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 } ]

[]

[ "I've read your code, and I don't see anything strange in it (I hope). It seems to be an error due to training a (whatever) model on TPUs rather than Transformers.\r\n\r\nDo you see [this](https://github.com/tensorflow/tensorflow/issues/29896) issue reported in TensorFlow's GitHub? It seems to be the same error, and someone gives indications about how to resolve.\r\n\r\n> ## Questions & Help\r\n> Hi,\r\n> \r\n> I am trying to fine-tune TF BERT on Imdb dataset on Colab TPU. Here is the full notebook:\r\n> \r\n> https://colab.research.google.com/drive/16ZaJaXXd2R1gRHrmdWDkFh6U_EB0ln0z\r\n> \r\n> Can anyone help me what I am doing wrong?\r\n> Thanks", "Thanks for the help. Actually my code was inspired from [this colab notebook. This](https://colab.research.google.com/github/CyberZHG/keras-bert/blob/master/demo/tune/keras_bert_classification_tpu.ipynb) notebook works perfectly but there is one major difference might be causing the problem:\r\n\r\n- I am force installing tensorflow 2.x for transformer notebook because transformer works only for TF>=2.0 but colab uses TF 1.15 on colab for TPUs\r\n\r\nSo I went to GCP and used a TPU for TF 2.x. The error changed to this:\r\n\r\n\r\nBut yes, you are right this issue might be related to TF in general.", "In the official docs of GCP [here](https://cloud.google.com/tpu/docs/supported-versions), they show the current list of supported TensorFlow and Cloud TPU versions; _only_ **TensorFlow 1.13, 1.14 and 1.15 are supported** (they don't mention TensorFlow 2.0).\r\nI think it's why you have the problem you've highlighted.\r\n\r\n> Thanks for the help. Actually my code was inspired from [this colab notebook. This](https://colab.research.google.com/github/CyberZHG/keras-bert/blob/master/demo/tune/keras_bert_classification_tpu.ipynb) notebook works perfectly but there is one major difference might be causing the problem:\r\n> \r\n> * I am force installing tensorflow 2.x for transformer notebook because transformer works only for TF>=2.0 but colab uses TF 1.15 on colab for TPUs\r\n> \r\n> So I went to GCP and used a TPU for TF 2.x. The error changed to this:\r\n> \r\n> \r\n> But yes, you are right this issue might be related to TF in general.", "tpu supports . tf2.1", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "Can you remind why you moved those options to initialization vs. at `encode` time?\r\n\r\nIs that a hard requirement of the native implem?", "Sure! The native implementation doesn't have `kwargs` so we need to define a static interface with pre-defined function arguments. This means that the configuration of the tokenizer is done by initializing its various parts and attaching them. There would be some unwanted overhead in doing this every time we `encode`.\r\nI think we generally don't need to change the behavior of the tokenizer while using it, so this shouldn't be a problem. Plus, I think it makes the underlying functions clearer, and easier to use.", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2211?src=pr&el=h1) Report\n> Merging [#2211](https://codecov.io/gh/huggingface/transformers/pull/2211?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/81db12c3ba0c2067f43c4a63edf5e45f54161042?src=pr&el=desc) will **increase** coverage by `<.01%`.\n> The diff coverage is `74.62%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2211?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2211 +/- ##\n==========================================\n+ Coverage 73.54% 73.54% +<.01% \n==========================================\n Files 87 87 \n Lines 14789 14919 +130 \n==========================================\n+ Hits 10876 10972 +96 \n- Misses 3913 3947 +34\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2211?