com/VinAIResearch/PhoBERT) works perfectly fine for me.\r\n\r\n```python\r\nimport torch\r\nfrom transformers import AutoModel, AutoTokenizer\r\n\r\nphobert = AutoModel.from_pretrained(\"vinai/phobert-base\")\r\n\r\n# For transformers v4.x+: \r\ntokenizer = AutoTokenizer.from_pretrained(\"vinai/phobert-base\", use_fast=False)\r\n```\r\n\r\nVersion:\r\n- `transformers` version: 4.1.0.dev0\r\n- Platform: Linux-5.4.0-1030-gcp-x86_64-with-glibc2.10\r\n- Python version: 3.8.5\r\n- PyTorch version (GPU?): 1.8.0.dev20201117 (True)\r\n- Tensorflow version (GPU?): 2.3.1 (False)\r\n- Using GPU in script?: <fill in>\r\n- Using distributed or parallel set-up in script?: <fill in>\r\n\r\n\r\nIt should also work with transformers 4.0.0.", "This issue has been automatically marked as stale and been closed because it has not had recent activity. Thank you for your contributions.\n\nIf you think this still needs to be addressed please comment on this thread." ]

[]

[]

[]

[ { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "Hey @rabeehk,\r\n\r\ncould you attach a code snippet that we can copy-paste to reproduce the error? Thanks!", "Hi Patrick, \r\nI checked and in finetune_trainer.py you consider these parameters only which are all of type float:\r\n` extra_model_params = (\"encoder_layerdrop\", \"decoder_layerdrop\", \"dropout\", \"attention_dropout\") \r\n`\r\nIn this case the issue would not happen, but if one of these parameters were boolean, lets say \"A\", if the user pass a config file like below to `finetune_trainer.py`\r\n\r\n```\r\n//config.json\r\n{\r\nA: false \r\n}\r\n```\r\n\r\nand if the code tries to updating the value of \"A\" in the config file as line 162, then this introduces the bug of not setting A.\r\nFor now since the variables you consider are float, this wont happen, so please feel free to close the bug. Still safer to change 162 with` if hasattr(training_args, p):`\r\nthanks. \r\n\r\n", "This issue has been automatically marked as stale and been closed because it has not had recent activity. Thank you for your contributions.\n\nIf you think this still needs to be addressed please comment on this thread." ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "Sounds right, will fix." ]

[]

[]

[]

[]

[]

[]

[ { "id": 1834052129, "node_id": "MDU6TGFiZWwxODM0MDUyMTI5", "url": "https://api.github.com/repos/huggingface/transformers/labels/High-Level%20feature", "name": "High-Level feature", "color": "f7c9a3", "default": false, "description": "" }, { "id": 1834067346, "node_id": "MDU6TGFiZWwxODM0MDY3MzQ2", "url": "https://api.github.com/repos/huggingface/transformers/labels/Documentation", "name": "Documentation", "color": "77cc3b", "default": false, "description": "" }, { "id": 1838412367, "node_id": "MDU6TGFiZWwxODM4NDEyMzY3", "url": "https://api.github.com/repos/huggingface/transformers/labels/model%20card", "name": "model card", "color": "92d5f4", "default": false, "description": "Related to pretrained model cards" }, { "id": 2139563322, "node_id": "MDU6TGFiZWwyMTM5NTYzMzIy", "url": "https://api.github.com/repos/huggingface/transformers/labels/cleanup", "name": "cleanup", "color": "e7fc49", "default": false, "description": "" } ]

[]

[ "@patrickvonplaten, unless we erase them from the history, it won't make git clone any faster.", "@sgugger it might make a `git checkout` slightly faster. I don't think the model cards were ever an issue in terms of git performance though (8% of number of files in the repo, 7% of total size of the repo)\r\n\r\n", "https://github.com/github/git-sizer is an awesome tool by the way\r\n\r\n```\r\n$ git-sizer\r\nProcessing blobs: 22486 \r\nProcessing trees: 25585 \r\nProcessing commits: 7749 \r\nMatching commits to trees: 7749 \r\nProcessing annotated tags: 33 \r\nProcessing references: 216 \r\n| Name | Value | Level of concern |\r\n| ---------------------------- | --------- | ------------------------------ |\r\n| Biggest checkouts | | |\r\n| * Maximum path length [1] | 135 B | * |\r\n\r\n[1] 030c0d2cdc80cf8dcf23a6ee55c20e979548a181 (refs/heads/master^{tree})\r\n```\r\n\r\nvs on datasets (cc @lhoestq):\r\n```\r\n$ git-sizer\r\nProcessing blobs: 9449 \r\nProcessing trees: 12905 \r\nProcessing commits: 1245 \r\nMatching commits to trees: 1245 \r\nProcessing annotated tags: 16 \r\nProcessing references: 110 \r\n| Name | Value | Level of concern |\r\n| ---------------------------- | --------- | ------------------------------ |\r\n| Biggest objects | | |\r\n| * Blobs | | |\r\n| * Maximum size [1] | 17.8 MiB | * |\r\n| | | |\r\n| Biggest checkouts | | |\r\n| * Number of directories [2] | 4.89 k | ** |\r\n| * Maximum path depth [3] | 16 | * |\r\n| * Maximum path length [3] | 231 B | ** |\r\n\r\n[1] 3fe8eaab7a337ea2a8d06daa5721fc5935ba3098 (75cafce7677d6f66c49c34e43cfbc425e1f50d30:datasets/anli/dummy/plain_text/0.1.0/dummy_data.zip)\r\n[2] 681c565a3d8f63535823a1d33438c2b76ba3c706 (refs/heads/master^{tree})\r\n[3] c94ea70f34be4ed2723fc1c647340792ba03879c (7cd045237bb77f3b32877d31aae87789ec57ffab^{tree})\r\n```", "**Update**: I deployed the new buttons/call to actions on the model pages. \r\nI also created a new Forum topic (@sgugger @Pierrci @patrickvonplaten) titled \"Model cards\" where users can suggest edits or creations of existing model cards, in case they don't have write access to the corresponding model repo:\r\n\r\nhttps://discuss.huggingface.co/t/about-the-model-cards-category/2777", "All out-standing model card PRs were merged. No more model cards PRs expected!\r\n\r\nWill migrate existing ones now." ]

[]

[]

[ "Hi! Do you want to submit a PR?", "Sure, I'll be happy to do it. I need permission... `remote: Permission to huggingface/transformers.git denied to NatLun137.`", "Hmmm I think you just tried to push on `huggingface/transformers`? You should fork the repo, apply your changes there and then open a PR here. I see you created your fork already, how did you open a PR then? Did you use the GitHub UI?" ]

[ { "id": 1838412367, "node_id": "MDU6TGFiZWwxODM4NDEyMzY3", "url": "https://api.github.com/repos/huggingface/transformers/labels/model%20card", "name": "model card", "color": "92d5f4", "default": false, "description": "Related to pretrained model cards" } ]

[]

[]

[]

[]

[ "another bit - `examples/seq2seq/test_data` is also used by `research_projects/seq2seq` - perhaps symlink? ", "Maybe a hard copy in that case, just in case the data changes/moves on the examples side.", "> Maybe a hard copy in that case, just in case the data changes/moves on the examples side.\r\n\r\nThen need to check which specific sub-dirs are needed - if I'm not mistaken it's only `test_data/wmt_en_ro/`\r\n\r\nI'd still use a symlink to avoid git repo bloat and this can always be easily fixed if there is a divergence down the road.", "Does this mean we no longer explicitly support pytorch_lightning? ", "> Does this mean we no longer explicitly support pytorch_lightning?\r\n\r\nI would rather like it if we ditch the custom transformer trainer and just use lightning." ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "@ayushtiku5 I tweeted about it and tagged you on the tweet: https://twitter.com/PatrickPlaten/status/1336681238485229568 - hope that's fine for you :-) ", "@ayushtiku5 can you please check if `HammingDiversityLogitsProcessor` and `PrefixConstrainedLogitsProcessor` can be speeded up with functions like `torch.scatter`, `torch.gather`, `torch.masked_fill`, `torch.index_fill`, `torch.index_add`, `torch.index_copy`? I believe there is room for improvement in \r\nhttps://github.com/huggingface/transformers/blob/master/src/transformers/generation_logits_process.py#L406-L409 and https://github.com/huggingface/transformers/blob/master/src/transformers/generation_logits_process.py#L471 (in the way like https://github.com/huggingface/transformers/pull/9557 and https://github.com/huggingface/transformers/pull/9600) but I have no experience using these processors to create good enough examples for speed testing and corner cases research." ]

[ { "id": 1838412367, "node_id": "MDU6TGFiZWwxODM4NDEyMzY3", "url": "https://api.github.com/repos/huggingface/transformers/labels/model%20card", "name": "model card", "color": "92d5f4", "default": false, "description": "Related to pretrained model cards" } ]

[]

[]

[ { "id": 2669577093, "node_id": "MDU6TGFiZWwyNjY5NTc3MDkz", "url": "https://api.github.com/repos/huggingface/transformers/labels/PR%20for%20Model%20Addition", "name": "PR for Model Addition", "color": "5319e7", "default": false, "description": "" } ]

[]

[ "@patrickvonplaten Never mind, just use this PR is ok. I am ok if our work can be merged into the master quickly. ", "@jplu @LysandreJik @sgugger , we all gave our thumbs-up in the old PR. It's a bit unfortunate that the authorship is slightly changed here, but the PR should be read to merge. ", "Squashed commit & cherry picked on the `master` branch so that the authorship is kept in df2af6d. Closing.", "@StillKeepTry thanks a lot for all of your hard work on this PR! Glad to welcome MPNet in the library!", "Thanks every reviewer for helping review our work. :)" ]

[]

[]

[ "After an offline discussion we decided to proceed differently. Then closing this PR." ]

[]

[]