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/\\_\\_init\\_\\_.py](https://codecov.io/gh/huggingface/transformers/pull/2211/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9fX2luaXRfXy5weQ==) | `98.79% <100%> (ø)` | :arrow_up: |\n| [src/transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2211/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHMucHk=) | `88.13% <67.01%> (-3.96%)` | :arrow_down: |\n| [src/transformers/tokenization\\_gpt2.py](https://codecov.io/gh/huggingface/transformers/pull/2211/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fZ3B0Mi5weQ==) | `96.21% <93.75%> (-0.37%)` | :arrow_down: |\n| [src/transformers/tokenization\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/2211/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fYmVydC5weQ==) | `96.2% <94.73%> (-0.15%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2211?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2211?src=pr&el=footer). Last update [81db12c...e6ec24f](https://codecov.io/gh/huggingface/transformers/pull/2211?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Testing the fast tokenizers is not trivial. Doing tokenizer-specific tests is alright, but as of right now the breaking changes make it impossible to implement the fast tokenizers in the common tests pipeline.\r\n\r\nIn order to do so, and to obtain full coverage of the tokenizers, we would have to either:\r\n- Split the current common tests in actual unit tests, rather than integration tests. As the arguments are passed in the `from_pretrained` (for the rust tokenizers) method rather than the `encode` (for the python tokenizers) method, having several chained calls to `encode` with varying arguments implies a big refactor of each test to test the rust tokenizers as well. Splitting those into unit tests would ease this task.\r\n- Copy/Paste the common tests into a separate file (not ideal).\r\n\r\nCurrently there are some integration tests for the rust GPT-2 and rust BERT tokenizers, which may offer sufficient coverage for this PR. We would need the aforementioned refactor to have full coverage, which can be attended to in a future PR. ", "I really like it. Great work @n1t0 \r\n\r\nI think now we should try to work on the other (python) tokenizers and see if we can find a middle ground behavior where they can both behave rather similarly, in particular for tests.\r\n\r\nAlso an open question: should we keep the \"slow\" python tokenizers that are easy to inspect? Could make sense, maybe renaming them to `BertTokenizerPython` for instance.", "Ok, I had to rewrite the whole history after the few restructuration PRs that have been merged. \r\n\r\nSince Python 2 has been dropped, I added `tokenizers` as a dependency (It is working for Python 3.5+).\r\nWe should now be ready to merge!\r\n\r\nWe should clearly keep the Python tokenizers and deprecate them slowly. In the end, I don't mind keeping them, but I'd like to avoid having to maintain both, especially if their API differ.", "This is ready for the final review!", "Ok, this is great, merging!" ]

[]

[ { "login": "n1t0", "id": 1217986, "node_id": "MDQ6VXNlcjEyMTc5ODY=", "avatar_url": "https://avatars.githubusercontent.com/u/1217986?v=4", "gravatar_id": "", "url": "https://api.github.com/users/n1t0", "html_url": "https://github.com/n1t0", "followers_url": "https://api.github.com/users/n1t0/followers", "following_url": "https://api.github.com/users/n1t0/following{/other_user}", "gists_url": "https://api.github.com/users/n1t0/gists{/gist_id}", "starred_url": "https://api.github.com/users/n1t0/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/n1t0/subscriptions", "organizations_url": "https://api.github.com/users/n1t0/orgs", "repos_url": "https://api.github.com/users/n1t0/repos", "events_url": "https://api.github.com/users/n1t0/events{/privacy}", "received_events_url": "https://api.github.com/users/n1t0/received_events", "type": "User", "site_admin": false } ]

[ "Follow sentencepiece github or Bert tensorflow GitHub. You will have some\nfeedback\n\nOn Wed, Dec 18, 2019 at 07:52 Younggyun Hahm <[email protected]>\nwrote:\n\n> ❓ Questions & Help\n>\n> I would like to train a new BERT model.\n> There are some way to train BERT tokenizer (a.k.a. wordpiece tokenizer) ?\n>\n> —\n> You are receiving this because you are subscribed to this thread.\n> Reply to this email directly, view it on GitHub\n> <https://github.com/huggingface/transformers/issues/2210?email_source=notifications&email_token=AIEAE4BMLLHVIADDR5PGZ63QZFQ27A5CNFSM4J4DE7PKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IBGFX2A>,\n> or unsubscribe\n> <https://github.com/notifications/unsubscribe-auth/AIEAE4CUY2ESKVEH4IPDL63QZFQ27ANCNFSM4J4DE7PA>\n> .\n>\n", "If you want to see some examples of custom implementation of **tokenizers** into Transformers' library, you can see how they have implemented [Japanese Tokenizer](https://github.com/huggingface/transformers/blob/master/transformers/tokenization_bert_japanese.py).\r\n\r\nIn general, you can read more information about adding a new model into Transformers [here](https://github.com/huggingface/transformers/blob/30968d70afedb1a9815164737cdc3779f2f058fe/templates/adding_a_new_model/README.md).\r\n\r\n> ## Questions & Help\r\n> I would like to train a new BERT model.\r\n> There are some way to train BERT tokenizer (a.k.a. wordpiece tokenizer) ?", "Checkout the [**`tokenizers`**](https://github.com/huggingface/tokenizers) repo.\r\n\r\nThere's an example of how to train a WordPiece tokenizer: https://github.com/huggingface/tokenizers/blob/master/bindings/python/examples/train_bert_wordpiece.py\r\n\r\n\r\n\r\n", "Hi @julien-c `tokenizers` package is great, but I found an issue when using the resulting tokenizer later with `transformers`.\r\n\r\nAssume I have this:\r\n```\r\nfrom tokenizers import BertWordPieceTokenizer\r\ninit_tokenizer = BertWordPieceTokenizer(vocab=vocab)\r\ninit_tokenizer.save(\"./my_local_tokenizer\")\r\n```\r\n\r\nWhen I am trying to load the file:\r\n```\r\nfrom transformers import AutoTokenizer\r\ntokenizer = AutoTokenizer.from_pretrained(\"./my_local_tokenizer\")\r\n```\r\nan error is thrown:\r\n```\r\nValueError: Unrecognized model in .... Should have a `model_type` key in its config.json, or contain one of the following strings in its name: ...\r\n```\r\n\r\nSeems format used by tokenizers is a single json file, whereas when I save transformers tokenizer it creates a dir with [config.json, special tokens map.json and vocab.txt].\r\n\r\ntransformers.__version__ = '4.15.0'\r\ntokenizers.__version__ '0.10.3'\r\n\r\nCan you please give me some hints how to fix this? Thx in advance", "@tkornuta better to open a post on the forum, but tagging @SaulLu for visibility", "Thanks for the ping julien-c! \r\n\r\n@tkornuta Indeed, for this kind of questions the [forum](https://discuss.huggingface.co/) is the best place to ask them: it also allows other users who would ask the same question as you to benefit from the answer. :relaxed: \r\n\r\nYour use case is indeed very interesting! With your current code, you have a problem because `AutoTokenizer` has no way of knowing which Tokenizer object we want to use to load your tokenizer since you chose to create a new one from scratch.\r\n\r\nSo in particular to instantiate a new `transformers` tokenizer with the Bert-like tokenizer you created with the `tokenizers` library you can do:\r\n\r\n```python\r\nfrom transformers import BertTokenizerFast\r\n\r\nwrapped_tokenizer = BertTokenizerFast(\r\n tokenizer_file=\"./my_local_tokenizer\",\r\n do_lower_case = FILL_ME,\r\n unk_token = FILL_ME,\r\n sep_token = FILL_ME,\r\n pad_token = FILL_ME,\r\n cls_token = FILL_ME,\r\n mask_token = FILL_ME,\r\n tokenize_chinese_chars =FILL_ME,\r\n strip_accents = FILL_ME\r\n)\r\n```\r\nNote: You will have to manually carry over the same parameters (`unk_token`, `strip_accents`, etc) that you used to initialize `BertWordPieceTokenizer` in the initialization of `BertTokenizerFast`.