[ "@jplu Thanks so much for your kind reviews! I will improve the code as you suggested.\r\n@patrickvonplaten I have confirmed that my TF and Pytorch have equivalent `generate` output on `num_beam=1` (greedy) on all (15) test cases . \r\n\r\nNevertheless, I just confirmed that the labels in the official test file is based on beam search where `num_beam=4` and this TFRag does not have `beam_search` yet since I am not sure whether I should wait for `tf_generation` refactor . \r\n\r\nIn the current Pytorch test, if explicitly set `num_beam=1` we will get exact same result as my TF implementation.\r\n\r\nSo to pass the current official test completely I will have to adapt `beam_search` into TFRag which I will try :)\r\nFor now, I slightly modify test cases to match Pytorch greedy output (we can revert back later when I finish `beam_search` )", "This is very much WIP -> need some more time for the `save/from_pretrained()`", "**UPDATED** Dec, 23 2020 : Finish `TFRagSequenceForGeneration` \r\n\r\n---\r\nSorry that I forgot to mention that the latest updated is still in very WIP (phase-2, after phase-1 which is core-part of `TFRagModel` and `TFRagToken` ), so not ready for review .\r\n\r\nAs discussed with Patrick, there is still an \"output_truncation\" issue (perhaps due to Caching & TFBart refactor) that Patrick will help take a look. \r\n\r\n**We will have to finish in Phase-2** (reason of TFRagToken tests fail at the moment ) :\r\n[ x] - bugs in the new test `test_rag_token_inference_nq_checkpoint() `\r\n[ x] - `beam_search` of `TFRagTokenForGeneration` and \r\n[ x ] - `TFRagSequenceForGeneration` <-- finished!\r\n [ ] - Apply jplu comments to clean up the code\r\n\r\nwhich will take some more time :) :) ", "Hi @jplu , @patrickvonplaten , regarding the graph issue, I just successfully made a[ colab notebook](https://colab.research.google.com/drive/1s-j9PB9yzrFsL6q5rZUQyf8_Lt6jDAkL?usp=sharing) capable of training `TFRagSequenceForGeneration` in graph mode using `context_input_ids` as inputs (instead of `input_ids` ) ... \r\nHopefully this is a reasonable work around on TFRag training.\r\n\r\nNow `TFRagToken` has the same output_truncation [issue as Pytorch 's 9098](https://github.com/huggingface/transformers/pull/9098) . If Patrick help me solve this, I will be able to finish the beam search ... which should finish all the main parts of TFRag ... \r\n\r\n(After this, the remaining is to clean codes as suggested by you guys, complete all fast tests, and fix the ugly hacks [ ie. pass `test_rag_token_inference_nq_checkpoint` without hacking ] )", "Hey @ratthachat,\r\n\r\ngreat work! I'll take care of the new (and hopefully last :D) `save_/from_pretrained()` issue today. And I'll also make sure that `TFRagTokenGeneration` works properly for `greedy_search`! I'll leave `beam_search` then for you :-) ", "Okey, I fixed the `from_pretrained` bug when `from_pt=True` and also fixed `greedy search` for `TFRagToken`. Thanks for the very in-detail error descriptions @ratthachat. \r\n\r\nI think the only thing left to do for now is to correctly implement `beam_search` for `TFRagToken`. As I understood it, you'd like to give it a try. Let me know if you want me to tackle this or if you need help :-)\r\n\r\nI extended your \"load from pt\" test slightly to make sure that weights are now always (I really hope so :D) correctly loaded. Also, I added one test for `greedy_search` for `RagToken`", "Hi @patrickvonplaten , I agree and all T5 related tests are deleted.\r\n\r\nTests related to TFBart are still not passed which likely due to TFBart bug.\r\nMore precisely, TFBart vs. Torch's Bart forward pass return `generator_enc_last_hidden_state.shape`\r\ndifferently if we provide `decoder_input_ids`\r\n\r\n```\r\ntexts = \"My friends are cool but they eat too many carbs. I really want them to be healthy, so I buy them vegetable.\"\r\ntexts2 = \"My friends are cool.\"\r\ninputs = tokenizer([texts], max_length=1024, return_tensors='tf')\r\ninputs2 = tokenizer([texts2], max_length=1024, return_tensors='tf')\r\n\r\ninput_ids=inputs['input_ids']\r\ninput_ids2=inputs2['input_ids']\r\nout = model(input_ids,decoder_input_ids=None)\r\nprint(out.encoder_last_hidden_state.shape) # RETURN (1, 27, 1024)\r\n\r\nout = model(input_ids,decoder_input_ids=input_ids2)\r\nprint(out.encoder_last_hidden_state.shape) # RETURN (1, 7, 1024)\r\n\r\n```\r\n\r\nIf we run the same snippet in Pytorch, they will both return `(1, 27, 1024)` . (tested on both official released and master)\r\nSo likely this is TFBart bug, **and it makes 4 TFRag fast-tests fail**. \r\n(shape of `out.encoder_last_hidden_state.shape` is not as expected)", "> Hi @patrickvonplaten , I agree and all T5 related tests are deleted.\r\n> \r\n> Tests related to TFBart are still not passed which likely due to TFBart bug.\r\n> More precisely, TFBart vs. Torch's Bart forward pass return `generator_enc_last_hidden_state.shape`\r\n> differently if we provide `decoder_input_ids`\r\n> \r\n> ```\r\n> texts = \"My friends are cool but they eat too many carbs. I really want them to be healthy, so I buy them vegetable.\"\r\n> texts2 = \"My friends are cool.\"\r\n> inputs = tokenizer([texts], max_length=1024, return_tensors='tf')\r\n> inputs2 = tokenizer([texts2], max_length=1024, return_tensors='tf')\r\n> \r\n> input_ids=inputs['input_ids']\r\n> input_ids2=inputs2['input_ids']\r\n> out = model(input_ids,decoder_input_ids=None)\r\n> print(out.encoder_last_hidden_state.shape) # RETURN (1, 27, 1024)\r\n> \r\n> out = model(input_ids,decoder_input_ids=input_ids2)\r\n> print(out.encoder_last_hidden_state.shape) # RETURN (1, 7, 1024)\r\n> ```\r\n> \r\n> If we run the same snippet in Pytorch, they will both return `(1, 27, 1024)` . (tested on both official released and master)\r\n> So likely this is TFBart bug, **and it makes 4 TFRag fast-tests fail**.\r\n> (shape of `out.encoder_last_hidden_state.shape` is not as expected)\r\n\r\nHey @ratthachat, \r\n\r\n> Hi @patrickvonplaten , I agree and all T5 related tests are deleted.\r\n> \r\n> Tests related to TFBart are still not passed which likely due to TFBart bug.\r\n> More precisely, TFBart vs. Torch's Bart forward pass return `generator_enc_last_hidden_state.shape`\r\n> differently if we provide `decoder_input_ids`\r\n> \r\n> ```\r\n> texts = \"My friends are cool but they eat too many carbs. I really want them to be healthy, so I buy them vegetable.\"\r\n> texts2 = \"My friends are cool.\"\r\n> inputs = tokenizer([texts], max_length=1024, return_tensors='tf')\r\n> inputs2 = tokenizer([texts2], max_length=1024, return_tensors='tf')\r\n> \r\n> input_ids=inputs['input_ids']\r\n> input_ids2=inputs2['input_ids']\r\n> out = model(input_ids,decoder_input_ids=None)\r\n> print(out.encoder_last_hidden_state.shape) # RETURN (1, 27, 1024)\r\n> \r\n> out = model(input_ids,decoder_input_ids=input_ids2)\r\n> print(out.encoder_last_hidden_state.shape) # RETURN (1, 7, 1024)\r\n> ```\r\n> \r\n> If we run the same snippet in Pytorch, they will both return `(1, 27, 1024)` . (tested on both official released and master)\r\n> So likely this is TFBart bug, **and it makes 4 TFRag fast-tests fail**.\r\n> (shape of `out.encoder_last_hidden_state.shape` is not as expected)\r\n\r\nHey @ratthachat,\r\n\r\nsorry to answer so late! I don't think that there is a TFBart bug to be honest. When inputting `decoder_input_ids` the behavior you described above is expected and is the same for PyTorch. \r\n\r\nIf you copy & paste the following code on master:\r\n\r\n```python\r\n#!/usr/bin/env python3\r\nfrom transformers import AutoModelForSeq2SeqLM, TFAutoModelForSeq2SeqLM\r\nfrom transformers import AutoTokenizer\r\n\r\nmodel = AutoModelForSeq2SeqLM.from_pretrained(\"facebook/bart-base\")\r\ntf_model = TFAutoModelForSeq2SeqLM.from_pretrained(\"facebook/bart-base\")\r\n\r\ntokenizer = AutoTokenizer.from_pretrained(\"facebook/bart-base\")\r\n\r\ninput_str_short = \"this is a string\"\r\ninput_str_long = \"this is a veeeeeeery veeeeeeeeery long string!!!\"\r\n\r\noutput_shape = model(input_ids=tokenizer(input_str_short, return_tensors=\"pt\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"pt\").input_ids)[0].shape\r\n\r\ntf_output_shape = tf_model(input_ids=tokenizer(input_str_short, return_tensors=\"tf\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"tf\").input_ids)[0].shape\r\n\r\nassert output_shape == tf_output_shape, \"Output shapes have to be the same\"\r\n```\r\n\r\nyou'll see that no assertion error is thrown. \r\n\r\nI think you might have to adapt a couple of TFRag tests to make it work. There also might be a small chance that you have to rebase your current PR to master because there is a weird version of Bart in this PR (but I doubt that a bit to be honest). \r\n\r\nPlease let me know if you need help for the tests! Think we are almost finished !!! :-) ", "Hi @patrickvonplaten , sorry that I did not explain clear enough.\r\nThe test I exactly adapted from Pytorch test the shape of `last_hidden_states` not the shape of `logits` . \r\n\r\nIe. please see\r\nhttps://github.com/ratthachat/transformers/blob/tfrag-draft-new/tests/test_modeling_tf_rag.py#L375\r\n\r\n```\r\nself.assertEqual(\r\n outputs.generator_enc_last_hidden_state.shape,\r\n (n_docs * decoder_input_ids.shape[0], self.max_combined_length, config.generator.hidden_size),\r\n )\r\n```\r\n\r\nFrom your example change output from `[0] (logits)` to `[2] (last_hidden_states)` , **we indeed got assertion error**. \r\n\r\n```\r\nfrom transformers import AutoModelForSeq2SeqLM, TFAutoModelForSeq2SeqLM\r\nfrom transformers import AutoTokenizer\r\n\r\nmodel = AutoModelForSeq2SeqLM.from_pretrained(\"facebook/bart-base\")\r\ntf_model = TFAutoModelForSeq2SeqLM.from_pretrained(\"facebook/bart-base\")\r\n\r\ntokenizer = AutoTokenizer.from_pretrained(\"facebook/bart-base\")\r\n\r\ninput_str_short = \"this is a string\"\r\ninput_str_long = \"this is a veeeeeeery veeeeeeeeery long string!!!\"\r\n\r\noutput_shape = model(input_ids=tokenizer(input_str_short, return_tensors=\"pt\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"pt\").input_ids)[2].shape\r\n\r\ntf_output_shape = tf_model(input_ids=tokenizer(input_str_short, return_tensors=\"tf\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"tf\").input_ids)[2].shape\r\n\r\nassert output_shape == tf_output_shape, \"Output shapes have to be the same\"\r\n```\r\n\r\n```\r\nAssertionError Traceback (most recent call last)\r\n<ipython-input-17-04a80fc987f7> in <module>()\r\n 14 tf_output_shape = tf_model(input_ids=tokenizer(input_str_short, return_tensors=\"tf\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"tf\").input_ids)[2].shape\r\n 15 \r\n---> 16 assert output_shape == tf_output_shape, \"Output shapes have to be the same\"\r\n\r\nAssertionError: Output shapes have to be the same\r\n```\r\n\r\nBTW, about the rebase, I really want to do it, but I could not solve the merging conflicts .", "> Hi @patrickvonplaten , sorry that I did not explain clear enough.\r\n> The test I exactly adapted from Pytorch test the shape of `last_hidden_states` not the shape of `logits` .\r\n> \r\n> Ie. please see\r\n> https://github.com/ratthachat/transformers/blob/tfrag-draft-new/tests/test_modeling_tf_rag.py#L375\r\n> \r\n> ```\r\n> self.assertEqual(\r\n> outputs.generator_enc_last_hidden_state.shape,\r\n> (n_docs * decoder_input_ids.shape[0], self.max_combined_length, config.generator.hidden_size),\r\n> )\r\n> ```\r\n> \r\n> From your example change output from `[0] (logits)` to `[2] (last_hidden_states)` , **we indeed got assertion error**.\r\n> \r\n> ```\r\n> from transformers import AutoModelForSeq2SeqLM, TFAutoModelForSeq2SeqLM\r\n> from transformers import AutoTokenizer\r\n> \r\n> model = AutoModelForSeq2SeqLM.from_pretrained(\"facebook/bart-base\")\r\n> tf_model = TFAutoModelForSeq2SeqLM.from_pretrained(\"facebook/bart-base\")\r\n> \r\n> tokenizer = AutoTokenizer.from_pretrained(\"facebook/bart-base\")\r\n> \r\n> input_str_short = \"this is a string\"\r\n> input_str_long = \"this is a veeeeeeery veeeeeeeeery long string!!!\"\r\n> \r\n> output_shape = model(input_ids=tokenizer(input_str_short, return_tensors=\"pt\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"pt\").input_ids)[2].shape\r\n> \r\n> tf_output_shape = tf_model(input_ids=tokenizer(input_str_short, return_tensors=\"tf\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"tf\").input_ids)[2].shape\r\n> \r\n> assert output_shape == tf_output_shape, \"Output shapes have to be the same\"\r\n> ```\r\n> \r\n> ```\r\n> AssertionError Traceback (most recent call last)\r\n> <ipython-input-17-04a80fc987f7> in <module>()\r\n> 14 tf_output_shape = tf_model(input_ids=tokenizer(input_str_short, return_tensors=\"tf\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"tf\").input_ids)[2].shape\r\n> 15 \r\n> ---> 16 assert output_shape == tf_output_shape, \"Output shapes have to be the same\"\r\n> \r\n> AssertionError: Output shapes have to be the same\r\n> ```\r\n> \r\n> BTW, about the rebase, I really want to do it, but I could not solve the merging conflicts .\r\n\r\nHey @ratthachat, please note that [2] are the `hidden_states` and not the `last_hidden_state`. The last hidden_state is as expected the same.\r\n\r\n```python\r\nfrom transformers import AutoModel, TFAutoModel\r\nfrom transformers import AutoTokenizer\r\n\r\nmodel = AutoModel.from_pretrained(\"facebook/bart-base\")\r\ntf_model = TFAutoModel.from_pretrained(\"facebook/bart-base\")\r\n\r\ntokenizer = AutoTokenizer.from_pretrained(\"facebook/bart-base\")\r\n\r\ninput_str_short = \"this is a string\"\r\ninput_str_long = \"this is a veeeeeeery veeeeeeeeery long string!!!\"\r\n\r\noutput_shape = model(input_ids=tokenizer(input_str_short, return_tensors=\"pt\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"pt\").input_ids).last_hidden_state.shape\r\n\r\ntf_output_shape = tf_model(input_ids=tokenizer(input_str_short, return_tensors=\"tf\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"tf\").input_ids).last_hidden_state.shape\r\n\r\nassert output_shape == tf_output_shape, \"Output shapes have to be the same\"\r\n```", "I'm not 100% sure if the merge was completely correct, but let's first focus on making all TFRag tests pass. The other test we can fix later :-) ", "I apologize @patrickvonplaten . I think I am now a bit confused.\r\nIn TFRag, `model.generator` is created using `TFAutoModelForSeq2SeqLM` (the same as Pytorch).\r\nhttps://github.com/ratthachat/transformers/blob/tfrag-draft-new/src/transformers/models/rag/modeling_tf_rag.py#L519\r\nhttps://github.com/ratthachat/transformers/blob/tfrag-draft-new/src/transformers/models/rag/modeling_rag.py#L356\r\n\r\nIn the above example, you used `TFAutoModel`, so it's not the same.\r\nSo if I changed to `TFAutoModelForSeq2SeqLM` and check `encoder_last_hidden_state.shape` , \r\n**[all fast tests, test this `encoder_last_hidden_state.shape` attribute]**\r\nI still got assertion error. \r\n\r\nI am not sure what's going on and I may miss something simple here. I apologize again.\r\n\r\n```\r\nfrom transformers import AutoModelForSeq2SeqLM, TFAutoModelForSeq2SeqLM\r\nfrom transformers import AutoTokenizer\r\n\r\nmodel = AutoModelForSeq2SeqLM.from_pretrained(\"facebook/bart-base\")\r\ntf_model = TFAutoModelForSeq2SeqLM.from_pretrained(\"facebook/bart-base\")\r\n\r\ntokenizer = AutoTokenizer.from_pretrained(\"facebook/bart-base\")\r\n\r\ninput_str_short = \"this is a string\"\r\ninput_str_long = \"this is a veeeeeeery veeeeeeeeery long string!!!\"\r\n\r\noutput_shape = model(input_ids=tokenizer(input_str_short, return_tensors=\"pt\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"pt\").input_ids).encoder_last_hidden_state.shape\r\ntf_output_shape = tf_model(input_ids=tokenizer(input_str_short, return_tensors=\"tf\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"tf\").input_ids).encoder_last_hidden_state.shape\r\n\r\nprint(output_shape, tf_output_shape)\r\n\r\nassert output_shape == tf_output_shape, \"Output shapes have to be the same\"\r\n```\r\n\r\n```\r\ntorch.Size([1, 6, 768]) (1, 17, 768)\r\n---------------------------------------------------------------------------\r\nAssertionError Traceback (most recent call last)\r\n<ipython-input-4-9790bfa59614> in <module>()\r\n 16 # print(output.keys(), tf_output.keys())\r\n 17 \r\n---> 18 assert output_shape == tf_output_shape, \"Output shapes have to be the same\"\r\n\r\nAssertionError: Output shapes have to be the same\r\n```", "Patrick, another issue is that after rebase, there is an error on `load_weight_prefix` which we invented for TFRag's name, so now the basic building block does not work.\r\n\r\n`TypeError: ('Keyword argument not understood:', 'load_weight_prefix') `\r\n", "> Patrick, another issue is that after rebase, there is an error on `load_weight_prefix` which we invented for TFRag's name, so now the basic building block does not work.\r\n> \r\n> `TypeError: ('Keyword argument not understood:', 'load_weight_prefix') `\r\n\r\nYeah sorry, I made a quick & dirty rebase so there might be errors! It would be awesome if you could fix them (if there are easy to fix)", "> from transformers import AutoModelForSeq2SeqLM, TFAutoModelForSeq2SeqLM\r\n> from transformers import AutoTokenizer\r\n> \r\n> model = AutoModelForSeq2SeqLM.from_pretrained(\"facebook/bart-base\")\r\n> tf_model = TFAutoModelForSeq2SeqLM.from_pretrained(\"facebook/bart-base\")\r\n> \r\n> tokenizer = AutoTokenizer.from_pretrained(\"facebook/bart-base\")\r\n> \r\n> input_str_short = \"this is a string\"\r\n> input_str_long = \"this is a veeeeeeery veeeeeeeeery long string!!!\"\r\n> \r\n> output_shape = model(input_ids=tokenizer(input_str_short, return_tensors=\"pt\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"pt\").input_ids).encoder_last_hidden_state.shape\r\n> tf_output_shape = tf_model(input_ids=tokenizer(input_str_short, return_tensors=\"tf\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"tf\").input_ids).encoder_last_hidden_state.shape\r\n> \r\n> print(output_shape, tf_output_shape)\r\n> \r\n> assert output_shape == tf_output_shape, \"Output shapes have to be the same\"\r\n\r\nI see thanks for the very descriptive error description! You're completely right -> that's a bug, great catch! I'll fix", "> > from transformers import AutoModelForSeq2SeqLM, TFAutoModelForSeq2SeqLM\r\n> > from transformers import AutoTokenizer\r\n> > model = AutoModelForSeq2SeqLM.from_pretrained(\"facebook/bart-base\")\r\n> > tf_model = TFAutoModelForSeq2SeqLM.from_pretrained(\"facebook/bart-base\")\r\n> > tokenizer = AutoTokenizer.from_pretrained(\"facebook/bart-base\")\r\n> > input_str_short = \"this is a string\"\r\n> > input_str_long = \"this is a veeeeeeery veeeeeeeeery long string!!!\"\r\n> > output_shape = model(input_ids=tokenizer(input_str_short, return_tensors=\"pt\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"pt\").input_ids).encoder_last_hidden_state.shape\r\n> > tf_output_shape = tf_model(input_ids=tokenizer(input_str_short, return_tensors=\"tf\").input_ids, decoder_input_ids=tokenizer(input_str_long, return_tensors=\"tf\").input_ids).encoder_last_hidden_state.shape\r\n> > print(output_shape, tf_output_shape)\r\n> > assert output_shape == tf_output_shape, \"Output shapes have to be the same\"\r\n> \r\n> I see thanks for the very descriptive error description! You're completely right -> that's a bug, great catch! I'll fix\r\n\r\nOk merged it: https://github.com/huggingface/transformers/pull/9944. Could you merge master once again into your PR and see whether the tests work now? :-) \r\n\r\n```bash\r\ngit fetch upstream master\r\ngit merge upstream/master\r\n```\r\n\r\nI don' think there will be any merge conflicts. Lemme know if you need any help :-) ", "Thanks so much Patrick. Tomorrow, I will try my best to fix the \"load_weights_prefix\" issue and will come back ❤️ ", "@patrickvonplaten I am sorry - bad news. \r\nEven though now I think all tests should be passed, I could not find a way to fix the `load_weights_prefix` issue above arised after conflict fixing 2 days ago.\r\n\r\nIt seems that this `load_weights_prefix` is sent as `kwarg` to all required functions correctly but failed with Keras not allowing this argument .\r\n(I attach the full `TraceBack` below).\r\n\r\nAt first, I thought that this might be due to the recent TF2.4 upgrade, but I tried downgrade back to TF2.3 and still got the same error. Could you please help take a look?\r\n\r\nTo reproduce, simply initiate the model:\r\n```\r\nfrom transformers import RagTokenizer, RagRetriever \r\nfrom transformers.models.rag.modeling_tf_rag import TFRagModel, TFRagSequenceForGeneration, TFRagTokenForGeneration\r\n\r\nPATH = \"facebook/rag-token-nq\"\r\ntokenizer = RagTokenizer.from_pretrained(PATH)\r\nretriever = RagRetriever.from_pretrained(PATH, index_name=\"exact\", use_dummy_dataset=True) \r\n\r\nmodel = TFRagModel.from_pretrained_question_encoder_generator('facebook/dpr-question_encoder-single-nq-base', \"facebook/bart-base\", generator_from_pt=True, question_encoder_from_pt=True, retriever=retriever) \r\n```\r\n\r\nProduced the following TraceBack:\r\n\r\n```\r\n---------------------------------------------------------------------------\r\nTypeError Traceback (most recent call last)\r\n<ipython-input-10-a201b416721e> in <module>()\r\n 1 \r\n----> 2 model = TFRagModel.from_pretrained_question_encoder_generator('facebook/dpr-question_encoder-single-nq-base', \"facebook/bart-base\", generator_from_pt=True, question_encoder_from_pt=True, retriever=retriever)\r\n\r\n8 frames\r\n/usr/local/lib/python3.6/dist-packages/transformers/models/rag/modeling_tf_rag.py in from_pretrained_question_encoder_generator(cls, question_encoder_pretrained_model_name_or_path, generator_pretrained_model_name_or_path, retriever, *model_args, **kwargs)\r\n 366 name=\"generator\",\r\n 367 load_weight_prefix=cls.load_weight_prefix,\r\n--> 368 **kwargs_generator,\r\n 369 )\r\n 370 \r\n\r\n/usr/local/lib/python3.6/dist-packages/transformers/models/auto/modeling_tf_auto.py in from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)\r\n 1105 if type(config) in TF_MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING.keys():\r\n 1106 return TF_MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING[type(config)].from_pretrained(\r\n-> 1107 pretrained_model_name_or_path, *model_args, config=config, **kwargs\r\n 1108 )\r\n 1109 raise ValueError(\r\n\r\n/usr/local/lib/python3.6/dist-packages/transformers/modeling_tf_utils.py in from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)\r\n 1245 \r\n 1246 # Instantiate model.\r\n-> 1247 model = cls(config, *model_args, **model_kwargs)\r\n 1248 \r\n 1249 if from_pt:\r\n\r\n/usr/local/lib/python3.6/dist-packages/transformers/models/bart/modeling_tf_bart.py in __init__(self, config, load_weight_prefix, *inputs, **kwargs)\r\n 1246 def __init__(self, config, load_weight_prefix=None, *inputs, **kwargs):\r\n 1247 super().__init__(config, *inputs, **kwargs)\r\n-> 1248 self.model = TFBartModel(config, load_weight_prefix=load_weight_prefix, name=\"model\")\r\n 1249 self.use_cache = config.use_cache\r\n 1250 # final_bias_logits is registered as a buffer in pytorch, so not trainable for the the sake of consistency.\r\n\r\n/usr/local/lib/python3.6/dist-packages/transformers/models/bart/modeling_tf_bart.py in __init__(self, config, *inputs, **kwargs)\r\n 1139 class TFBartModel(TFBartPretrainedModel):\r\n 1140 def __init__(self, config: BartConfig, *inputs, **kwargs):\r\n-> 1141 super().__init__(config, *inputs, **kwargs)\r\n 1142 \r\n 1143 self.model = TFBartMainLayer(config, name=\"model\")\r\n\r\n/usr/local/lib/python3.6/dist-packages/transformers/modeling_tf_utils.py in __init__(self, config, *inputs, **kwargs)\r\n 629 \r\n 630 def __init__(self, config, *inputs, **kwargs):\r\n--> 631 super().__init__(*inputs, **kwargs)\r\n 632 if not isinstance(config, PretrainedConfig):\r\n 633 raise ValueError(\r\n\r\n/usr/local/lib/python3.6/dist-packages/tensorflow/python/training/tracking/base.py in _method_wrapper(self, *args, **kwargs)\r\n 455 self._self_setattr_tracking = False # pylint: disable=protected-access\r\n 456 try:\r\n--> 457 result = method(self, *args, **kwargs)\r\n 458 finally:\r\n 459 self._self_setattr_tracking = previous_value # pylint: disable=protected-access\r\n\r\n/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/training.py in __init__(self, *args, **kwargs)\r\n 260 # self.non_trainable_weights\r\n 261 generic_utils.validate_kwargs(kwargs, {'trainable', 'dtype', 'dynamic',\r\n--> 262 'name', 'autocast'})\r\n 263 super(Model, self).__init__(**kwargs)\r\n 264 # By default, Model is a subclass model, which is not in graph network.\r\n\r\n/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/utils/generic_utils.py in validate_kwargs(kwargs, allowed_kwargs, error_message)\r\n 776 for kwarg in kwargs:\r\n 777 if kwarg not in allowed_kwargs:\r\n--> 778 raise TypeError(error_message, kwarg)\r\n 779 \r\n 780 \r\n\r\nTypeError: ('Keyword argument not understood:', 'load_weight_prefix')\r\n```", "> @patrickvonplaten I am sorry - bad news.\r\n> Even though now I think all tests should be passed, I could not find a way to fix the `load_weights_prefix` issue above arised after conflict fixing 2 days ago.