\r\n\r\nI refer you to the [section \"Building a tokenizer, block by block\" of the course ](https://huggingface.co/course/chapter6/9?fw=pt#building-a-wordpiece-tokenizer-from-scratch) where we explained how you can build a tokenizer from scratch with the `tokenizers` library and use it to instantiate a new tokenizer with the `transformers` library. We have even treated the example of a Bert type tokenizer in this chapter :smiley:.\r\n\r\nMoreover, if you just want to generate a new vocabulary for BERT tokenizer by re-training it on a new dataset, the easiest way is probably to use the `train_new_from_iterator` method of a fast `transformers` tokenizer which is explained in the [section \"Training a new tokenizer from an old one\" of our course](https://huggingface.co/course/chapter6/2?fw=pt). :blush: \r\n\r\nI hope this can help you!\r\n \r\n\r\n", "Hi @SaulLu thanks for the answer! I managed to find the solution here:\r\nhttps://huggingface.co/docs/transformers/fast_tokenizer\r\n\r\n```\r\n# 1st solution: Load the HF.tokenisers tokenizer.\r\nloaded_tokenizer = Tokenizer.from_file(decoder_tokenizer_path)\r\n# \"Wrap\" it with HF.transformers tokenizer.\r\ntokenizer = PreTrainedTokenizerFast(tokenizer_object=loaded_tokenizer)\r\n\r\n# 2nd solution: Load from tokenizer file\r\ntokenizer = PreTrainedTokenizerFast(tokenizer_file=decoder_tokenizer_path)\r\n```\r\n\r\nNow I also see that somehow I have missed the information at the bottom of the section that you mention on building tokenizer that is also stating that - sorry.\r\n\r\n\r\n", "Hey @SaulLu sorry for bothering, but struggling with yet another problem/question.\r\n\r\nWhen I am loading the tokenizer created in HF.tokenizers my special tokens are \"gone\", i.e. \r\n```\r\n# Load from tokenizer file\r\ntokenizer = PreTrainedTokenizerFast(tokenizer_file=decoder_tokenizer_path)\r\ntokenizer.pad_token # <- this is None\r\n```\r\n\r\nWithout this when I am using padding:\r\n```\r\nencoded = tokenizer.encode(input, padding=True) # <- raises error - lack of pad_token\r\n```\r\n\r\nI can add them to tokenizer from HF.transformers e.g. like this:\r\n```\r\ntokenizer.add_special_tokens({'pad_token': '[PAD]'}) # <- this works!\r\n```\r\n\r\nIs there a similar method for setting special tokens to tokenizer in HF.tokenizers that will enable me to load the tokenizer in HF.transformers?\r\n\r\nI have all tokens in my vocabulary and tried the following\r\n````\r\n# Pass as arguments to constructor:\r\ninit_tokenizer = BertWordPieceTokenizer(vocab=vocab) \r\n #special_tokens=[\"[UNK]\", \"[CLS]\", \"[SEP]\", \"[PAD]\", \"[MASK]\"]) <- error: wrong keyword\r\n #bos_token = \"[CLS]\", eos_token = \"[SEP]\", unk_token = \"[UNK]\", sep_token = \"[SEP]\", <- wrong keywords\r\n #pad_token = \"[PAD]\", cls_token = \"[CLS]\", mask_token = \"[MASK]\", <- wrong keywords\r\n# Use tokenizer.add_special_tokens() method:\r\n#init_tokenizer.add_special_tokens({'pad_token': '[PAD]'}) <- error: must be a list\r\n#init_tokenizer.add_special_tokens([\"[PAD]\", \"[CLS]\", \"[SEP]\", \"[UNK]\", \"[MASK]\", \"[BOS]\", \"[EOS]\"]) <- doesn't work (means error when calling encode(padding=True))\r\n#init_tokenizer.add_special_tokens(['[PAD]']) # <- doesn't work\r\n\r\n# Set manually.\r\n#init_tokenizer.pad_token = \"[PAD]\" # <- doesn't work\r\ninit_tokenizer.pad_token_id = vocab[\"[PAD]\"] # <- doesn't work\r\n```\r\n\r\nAm I missing something obvious? Thanks in advance!", "@tkornuta, I'm sorry I missed your second question!\r\n\r\nThe `BertWordPieceTokenizer` class is just an helper class to build a `tokenizers.Tokenizers` object with the architecture proposed by the Bert's authors. The `tokenizers` library is used to build tokenizers and the `transformers` library to wrap these tokenizers by adding useful functionality when we wish to use them with a particular model (like identifying the padding token, the separation token, etc). \r\n\r\nTo not miss anything, I would like to comment on several of your remarks\r\n\r\n#### Remark 1\r\n> [@tkornuta] When I am loading the tokenizer created in HF.tokenizers my special tokens are \"gone\", i.e.