\r\n> \r\n> It seems that this `load_weights_prefix` is sent as `kwarg` to all required functions correctly but failed with Keras not allowing this argument .\r\n> (I attach the full `TraceBack` below).\r\n> \r\n> At first, I thought that this might be due to the recent TF2.4 upgrade, but I tried downgrade back to TF2.3 and still got the same error. Could you please help take a look?\r\n> \r\n> To reproduce, simply initiate the model:\r\n> \r\n> ```\r\n> from transformers import RagTokenizer, RagRetriever \r\n> from transformers.models.rag.modeling_tf_rag import TFRagModel, TFRagSequenceForGeneration, TFRagTokenForGeneration\r\n> \r\n> PATH = \"facebook/rag-token-nq\"\r\n> tokenizer = RagTokenizer.from_pretrained(PATH)\r\n> retriever = RagRetriever.from_pretrained(PATH, index_name=\"exact\", use_dummy_dataset=True) \r\n> \r\n> model = TFRagModel.from_pretrained_question_encoder_generator('facebook/dpr-question_encoder-single-nq-base', \"facebook/bart-base\", generator_from_pt=True, question_encoder_from_pt=True, retriever=retriever) \r\n> ```\r\n> \r\n> Produced the following TraceBack:\r\n> \r\n> ```\r\n> ---------------------------------------------------------------------------\r\n> TypeError Traceback (most recent call last)\r\n> <ipython-input-10-a201b416721e> in <module>()\r\n> 1 \r\n> ----> 2 model = TFRagModel.from_pretrained_question_encoder_generator('facebook/dpr-question_encoder-single-nq-base', \"facebook/bart-base\", generator_from_pt=True, question_encoder_from_pt=True, retriever=retriever)\r\n> \r\n> 8 frames\r\n> /usr/local/lib/python3.6/dist-packages/transformers/models/rag/modeling_tf_rag.py in from_pretrained_question_encoder_generator(cls, question_encoder_pretrained_model_name_or_path, generator_pretrained_model_name_or_path, retriever, *model_args, **kwargs)\r\n> 366 name=\"generator\",\r\n> 367 load_weight_prefix=cls.load_weight_prefix,\r\n> --> 368 **kwargs_generator,\r\n> 369 )\r\n> 370 \r\n> \r\n> /usr/local/lib/python3.6/dist-packages/transformers/models/auto/modeling_tf_auto.py in from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)\r\n> 1105 if type(config) in TF_MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING.keys():\r\n> 1106 return TF_MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING[type(config)].from_pretrained(\r\n> -> 1107 pretrained_model_name_or_path, *model_args, config=config, **kwargs\r\n> 1108 )\r\n> 1109 raise ValueError(\r\n> \r\n> /usr/local/lib/python3.6/dist-packages/transformers/modeling_tf_utils.py in from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)\r\n> 1245 \r\n> 1246 # Instantiate model.\r\n> -> 1247 model = cls(config, *model_args, **model_kwargs)\r\n> 1248 \r\n> 1249 if from_pt:\r\n> \r\n> /usr/local/lib/python3.6/dist-packages/transformers/models/bart/modeling_tf_bart.py in __init__(self, config, load_weight_prefix, *inputs, **kwargs)\r\n> 1246 def __init__(self, config, load_weight_prefix=None, *inputs, **kwargs):\r\n> 1247 super().__init__(config, *inputs, **kwargs)\r\n> -> 1248 self.model = TFBartModel(config, load_weight_prefix=load_weight_prefix, name=\"model\")\r\n> 1249 self.use_cache = config.use_cache\r\n> 1250 # final_bias_logits is registered as a buffer in pytorch, so not trainable for the the sake of consistency.\r\n> \r\n> /usr/local/lib/python3.6/dist-packages/transformers/models/bart/modeling_tf_bart.py in __init__(self, config, *inputs, **kwargs)\r\n> 1139 class TFBartModel(TFBartPretrainedModel):\r\n> 1140 def __init__(self, config: BartConfig, *inputs, **kwargs):\r\n> -> 1141 super().__init__(config, *inputs, **kwargs)\r\n> 1142 \r\n> 1143 self.model = TFBartMainLayer(config, name=\"model\")\r\n> \r\n> /usr/local/lib/python3.6/dist-packages/transformers/modeling_tf_utils.py in __init__(self, config, *inputs, **kwargs)\r\n> 629 \r\n> 630 def __init__(self, config, *inputs, **kwargs):\r\n> --> 631 super().__init__(*inputs, **kwargs)\r\n> 632 if not isinstance(config, PretrainedConfig):\r\n> 633 raise ValueError(\r\n> \r\n> /usr/local/lib/python3.6/dist-packages/tensorflow/python/training/tracking/base.py in _method_wrapper(self, *args, **kwargs)\r\n> 455 self._self_setattr_tracking = False # pylint: disable=protected-access\r\n> 456 try:\r\n> --> 457 result = method(self, *args, **kwargs)\r\n> 458 finally:\r\n> 459 self._self_setattr_tracking = previous_value # pylint: disable=protected-access\r\n> \r\n> /usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/training.py in __init__(self, *args, **kwargs)\r\n> 260 # self.non_trainable_weights\r\n> 261 generic_utils.validate_kwargs(kwargs, {'trainable', 'dtype', 'dynamic',\r\n> --> 262 'name', 'autocast'})\r\n> 263 super(Model, self).__init__(**kwargs)\r\n> 264 # By default, Model is a subclass model, which is not in graph network.\r\n> \r\n> /usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/utils/generic_utils.py in validate_kwargs(kwargs, allowed_kwargs, error_message)\r\n> 776 for kwarg in kwargs:\r\n> 777 if kwarg not in allowed_kwargs:\r\n> --> 778 raise TypeError(error_message, kwarg)\r\n> 779 \r\n> 780 \r\n> \r\n> TypeError: ('Keyword argument not understood:', 'load_weight_prefix')\r\n> ```\r\n\r\nShould be fine now - can you check? ", "@patrickvonplaten all fast and slow tests are now pass 😄 \r\nJust that in one slow test : `test_rag_token_generate_batch` , my colab P100 ran out of memory if we used all 15 inputs.\r\nIf I reduce the size of inputs to 5, the test passed. \r\n(any 5 of 15 will passed indicated that all outputs are correct)", "@ratthachat - you've really done an amazing job here! The PR looks very nice to me overall. \r\nOne thing, I'd like to change before trying to merge is to delete the `_generate_beam_search` and `_generate_no_beam_search` methods and use the default ones instead. I can definitely help you get this done here. Do you know what the differences are in `_generate_beam_search` that you added to `modeling_tf_rag.py` compared to the one in `modeling_tf_utils.py`? Happy to help you here\r\n\r\nApart from that, I only left a couple of nits.", "Hi Patrick, thanks for all your super kind helps all these time! 😄 ❤️ \r\nI improved docstrings as suggested.\r\n\r\nAbout` _generate_beam_search` and `_generate_no_beam_search`, actually there are exactly **20 lines differences**.\r\nI made a notebook to show **20 lines differences** -- Version1 (`TFRag`) and Version2 (`generation_tf_utils.py`) \r\nhttps://www.kaggle.com/ratthachat/generate-beam-and-no-searches/\r\n (Please clicking version on the top-right and see `diff` )\r\n\r\nMainly, I simply fix both functions to accept `**kwarg` arguments from `TFRag` (in particular `kwargs[\"encoder_outputs\"]` ).\r\nHowever, I did not directly change/PR `generation_tf_utils.py` for two reasons : \r\n\r\n(1) I am not sure if this changes will affect other TF models or not and I don't have enough resources to check them all\r\n(2) As we once discussed, there will be big 'generation refactor' in `generation_tf_utils.py` like Pytorch soon, and it should be a great chance there to fix & test TFRag at that time together with other TF models.\r\n\r\nWhat do you think ?\r\n\r\n", "All the slow tests passed for TFBart & TFRag -> PR is ready for review. Started running the whole suite of SLOW tests, just to be sure - will report any unexpected behavior here.\r\n\r\n@LysandreJik @sgugger @jplu - it would be great if you can review the PR.", "> I am very much not in favor of changing the core modeling utils to fit the needs of one model, so really dislike the change related to `from_pretrained`. I understand the problems of scope, but it seems to me that it's only there to be able to write the convenience init `from_encoder_decoder_pretrained` and the like which is not strictly necessary (one can load the encoder and decoder out of the model then instantiate it by passing the encoder and the decoder, it's just three lines of code instead of one).\r\n> \r\n> I would favor this option for now while we look into different solutions for the weight loading.\r\n\r\nI also think that it's not clean at all to introduce this hack to be able to load the pretrained weights correctly, but there is no way around it really. \r\n\r\nIt is not possible to load the encoder and decoder separately and to pass them into init if one wants to save the model afterward. Even with the current hack this is not possible because there is no way of knowing what the correct scope of the overall model is when the submodel is loaded separately. Try this in the branch of this PR:\r\n\r\n```python\r\nfrom transformers import TFBartForConditionalGeneration, TFDPRQuestionEncoder, TFRagModel, RagRetriever\r\n\r\nretriever = RagRetriever.from_pretrained(\"facebook/rag-sequence-nq\", index_name=\"exact\", use_dummy_dataset=True)\r\n\r\nencoder = TFDPRQuestionEncoder.from_pretrained(\"facebook/dpr-question_encoder-single-nq-base\")\r\ngenerator = TFBartForConditionalGeneration.from_pretrained(\"facebook/bart-large-cnn\")\r\n\r\nrag = TFRagModel(question_encoder=encoder, generator=generator, retriever=retriever)\r\n\r\nrag.save_pretrained(\"rag_temp\")\r\n\r\nnew_rag = TFRagModel.from_pretrained(\"rag_temp\", retriever=retriever) # ERROR => the weights are randomly initialized here\r\n```\r\n\r\nThis means that ```from_encoder_decoder_pretrained``` is not just a convenience function, but actually the only way to correctly load two \"sub\" models into a composite model class. Otherwise the weights are not saved correctly and can then not be loaded again.\r\n\r\nAlso one other thing I want to point out here is that the change/hack added to `TFPretrainedModel`'s `from_pretrained` method in `modeling_tf_utils.py` is not just done for RAG but will make composite TF model classes such as `TFEncoderDecoderModel` possible.\r\nAt the same time, I also think it's a big hack that is not beautiful in any way. \r\n\r\nAt the moment, I don't really see another solution here because we cannot just overwrite the `from_pretrained(...)` method for TFRag since the `load_preifx_weight` is passed to BART's and DPR's `from_pretrained`'s method. What we could do instead however is to add a whole new `from_pretrained_with_prefix(....)` function to `modeling_tf_utils.py` instead of changing the existing method. \r\n\r\nWhat do you think? @sgugger @LysandreJik & @jplu ", "> At the moment, I don't really see another solution here because we cannot just overwrite the from_pretrained(...) method for TFRag since the load_preifx_weight is passed to BART's and DPR's from_pretrained's method. What we could do instead however is to add a whole new from_pretrained_with_prefix(....) function to modeling_tf_utils.py instead of changing the existing method.\r\n\r\nFrom a short term view, I don't see another solution either. We clearly have an issue with the weights naming for Seq2Seq models and the found workaround we had until now (the `tf.compat.v1...`) reaches its limits with RAG as it requires changes where it should not.\r\n\r\nFor me this has to be rethought because I think that all the convert issues have to be handled in the `modeling_tf_pytorch_utils.py` script and not elsewhere, and we should stop to force TF to have the same names than PT but more handle how to convert \"proper\" TF weight names to \"proper\" PT weight names (and the other way around). I think that if we continue without having a better design for this we go to a more complex implementation and then understanding. We should also keep in mind that the day TF removes the V1 compatibility, none of these models will work as expected. Hence, I would accept this as a temporary solution, but we clearly need to review this TF naming part thoroughly.", "As discussed a bit offline, @LysandreJik will take a final review and then we'll merge the approach of this PR.\r\n\r\nWe should integrate the suggestion from @sgugger which is to change `prefix` to a private function arg `_prefix` to keep the design flexible for future changes.\r\n\r\nOnce @LysandreJik has done the review, I'll do a last refactor & then we can merge I think @ratthachat :-)", "Thanks so much everyone for the merge! Especially @jplu who gave insightful comments on several earlier versions and @patrickvonplaten who has collaborated and has greatly helped in all aspects of this works!! \r\n(Honestly it's not possible without Patrick's help)\r\n\r\nAbout the single mismatched generated answer (\"step by step\" vs. \"evolution\"), I will investigate this point further. Strangely, in earlier versions all tests are passed meaning all outputs are equivalent." ]

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

[]

[ "I ported this model for easy use in Hugging Face Transformers. Try using the code below!\r\n\r\n### 1. Create models and tokenizers\r\n```python\r\n>> from transformers import AutoModelForSeq2SeqLM, PreTrainedTokenizerFast\r\n\r\n>>> model = AutoModelForSeq2SeqLM.from_pretrained(\"hyunwoongko/ctrlsum-cnndm\")\r\n>>> # model = AutoModelForSeq2SeqLM.from_pretrained(\"hyunwoongko/ctrlsum-arxiv\")\r\n>>> # model = AutoModelForSeq2SeqLM.from_pretrained(\"hyunwoongko/ctrlsum-bigpatent\")\r\n\r\n>>> tokenizer = PreTrainedTokenizerFast.from_pretrained(\"hyunwoongko/ctrlsum-cnndm\")\r\n>>> # tokenizer = PreTrainedTokenizerFast.from_pretrained(\"hyunwoongko/ctrlsum-arxiv\")\r\n>>> # tokenizer = PreTrainedTokenizerFast.from_pretrained(\"hyunwoongko/ctrlsum-bigpatent\")\r\n```\r\n\r\n### 2. Unconditioned summarization\r\n```python\r\n>>> data = tokenizer(\"My name is Kevin. I love dogs. I loved dogs from 1996. Today, I'm going to walk on street with my dogs\", return_tensors=\"pt\")\r\n>>> input_ids, attention_mask = data[\"input_ids\"], data[\"attention_mask\"]\r\n>>> tokenizer.batch_decode(model.generate(input_ids, attention_mask=attention_mask, num_beams=5))[0]\r\n'</s>My name is Kevin. I loved dogs from 1996.</s>'\r\n```\r\n### 3. Conditioned summarization\r\n- You can input condition token using `TOKEN => CONTENTS` structure\r\n```python\r\n>>> data = tokenizer(\"today plan => My name is Kevin. I love dogs. I loved dogs from 1996. Today, I'm going to walk on street with my dogs\", return_tensors=\"pt\")\r\n>>> input_ids, attention_mask = data[\"input_ids\"], data[\"attention_mask\"]\r\n>>> tokenizer.batch_decode(model.generate(input_ids, attention_mask=attention_mask, num_beams=5))[0]\r\n\"</s> Today, I'm going to walk on street with my dogs. I loved dogs from 1996</s>\"\r\n```\r\n\r\n### 4. Prompt summarization\r\n- You can also input `decoder_input_ids` for input prompt.\r\n```python\r\n>>> data = tokenizer(\"Q:What is my name? A: => My name is Kevin. I love dogs. I loved dogs from 1996. Today, I'm going to walk on street with my dogs\", return_tensors=\"pt\")\r\n>>> input_ids, attention_mask = data[\"input_ids\"], data[\"attention_mask\"]\r\n>>> tokenizer.batch_decode(model.generate(input_ids, attention_mask=attention_mask, num_beams=5, decoder_input_ids=tokenizer(\"Q:What is My name? A:\", return_tensors=\"pt\")[\"input_ids\"][:, :-1]))[0]\r\n'<s>Q:What is My name? A: Kevin.</s>'\r\n```" ]