\r\n\r\nTo carry your special tokens in your `HF.transformers` tokenizer, I refer you to this section of my previous answer\r\n> [@SaulLu] So in particular to instantiate a new transformers tokenizer with the Bert-like tokenizer you created with the tokenizers library you can do:\r\n> ```python\r\n> from transformers import BertTokenizerFast\r\n>\r\n> wrapped_tokenizer = BertTokenizerFast(\r\n> tokenizer_file=\"./my_local_tokenizer\",\r\n> do_lower_case = FILL_ME,\r\n> unk_token = FILL_ME,\r\n> sep_token = FILL_ME,\r\n> pad_token = FILL_ME,\r\n> cls_token = FILL_ME,\r\n> mask_token = FILL_ME,\r\n> tokenize_chinese_chars =FILL_ME,\r\n> strip_accents = FILL_ME\r\n> )\r\n>```\r\n> Note: You will have to manually carry over the same parameters (unk_token, strip_accents, etc) that you used to initialize BertWordPieceTokenizer in the initialization of BertTokenizerFast.\r\n\r\n\r\n#### Remark 2\r\n> [@tkornuta] Is there a similar method for setting special tokens to tokenizer in HF.tokenizers that will enable me to load the tokenizer in HF.transformers?\r\n\r\nNothing prevents you from overloading the `BertWordPieceTokenizer` class in order to define the properties that interest you. On the other hand, there will be no automatic porting of the values of these new properties in the `HF.transformers` tokenizer properties (you have to use the method mentioned below or the methode `.add_special_tokens({'pad_token': '[PAD]'})` after having instanciated your `HF.transformers` tokenizer ).\r\n\r\nDoes this answer your questions? :relaxed: ", "Hi @SaulLu yeah, I was asking about this \"automatic porting of special tokens\". As I set them already when training the tokenizer in HF.tokenizers, hence I am really not sure why they couldn't be imported automatically when loading tokenizer in HF.transformers...\r\n\r\nAnyway, thanks, your answer is super useful. \r\n\r\nMoreover, thanks for the hint pointing to forum! I will use it next time for sure! :)" ]

[]

[]

[ "duplicate to 2208" ]

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

[]

[ "Do you mean the one already into Transformers in the [glue.py](https://github.com/huggingface/transformers/blob/d46147294852694d1dc701c72b9053ff2e726265/transformers/data/processors/glue.py) at line 30 or a different function? \r\n\r\n> glue_convert_examples_to_features", "A different one. Does this proposal makes sense?", "> A different one. Does this proposal makes sense?\r\n\r\nDifferent in which way? Describe to us please the goal and an high-level implementation.", "Thanks for the reply! First of all I just want to clarify that I am not sure that my suggestion makes indeed sense. I will try to clarify: 1) This implementation ``` def glue_convert_examples_to_features(examples, tokenizer,``` is for glue datasets, so it does not cover what I suggest 2) in line 112 we see that it can only support \"classification\" and \"regression\". The classification is in sentence level. I want classification in tokens level.\r\n\r\nSo to my understanding, this function will give back one ```features``` tensor that is 1-1 correspondence with the one label for this sentence. In my case we would like n ```features``` tensors, that will have 1-1 correspondence with the labels for this sentence, where n the number of tokens of the sentence.", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2207?src=pr&el=h1) Report\n> Merging [#2207](https://codecov.io/gh/huggingface/transformers/pull/2207?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/f061606277322a013ec2d96509d3077e865ae875?src=pr&el=desc) will **increase** coverage by `1.13%`.\n> The diff coverage is `83.33%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2207?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2207 +/- ##\n==========================================\n+ Coverage 80.32% 81.46% +1.13% \n==========================================\n Files 122 122 \n Lines 18342 18345 +3 \n==========================================\n+ Hits 14734 14945 +211 \n+ Misses 3608 3400 -208\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2207?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2207/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2ZpbGVfdXRpbHMucHk=) | `70.42% <83.33%> (-1.01%)` | :arrow_down: |\n| [transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/2207/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX29wZW5haS5weQ==) | `81.51% <0%> (+1.32%)` | :arrow_up: |\n| [transformers/modeling\\_ctrl.py](https://codecov.io/gh/huggingface/transformers/pull/2207/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2N0cmwucHk=) | `96.47% <0%> (+2.2%)` | :arrow_up: |\n| [transformers/modeling\\_xlnet.py](https://codecov.io/gh/huggingface/transformers/pull/2207/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3hsbmV0LnB5) | `74.54% <0%> (+2.32%)` | :arrow_up: |\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/2207/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `71.76% <0%> (+12.35%)` | :arrow_up: |\n| [transformers/tests/modeling\\_tf\\_common\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2207/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2NvbW1vbl90ZXN0LnB5) | `97.41% <0%> (+17.24%)` | :arrow_up: |\n| [transformers/modeling\\_tf\\_pytorch\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2207/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3B5dG9yY2hfdXRpbHMucHk=) | `92.15% <0%> (+83%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2207?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2207?src=pr&el=footer). Last update [f061606...14cc752](https://codecov.io/gh/huggingface/transformers/pull/2207?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "is this issue related to https://github.com/scipy/scipy/issues/11237, which also started happening yesterday.", "Indeed, this is related to that issue. I've just tested on the CircleCI machine directly, the segmentation fault happens when importing torch after tensorflow, when scipy is installed on the machine.", "@LysandreJik I get this error on transfomers 2.2.2 on PyPi. When it will be updated?\r\n\r\n```\r\n>>> import transformers\r\nSegmentation fault (core dumped)\r\n```\r\n\r\n```\r\nroot@261246f307ae:~/src# python --version\r\nPython 3.6.8\r\n```", "This is due to an upstream issue related to scipy 1.4.0. Please pin your scipy version to one earlier than 1.4.0 and you should see this segmentation fault resolved.", "@LysandreJik thank you this issue is destroying everything 🗡 ", "Did pinning the scipy version fix your issue?", "@LysandreJik I'm now pinning `scipy==1.3.3` that is the latest version before RC1.4.x", "I confirm that with 1.3.3 works, but they have just now pushed `scipy==1.4.1`. We have tested it and it works as well.\r\n\r\n - https://github.com/scipy/scipy/issues/11237#issuecomment-567550894\r\nThank you!", "Glad you could make it work!" ]

[]

[]

[ "As said in #2117 by @rlouf (an author of Transformers), **at the moment** you can use `PreTrainedEncoderDecoder` with only **BERT** model both as encoder and decoder.\r\n\r\nIn more details, he said: \"_Indeed, as I specified in the article, PreTrainedEncoderDecoder only works with BERT as an encoder and BERT as a decoder. GPT2 shouldn't take too much work to adapt, but we haven't had the time to do it yet. Try PreTrainedEncoderDecoder.from_pretrained('bert-base-uncased', 'bert-base-uncased') should work. Let me know if it doesn't._\".