[]

[]

[ "analogs to #8923", "this one is probably safe to close - probably your link was satisfactory, @patrickvonplaten " ]

[ { "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!" } ]

[ { "login": "LysandreJik", "id": 30755778, "node_id": "MDQ6VXNlcjMwNzU1Nzc4", "avatar_url": "https://avatars.githubusercontent.com/u/30755778?v=4", "gravatar_id": "", "url": "https://api.github.com/users/LysandreJik", "html_url": "https://github.com/LysandreJik", "followers_url": "https://api.github.com/users/LysandreJik/followers", "following_url": "https://api.github.com/users/LysandreJik/following{/other_user}", "gists_url": "https://api.github.com/users/LysandreJik/gists{/gist_id}", "starred_url": "https://api.github.com/users/LysandreJik/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/LysandreJik/subscriptions", "organizations_url": "https://api.github.com/users/LysandreJik/orgs", "repos_url": "https://api.github.com/users/LysandreJik/repos", "events_url": "https://api.github.com/users/LysandreJik/events{/privacy}", "received_events_url": "https://api.github.com/users/LysandreJik/received_events", "type": "User", "site_admin": false }, { "login": "SaulLu", "id": 55560583, "node_id": "MDQ6VXNlcjU1NTYwNTgz", "avatar_url": "https://avatars.githubusercontent.com/u/55560583?v=4", "gravatar_id": "", "url": "https://api.github.com/users/SaulLu", "html_url": "https://github.com/SaulLu", "followers_url": "https://api.github.com/users/SaulLu/followers", "following_url": "https://api.github.com/users/SaulLu/following{/other_user}", "gists_url": "https://api.github.com/users/SaulLu/gists{/gist_id}", "starred_url": "https://api.github.com/users/SaulLu/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/SaulLu/subscriptions", "organizations_url": "https://api.github.com/users/SaulLu/orgs", "repos_url": "https://api.github.com/users/SaulLu/repos", "events_url": "https://api.github.com/users/SaulLu/events{/privacy}", "received_events_url": "https://api.github.com/users/SaulLu/received_events", "type": "User", "site_admin": false } ]

[ "I'm taking a look at why this is, it seems that we have a differing behavior between `from_pretrained` and the initialization method. I tried loading that `spiece.model` directly with `from_pretrained` and it behaves normally, and so does pointing to a directory containing solely that file.\r\n\r\nI'm taking a look and will come back to you.", "@LysandreJik the problem is that `unique_no_split_tokens` is not initialised when you create a tokenizer from `__init__`.\r\n\r\nSee: https://stackoverflow.com/questions/64631665/what-is-the-difference-in-robertatokenizer-and-from-pretrained-way-of-initia/64640570#64640570", "Indeed, thanks for investigating @cronoik. Do you want to open a PR with a fix?", "Yes, I can but I would like to discuss this before because it affects the core of the library and all tokenizers (I have only checked the slow tokenizers yet, but it probably applies to the fast tokenizers as well.). \r\n\r\nWhen a user calls `.from_pretrained`, the tokenizer is created with `__init__` in the `._from_pretrained` method of the `PreTrainedTokenizerBase` class ([line 1868](https://github.com/huggingface/transformers/blob/b01ddc9577b87f057e163d49563ee3f74f4810cf/src/transformers/tokenization_utils_base.py#L1868)). The problem is now, that `._from_pretrained` does some magic from line [1881](https://github.com/huggingface/transformers/blob/b01ddc9577b87f057e163d49563ee3f74f4810cf/src/transformers/tokenization_utils_base.py#L1881) to [1909](https://github.com/huggingface/transformers/blob/b01ddc9577b87f057e163d49563ee3f74f4810cf/src/transformers/tokenization_utils_base.py#L1909), that is not executed when you create the tokenizer from `__init__` directly. \r\n\r\nSo, simply said all I need to do is to move this magic to the `__init__` method and remove it from the `._from_pretrained`?\r\n\r\n@LysandreJik ", "Pinging @thomwolf for advice on tokenizers loading methods.", "This issue has been stale for 1 month.", "@LysandreJik Maybe the [documentation](https://huggingface.co/transformers/main_classes/tokenizer.html) should be updated to at least tell the people that the recommended way to initialize a tokenizer is `from_pretrained` and that is not guaranteed that `__init__` will work properly?", "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/master/CONTRIBUTING.md) are likely to be ignored.", "It is closed without solving the 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/master/CONTRIBUTING.md) are likely to be ignored.", "Pinging @SaulLu here, as I also encountered this.", "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/master/CONTRIBUTING.md) are likely to be ignored." ]

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

[]

[ "model weights available?", "No. :( " ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "you may replace is_world_master by is_world_process_zero.", "Any solutions to this? I am not able to use ray tune backend in any way", "If you get this error still it's probably because you're using mismatched library versions/example versions. Using the latest examples or a more recent version (v4.1.x) should patch this.\r\n\r\nIf it doesn't, then please open a new issue and fill in the issue template so that we may help you. Thank you." ]

[]

[]

[ "Hello!\r\n\r\nFor now, as the TF models are implemented right now, this is the normal behavior, if you want to have a different size you have to set it manually yourself before to create the saved model. It is planned to make the size of the sequence length variable when creating a saved model but we don't know when. If you don't know how to do it, I can show you how :)", "Hi Julien! Thanks for your response. Could you please show me how to manually set that?", "To do that you can run the following lines:\r\n```\r\nfrom transformers import TFDistilBertModel, DistilBertTokenizer\r\ntf_model = TFDistilBertModel.from_pretrained(model_name, from_pt=True)\r\ntokenizer = DistilBertTokenizer.from_pretrained(model_name)\r\ninputs = tokenizer(\"My test sentence\", padding=\"max_length\", max_length=128, return_tensors=\"tf\")\r\ntf_model._saved_model_inputs_spec = None\r\nmodel._set_save_spec(inputs)\r\ntf.saved_model.save(tf_model, path)\r\n```", "Thanks!" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "The failed circleci test is not related to my PR. :)", "Hi @cronoik! There's been a mistake done yesterday, and the history of your branch was messed up by mistake (see the file changes, +/-). Do you mind closing this PR and opening another one? No need to do anything on the branch, just closing this one and opening a new one should be enough. Thank you!" ]

[]

[]

[ "I think I'll take a Flax 101 class before reviewing this PR.", "Requires more discussion" ]

[]

[]

[]

[ { "id": 1838412367, "node_id": "MDU6TGFiZWwxODM4NDEyMzY3", "url": "https://api.github.com/repos/huggingface/transformers/labels/model%20card", "name": "model card", "color": "92d5f4", "default": false, "description": "Related to pretrained model cards" } ]

[ { "login": "LysandreJik", "id": 30755778, "node_id": "MDQ6VXNlcjMwNzU1Nzc4", "avatar_url": "https://avatars.githubusercontent.com/u/30755778?v=4", "gravatar_id": "", "url": "https://api.github.com/users/LysandreJik", "html_url": "https://github.com/LysandreJik", "followers_url": "https://api.github.com/users/LysandreJik/followers", "following_url": "https://api.github.com/users/LysandreJik/following{/other_user}", "gists_url": "https://api.github.com/users/LysandreJik/gists{/gist_id}", "starred_url": "https://api.github.com/users/LysandreJik/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/LysandreJik/subscriptions", "organizations_url": "https://api.github.com/users/LysandreJik/orgs", "repos_url": "https://api.github.com/users/LysandreJik/repos", "events_url": "https://api.github.com/users/LysandreJik/events{/privacy}", "received_events_url": "https://api.github.com/users/LysandreJik/received_events", "type": "User", "site_admin": false } ]

[ "Closing this PR and opening a new one on the same branch due to Github issues." ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "Could you elaborate on the use case? It seems dangerous and magical to me. When passing a parameter to a function that is not in the signature, the user gets a `ValueError`.", "Sure! an EagerTensor doesn't have a `.name` attribute so we assume for that case that the values are given in the parameters order. That's ok because we don't have the choice, but why not having the same behavior in case someone decides to name the tensors as he wishs.\r\n\r\nThis is very picky, and I won't fight at all if not accepted ahah", "Mmm, but in this test we're not eager tensors since there is a `.name` attribute, or am I missing something?", "While I was trying to explain this, a use case came to my mind, and indeed this behavior is not correct for an edge use case:\r\n```\r\nfrom transformers import AutoTokenizer, TFBertForSequenceClassification, BertConfig\r\nimport tensorflow as tf\r\nimport datasets\r\n\r\nconfig = BertConfig.from_pretrained(\"bert-base-cased\", num_labels=6)\r\ntokenizer = AutoTokenizer.from_pretrained(\"bert-base-cased\")\r\nds = datasets.load_dataset('emotion')\r\nencoded_train = ds['train'].map(lambda examples: tokenizer(examples['text'], truncation=True, padding='max_length', max_length=128), batched = True)\r\nencoded_train.set_format(type='tensorflow', columns=['input_ids', 'attention_mask', 'label'])\r\nfeatures_train = {x: encoded_train[x].to_tensor(default_value=0, shape=[None, 128]) for x in ['input_ids', 'attention_mask']}\r\ntrain_ds = tf.data.Dataset.from_tensor_slices((features_train, encoded_train[\"label\"])).batch(16)\r\ninput_ids = tf.keras.Input(shape=(128,), dtype='int32', name=\"input_ids\")\r\nattention_mask = tf.keras.Input(shape=(128, ), dtype='int32', name=\"attention_mask\")\r\ntransformer = TFBertForSequenceClassification.from_pretrained(\"bert-base-cased\", num_labels=6)\r\nencoded = transformer([input_ids, attention_mask])\r\nlogits = encoded[0]\r\nmodel = tf.keras.models.Model(inputs = [input_ids, attention_mask], outputs = logits)\r\n\r\nmodel.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=3e-5, epsilon=1e-08, clipnorm=1.0), \r\n loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True), \r\n metrics=[tf.keras.metrics.SparseCategoricalAccuracy('accuracy')])\r\nmodel.fit(train_ds, epochs=1, steps_per_epoch=1)\r\n```\r\nWe get:\r\n```\r\nValueError: The tensor named IteratorGetNext:1 does not belong to the authorized list of names ['input_ids', 'attention_mask', 'token_type_ids', 'position_ids', 'head_mask', 'inputs_embeds', 'output_attentions', 'output_hidden_states', 'return_dict', 'labels', 'training'].\r\n```\r\n\r\nWhich is normal because `.fit()` wraps the dataset into an iterator, and then the tensors are renamed accordingly. Thanks @sgugger for asking the question :)", "Thanks for explaining, I understand better now :-)", "Ok,just realized it is even worse, the inputs gets an ID, here `IteratorGetNext:1` and `IteratorGetNext:0` but the order of the list is never guaranteed. I'm trying to think to a fix for this.", "Ok, as long as we are naming the inputs accordingly to the parameters, the order is safe. For example:\r\n```\r\ninput_ids = tf.keras.Input(shape=(128,), dtype='int32', name=\"input_ids\")\r\nattention_mask = tf.keras.Input(shape=(128, ), dtype='int32', name=\"attention_mask\")\r\n\r\nmodel = tf.keras.models.Model(inputs = [input_ids, attention_mask], outputs = ...)\r\n```\r\n\r\nIs perfectly fine and works as expected, but:\r\n```\r\ninput_ids = tf.keras.Input(shape=(128,), dtype='int32')\r\nattention_mask = tf.keras.Input(shape=(128, ), dtype='int32')\r\n\r\nmodel = tf.keras.models.Model(inputs = [input_ids, attention_mask], outputs = ...)\r\n```\r\n\r\nBrings an undefined behavior into the order.\r\n\r\nNevertheless, there is still an issue. Let's imagine this case:\r\n```\r\ninput_embeds = tf.keras.Input(shape=(768,), dtype='float32')\r\nattention_mask = tf.keras.Input(shape=(128, ), dtype='int32')\r\n\r\nmodel = tf.keras.models.Model(inputs = [input_embeds, attention_mask], outputs = ...)\r\n```\r\n\r\nWon't work because internally, the `input_ids` parameter will take the value of the `input_embeds` input. This can be solved by integrating the names of each parameter directly inside the model, but we cannot do this because of a bug in TF <= 2.4, and will be solved in the TF 2.5 release. So as long as this release is not out, we cannot fix this, so we have to live with this bug, even though this is an edge use case.\r\n\r\nWhat do you think?", "I think we should document that this does not work and encourage users to use named inputs then.", "I have completed the documentation of the `input_processing` function. Does-it sounds enough as explanation for you?", "LGTM!", "LGTM! @LysandreJik feel free to merge if the PR gets your approval!" ]

[ { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "Hello! We really cannot help you with this little information. Please respect the issue template with your environment information, the exact command you used to launch the script, and the full stack trace.\r\n\r\nThank you for your understanding.", "@rabeehk Did you manage to solve this? Experiencing the same issue fine-tuning mT5 on toy data.", "yes I did managed, the issue was I needed to set a longer max_length for the decoder.", "\r\nif you want to debug your codes, go to the place where huggingface computes the final metrics, like bleu, ... and there you can check that prediction max length and targets max-length need to match", "This issue has been automatically marked as stale and been closed because it has not had recent activity. Thank you for your contributions.\n\nIf you think this still needs to be addressed please comment on this thread." ]

[ { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "This issue has been automatically marked as stale and been closed because it has not had recent activity. Thank you for your contributions.\n\nIf you think this still needs to be addressed please comment on this thread." ]

[ { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "This issue has been automatically marked as stale and been closed because it has not had recent activity. Thank you for your contributions.\n\nIf you think this still needs to be addressed please comment on this thread." ]

[]

[]

[ "Thank you all for your reviews. Will do the last changes and update from `LMHead` to `CausalLM`.\r\n\r\n@jplu the GA machines are indeed less beefy than the CircleCI ones (and there's a reason for that, we pay CircleCI but not GA).", "Closing PR and opening it again because the history is messed up." ]

[]

[]

[ "trying to self-heal PR", "I guess continuing over here: https://github.com/huggingface/transformers/pull/8997", "ok, succeeded at restoring this PR after force pushed mess up.\r\n\r\nThanks to this recipe: https://gist.github.com/robertpainsi/2c42c15f1ce6dab03a0675348edd4e2c", "let's please monitor new PRs closely, so that it doesn't somehow break the job while testing things. thank you.", "> Crazy that even the GitHub API is unreliable on that front.\r\n\r\nI haven't seen it with my own eyes (other than when force pushed history of master was rewritten yesterday), but I found more than one report of it being unreliable on various forums.\r\n\r\nIt's possible that they are referring to the situation when someone force pushes into the PR branch, thus changing the local history which could impact the forking/branching point (== `base.sha`), but github API continues to report the original `base.sha` for awhile - I think based on reports due to caching.\r\n\r\nSo this workaround going into user's branch and derives an up-to-date branching point from their branch - at a cost of needing to clone their forked repo." ]

[]

[]

[ "Thank God they removed that horrible thing! Instead of the complex parsing, maybe a `try`/`except` would be cleaner?", " no horrible formatting that way! good idea - done.", "@stas00, could it be possible to apply your fix to the code of version `3.5.X` and release the next **minor** version (like v3.5.2)?", "@LysandreJik, your help is needed here. I don't know anything about how old branches maintenance is done. \r\n\r\nThis PR was merged in 4x series and @vyshkant is requesting this fix applied to 3.5.x for the next release.\r\n\r\nThank you.", "No we won't do fixes for old versions, either upgrade to v4 or use PyTorch < 1.8 if you want to stick to v3.5." ]

[]

[]

[ "Pinging @sgugger, the king of model outputs!", "That was very quick :) Thank you @sgugger and @LysandreJik !" ]

[]

[]

[ "Hello, 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://discusss.huggingface.co) instead?\r\n\r\nThanks!" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "Closing in favor of #8976 " ]

[]

[]

[]

[]

[]

[ "@LysandreJik I had to remove some tests that were testing the decoder mode for MobileBert. \r\n\r\nOne test still fails (flax), the error seems unrelated to this PR unless I am missing something?" ]

[]

[]

[ "@patrickvonplaten I have added an integration test with a pre-trained weight in [https://github.com/StillKeepTry/transformers/blob/dfc18d59da04c38723553354ee1799ce204f52c8/tests/test_modeling_mpnet.py#L240](https://github.com/StillKeepTry/transformers/blob/dfc18d59da04c38723553354ee1799ce204f52c8/tests/test_modeling_mpnet.py#L240)\r\n\r\n", "> @patrickvonplaten I have added an integration test with a pre-trained weight in https://github.com/StillKeepTry/transformers/blob/dfc18d59da04c38723553354ee1799ce204f52c8/tests/test_modeling_mpnet.py#L240\r\n\r\nThat's awesome thanks a lot!", "Think you need to run `make style` and then the last test should pass as well :-)", "@jplu \r\n\r\nI have updated the inputs handling in the TF file now, and rebase and fix the conflicting files. \r\n\r\nBesides, I have used `make style` multiple times but it still reports an error in `check_code_quality`. And I have checked and it seems the problem is not from my added part in the [https://github.com/StillKeepTry/transformers/blob/master/src/transformers/__init__.py](https://github.com/StillKeepTry/transformers/blob/master/src/transformers/__init__.py), despite it reports an error. ", "@LysandreJik Thanks for pointing out this problem. I have fixed it.", "The `make style` issue is probably because of the isort version installed, maybe you can try uninstalling black/isort and doing the following at the root of the repo:\r\n```\r\npip uninstall black isort\r\npip install -e .[quality]\r\n```\r\nIf you want I can run `make style` and push on your branch so that it's ready to be merged.", "@LysandreJik Thank you. You are right. It is because I have installed black and isort before. ", "Great, thanks! The quality test really doesn't like you haha! \r\n\r\nThis time I think it's because of the \"# Copied from xxx ...\" which still uses the old scheme (like `transformers.modeling_roberta.RobertaLMHead`) instead of the new scheme (like `transformers.models.roberta.modeling_roberta.RobertaLMHead`).", "It seems ok now :) ...", "@jplu I have fixed your comments now.", "Thanks!!\r\n\r\n@sgugger @patrickvonplaten @LysandreJik I'm seeing that the `TFMPNetForPreTraining` and `MPNetForPreTraining` are missing from the TF and PT file. Should they be added? Otherwise it is fine for me :)", "> Thanks!!\r\n> \r\n> @sgugger @patrickvonplaten @LysandreJik I'm seeing that the `TFMPNetForPreTraining` and `MPNetForPreTraining` are missing from the TF and PT file. Should they be added? Otherwise it is fine for me :)\r\n\r\nI observe that some models also lack `TFXXXForPreTraining` and `XXXForPreTraining`. I am willing to add them in the next stage. ", "Hey @StillKeepTry, \r\n\r\nwe are super sorry, we had a problem yesterday with git and this is why your git history is cluttered with wrong commits earlier. I cleaned your PR and pushed it to a new branch on master here: https://github.com/huggingface/transformers/pull/9004 . \r\nIt should include all the commits you had earlier. I think we all gave our thumbs-up, so we could merge the other pull request to master (which would require the least amount of work from your side). \r\n\r\nHowever if you want to be the main author of the PR (which is 100% understandable and which is what I would want!), can you do the following steps to open a new clean PR which was exactly like before:\r\n\r\nIn your repo (https://github.com/StillKeepTry/transformers), assuming that the remote to the original hugging face repo (https://github.com/huggingface/transformers.git) is called `upstream`:\r\n\r\n```\r\n$ git fetch upstream\r\n$ git checkout upstream/master\r\n$ git checkout -b add_mp_net_new\r\n# now we'll cherry pick all of your commits\r\n$ git cherry-pick 7361516^..78dcc71\r\n$ git push\r\n# => now you should be able to open new PR with exactly the commits you had previously\r\n```\r\n\r\nLemme know if you need help doing this (or if you don't mind merging https://github.com/huggingface/transformers/pull/9004 - but it would be fairer to you if you're also officially the main author!).\r\n\r\nBig sorry again!", "@patrickvonplaten Never mind, just use your PR. I am ok if our work can be merged into the master quickly. ", "Hello! I can not see the data collator for permuted and masked language models. Was it added also inside HuggingFace? There is an already proposed way to do this collator inside the trainer? \r\n\r\nThanks!", "@gaceladri we have an example for permutation language modeling, check it out here: https://github.com/huggingface/transformers/blob/master/examples/pytorch/language-modeling/run_plm.py", "Hi @LysandreJik, thank you for your kind response. This data collator that you pointed me out, is the collator from permuted language model used in XLNet right? I am unsure that this is a collator to replicate MPNet that mask tokens, not indices and also do the permutation. Sure that I am misunderstanding something..." ]

[]

[]

[ "Merging to avoid merge conflicts. Can address comments in a follow-up PR." ]

[]

[]

[ "You are right, there are never used. These seem to have been added by mistake during the original implementation. I would be all for a cleanup, as long as it doesn't touch to anything other than these attributes." ]

[]

[ { "login": "patrickvonplaten", "id": 23423619, "node_id": "MDQ6VXNlcjIzNDIzNjE5", "avatar_url": "https://avatars.githubusercontent.com/u/23423619?v=4", "gravatar_id": "", "url": "https://api.github.com/users/patrickvonplaten", "html_url": "https://github.com/patrickvonplaten", "followers_url": "https://api.github.com/users/patrickvonplaten/followers", "following_url": "https://api.github.com/users/patrickvonplaten/following{/other_user}", "gists_url": "https://api.github.com/users/patrickvonplaten/gists{/gist_id}", "starred_url": "https://api.github.com/users/patrickvonplaten/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/patrickvonplaten/subscriptions", "organizations_url": "https://api.github.com/users/patrickvonplaten/orgs", "repos_url": "https://api.github.com/users/patrickvonplaten/repos", "events_url": "https://api.github.com/users/patrickvonplaten/events{/privacy}", "received_events_url": "https://api.github.com/users/patrickvonplaten/received_events", "type": "User", "site_admin": false } ]

[ "Exactly same issue while using transformers. I'm using Pytorch 1.7. The solution presented solved the issue" ]

[ { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "Thanks for your proposed solution @alexyalunin ! \r\n\r\nI don't really think that this is a problem on our side. I think MLFlow should better handle this no? ", "Probably, but since you use mlflow inside your library people might expect it working with your models. I won't open the issue in mlflow repo, I leave it to someone who encounters this error again. @patrickvonplaten You can close this issue then. ", "This issue has been automatically marked as stale and been closed because it has not had recent activity. Thank you for your contributions.\n\nIf you think this still needs to be addressed please comment on this thread." ]

[]

[]

[ "While I understand where you're coming from, why not define the loss function outside of the model, like it's generally done for PyTorch models? If you do not pass the labels, you get the logits from the model. You can use these logits with your labels to compute any loss you want.", "Ok. Good I asked first because I did not know this general practice. 😄 \r\n\r\nSo, if I want to use the `Trainer`, I will then be required to only override:\r\nhttps://github.com/huggingface/transformers/blob/9d7d0005b046a95d9d59354714bb6c3547a612fe/src/transformers/trainer.py#L1114-L1120\r\n&rarr; I would need to split the labels from the input, feed it into the model as usual and then compute the loss afterwards manually. Or just ignore the default computed loss and compute my own loss myself and override it.\r\n\r\nMy own loss computation can then still be like this:\r\nhttps://github.com/huggingface/transformers/blob/9d7d0005b046a95d9d59354714bb6c3547a612fe/src/transformers/models/bert/modeling_bert.py#L1383-L1391\r\n\r\nI think this is even easier. Thank you. Not sure why I did not see this ...", "Yes, the trainer has a `compute_loss` method that is simple to override for that exact purpose. Glad you're satisfied with the outcome, and thanks for opening such a detailed issue.", "I stumbled over this exact method when the `Trainer` was introduced but did not realize that the models still return the raw logits that I then can use for custom loss computation ...\r\nWell, I try to research before opening issues, and looking through the source code often helps understand some details but the code base keeps growing and changing, so it's sometimes hard to keep up and not miss some obvious things.\r\n😄 " ]

[]

[]

[]

[ { "id": 1838412367, "node_id": "MDU6TGFiZWwxODM4NDEyMzY3", "url": "https://api.github.com/repos/huggingface/transformers/labels/model%20card", "name": "model card", "color": "92d5f4", "default": false, "description": "Related to pretrained model cards" } ]

[]

[]

[]

[]

[ "Hey @marcoabrate, \r\nI tried to reproduce your error in this colab without success: https://colab.research.google.com/drive/1nBCEtP773LplNodOSw5OBW-rJ84gizW5?usp=sharing can you check again ?", "You are right @patrickvonplaten \r\nTo reproduce\r\n\r\n```\r\n!pip install -U transformers\r\n\r\nfrom transformers import PegasusTokenizer\r\ntokenizer = PegasusTokenizer.from_pretrained('google/pegasus-large')\r\n\r\n!pip install sentencepiece\r\n\r\ntokenizer = PegasusTokenizer.from_pretrained('google/pegasus-large')\r\n```\r\n\r\nVery weird", "Having the same issue. Although solely in my Jupyter notebook. Running the code from a file is working fine..", "Probably it's a way the Jupyter kernel works. Indeed, if you restart the kernel and install `sentencepiece` before is working.", "as @marcoabrate said, restated kernel at my project and without code changes everything started working", "I can confirm that this solution is still valid today. I encountered the same issue today.\r\nI solved it by adding ` !pip install sentencepiece` and then fully restart the Jupiter environment and rerun.", "I am unable to install the SentencePiece library, this is the error i get when i do pip3 install sentencepiece:\r\n\r\nerror: legacy-install-failure\r\n", "Restarting the kernel and using\r\n!pip install Transformers==3.2.0 instead of !pip install Transformers, worked for me", "After using !pip install sentencepiece just restart the kernel and run the cells. It will work fine\r\n" ]

[]

[]

[]

[]

[]

[ "Thanks for working on this @jplu. I think we should take the opportunity to think about this issue: https://github.com/huggingface/transformers/issues/8793.\r\n\r\nThe problem with the `add_pooling_layer` option and how it's currently done in PyTorch models is that when doing a training initialized from a model checkpoints that *contains the pooling layer*, like `bert-base-cased`:\r\n\r\n```py\r\nmodel = BertForMaskedLM.from_pretrained(\"bert-base-cased\")\r\n# Fine-tune the model on an MLM task\r\n```\r\n\r\nwe're losing the pooling layer doing so. It's not a big deal here as we're doing an MLM task, however, if we want to use that model for a downstream task:\r\n\r\n```py\r\nmodel.save_pretrained(\"bert-base-cased-finetuned-mlm\")\r\nclassifier_model = BertForSequenceClassification.from_pretrained(\"bert-base-cased-finetuned-mlm\")\r\n```\r\nwe're now having a classifier model that has a randomly initialized pooling layer, whereas the weights that were stored in the `bert-base-cased` original checkpoint would have been better than a randomly initialized layer.\r\n\r\nThe issue is that right now, we have no way of specifying if we want to keep the pooling layer or not in such a setup. I would argue that controlling it from the configuration would really be useful here, rather than setting it to `add_pooling_layer=False` in architectures that do not need it.\r\n\r\ncc @jplu @sgugger @patrickvonplaten ", "Indeed, it starts to be more complicated than we thought at the beginning, but the case you are raising is a very good one!!\r\n\r\nI think that controlling this from the config to have the same behavior would be more flexible, I +1 this proposal!", "> Thanks for working on this @jplu. I think we should take the opportunity to think about this issue: #8793.\r\n> \r\n> The problem with the `add_pooling_layer` option and how it's currently done in PyTorch models is that when doing a training initialized from a model checkpoints that _contains the pooling layer_, like `bert-base-cased`:\r\n> \r\n> ```python\r\n> model = BertForMaskedLM.from_pretrained(\"bert-base-cased\")\r\n> # Fine-tune the model on an MLM task\r\n> ```\r\n> \r\n> we're losing the pooling layer doing so. It's not a big deal here as we're doing an MLM task, however, if we want to use that model for a downstream task:\r\n> \r\n> ```python\r\n> model.save_pretrained(\"bert-base-cased-finetuned-mlm\")\r\n> classifier_model = BertForSequenceClassification.from_pretrained(\"bert-base-cased-finetuned-mlm\")\r\n> ```\r\n> \r\n> we're now having a classifier model that has a randomly initialized pooling layer, whereas the weights that were stored in the `bert-base-cased` original checkpoint would have been better than a randomly initialized layer.\r\n> \r\n> The issue is that right now, we have no way of specifying if we want to keep the pooling layer or not in such a setup. I would argue that controlling it from the configuration would really be useful here, rather than setting it to `add_pooling_layer=False` in architectures that do not need it.\r\n> \r\n> cc @jplu @sgugger @patrickvonplaten\r\n\r\nI remember that we were thinking about adding a config param for `add_pooling_layer` for PT: https://github.com/huggingface/transformers/pull/7272 and decided not to. I still think the cleaner solution is to **not** add a config param because it's a very weird use-case IMO. Why wouldn't the user just use a `BertForPreTraining` model for his use case? But I'm also fine with adding a config param instead. It's not a big deal to me...but in this case I'd definitely prefer to not add it to the general `PretrainedConfig`, but to each model's config.", "Good point regarding the `BertForPreTraining`. I think this is a use-case (you want to keep a layer from another architecture) where you would want to build your own architectures for that, to have complete control over the layers.\r\n\r\nI think we might be missing some documentation on how to do that, and on how creating an architecture that inherits from `PreTrainedModel` works, but this is a discussion for another time.\r\n\r\nOk to keep it this way.", "LGTM for me!" ]