\r\n\r\n> ## Bug\r\n> Model I am using (Bert, XLNet....):\r\n> \r\n> Language I am using the model on (English, Chinese....):\r\n> \r\n> The problem arise when using:\r\n> \r\n> * [X ] the official example scripts: (give details)\r\n> * [ ] my own modified scripts: (give details)\r\n> \r\n> The tasks I am working on is:\r\n> \r\n> * [ ] an official GLUE/SQUaD task: (give the name)\r\n> * [ X] my own task or dataset: (give details)\r\n> \r\n> ## To Reproduce\r\n> Steps to reproduce the behavior:\r\n> \r\n> 1. Error while doing inference.\r\n> \r\n> ```\r\n> from transformers import PreTrainedEncoderDecoder, BertTokenizer\r\n> model = PreTrainedEncoderDecoder.from_pretrained('bert-base-uncased','gpt2')\r\n> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')\r\n> encoder_input_ids=tokenizer.encode(\"Hi How are you\")\r\n> import torch\r\n> ouput = model(torch.tensor( encoder_input_ids).unsqueeze(0))\r\n> ```\r\n> \r\n> and the error is\r\n> \r\n> ```\r\n> TypeError: forward() missing 1 required positional argument: 'decoder_input_ids'\r\n> ```\r\n> \r\n> During inference why is decoder input is expected ?\r\n> \r\n> Let me know if I'm missing anything?\r\n> \r\n> ## Environment\r\n> OS: ubuntu\r\n> Python version: 3.6\r\n> PyTorch version:1.3.0\r\n> PyTorch Transformers version (or branch):2.2.0\r\n> Using GPU ? Yes\r\n> Distributed of parallel setup ? No\r\n> Any other relevant information:\r\n> \r\n> ## Additional context", "@TheEdoardo93 it doesn't matter whether it is GPT2 or bert. Both has the same error :\r\nI'm trying to play with GPT2 that's why I pasted my own code.\r\n\r\nUsing BERT as an Encoder and Decoder\r\n```\r\n>>> model = PreTrainedEncoderDecoder.from_pretrained('bert-base-uncased','bert-base-uncased')\r\n>>> ouput = model(torch.tensor( encoder_input_ids).unsqueeze(0))\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"/home/guest_1/.local/lib/python3.6/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n result = self.forward(*input, **kwargs)\r\nTypeError: forward() missing 1 required positional argument: 'decoder_input_ids'\r\n```", "First of all, authors of Transformers are working on the implementation of `PreTrainedEncoderDecoder` object, so it's not a definitive implementation, e.g. the code lacks of the implementation of some methods. Said so, I've tested your code and I've revealed how to working with `PreTrainedEncoderDecoder` **correctly without bugs**. You can see my code below.\r\n\r\nIn brief, your problem occurs because you have not passed _all_ arguments necessary to the `forward` method. By looking at the source code [here](https://github.com/huggingface/transformers/blob/master/transformers/modeling_encoder_decoder.py), you can see that this method accepts **two** parameters: `encoder_input_ids` and `decoder_input_ids`. In your code, you've passed _only one_ parameter, and the Python interpreter associates your `encoder_input_ids` to the `encoder_input_ids` of the `forward` method, but you don't have supply a value for `decoder_input_ids` of the `forward` method, and this is the cause that raise the error.\r\n\r\n```\r\nPython 3.6.9 |Anaconda, Inc.| (default, Jul 30 2019, 19:07:31) \r\n[GCC 7.3.0] on linux\r\nType \"help\", \"copyright\", \"credits\" or \"license\" for more information.\r\n>>> import transformers\r\n/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:541: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\r\n _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\r\n/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:542: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\r\n _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\r\n/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:543: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\r\n _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\r\n/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:544: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\r\n _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\r\n/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:545: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\r\n _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\r\n/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:550: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\r\n np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\r\n>>> from transformers import PreTrainedEncoderDecoder\r\n>>> from transformers import BertTokenizer\r\n>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')\r\n>>> model = PreTrainedEncoderDecoder.from_pretrained('bert-base-uncased', 'bert-base-uncased')\r\n>>> text='Hi How are you'\r\n>>> import torch\r\n>>> input_ids = torch.tensor(tokenizer.encode(text)).unsqueeze(0)\r\n>>> input_ids\r\ntensor([[ 101, 7632, 2129, 2024, 2017, 102]])\r\n>>> output = model(input_ids) # YOUR PROBLEM IS HERE\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n result = self.forward(*input, **kwargs)\r\nTypeError: forward() missing 1 required positional argument: 'decoder_input_ids'\r\n>>> output = model(input_ids, input_ids) # SOLUTION TO YOUR PROBLEM\r\n>>> output\r\n(tensor([[[ -6.3390, -6.3664, -6.4600, ..., -5.5354, -4.1787, -5.8384],\r\n [ -6.3550, -6.3077, -6.4661, ..., -5.3516, -4.1338, -4.0742],\r\n [ -6.7090, -6.6050, -6.6682, ..., -5.9591, -4.7142, -3.8219],\r\n [ -7.7608, -7.5956, -7.6634, ..., -6.8113, -5.7777, -4.1638],\r\n [ -8.6462, -8.5767, -8.6366, ..., -7.9503, -6.5382, -5.0959],\r\n [-12.8752, -12.3775, -12.2770, ..., -10.0880, -10.7659, -9.0092]]],\r\n grad_fn=<AddBackward0>), tensor([[[ 0.0929, -0.0264, -0.1224, ..., -0.2106, 0.1739, 0.1725],\r\n [ 0.4074, -0.0593, 0.5523, ..., -0.6791, 0.6556, -0.2946],\r\n [-0.2116, -0.6859, -0.4628, ..., 0.1528, 0.5977, -0.9102],\r\n [ 0.3992, -1.3208, -0.0801, ..., -0.3213, 0.2557, -0.5780],\r\n [-0.0757, -1.3394, 0.1816, ..., 0.0746, 0.4032, -0.7080],\r\n [ 0.5989, -0.2841, -0.3490, ..., 0.3042, -0.4368, -0.2097]]],\r\n grad_fn=<NativeLayerNormBackward>), tensor([[-9.3097e-01, -3.3807e-01, -6.2162e-01, 8.4082e-01, 4.4154e-01,\r\n -1.5889e-01, 9.3273e-01, 2.2240e-01, -4.3249e-01, -9.9998e-01,\r\n -2.7810e-01, 8.9449e-01, 9.8638e-01, 6.4763e-02, 9.6649e-01,\r\n -7.7835e-01, -4.4046e-01, -5.9515e-01, 2.7585e-01, -7.4638e-01,\r\n 7.4700e-01, 9.9983e-01, 4.4468e-01, 2.8673e-01, 3.6586e-01,\r\n 9.7642e-01, -8.4343e-01, 9.6599e-01, 9.7235e-01, 7.2667e-01,\r\n -7.5785e-01, 9.2892e-02, -9.9089e-01, -1.7004e-01, -6.8200e-01,\r\n -9.9283e-01, 2.6244e-01, -7.9871e-01, 2.3397e-02, 4.6413e-02,\r\n -9.3371e-01, 2.7699e-01, 9.9995e-01, -3.2671e-01, 2.1108e-01,\r\n -2.0636e-01, -1.0000e+00, 1.9622e-01, -9.3330e-01, 6.8736e-01,\r\n 6.4731e-01, 5.3773e-01, 9.2759e-02, 4.1069e-01, 4.0360e-01,\r\n 1.9002e-01, -1.7049e-01, 7.5259e-03, -2.0453e-01, -5.7574e-01,\r\n -5.3062e-01, 3.9367e-01, -7.0627e-01, -9.2865e-01, 6.8820e-01,\r\n 3.2698e-01, -3.3506e-02, -1.2323e-01, -1.5304e-01, -1.8077e-01,\r\n 9.3398e-01, 2.6375e-01, 3.7505e-01, -8.9548e-01, 1.1777e-01,\r\n 2.2054e-01, -6.3351e-01, 1.0000e+00, -6.9228e-01, -9.8653e-01,\r\n 6.9799e-01, 4.0303e-01, 5.2453e-01, 2.3217e-01, -1.2151e-01,\r\n -1.0000e+00, 5.6760e-01, 2.9295e-02, -9.9318e-01, 8.3171e-02,\r\n 5.2939e-01, -2.3176e-01, -1.5694e-01, 4.9278e-01, -4.2614e-01,\r\n -3.8079e-01, -2.6060e-01, -6.9055e-01, -1.7180e-01, -1.9810e-01,\r\n -2.7986e-02, -7.2085e-02, -3.7635e-01, -3.7743e-01, 1.3508e-01,\r\n -4.3892e-01, -6.1321e-01, 1.7726e-01, -3.5434e-01, 6.4734e-01,\r\n 4.0373e-01, -2.8194e-01, 4.5104e-01, -9.7876e-01, 6.1044e-01,\r\n -2.3526e-01, -9.9035e-01, -5.1350e-01, -9.9280e-01, 6.8329e-01,\r\n -2.1623e-01, -1.4641e-01, 9.8273e-01, 3.7345e-01, 4.8171e-01,\r\n -5.6467e-03, -7.3005e-01, -1.0000e+00, -7.2252e-01, -5.1978e-01,\r\n 7.0765e-02, -1.5036e-01, -9.8355e-01, -9.7384e-01, 5.8453e-01,\r\n 9.6710e-01, 1.4193e-01, 9.9981e-01, -2.1194e-01, 9.6675e-01,\r\n 2.3627e-02, -4.1555e-01, 1.9872e-01, -4.0593e-01, 6.5180e-01,\r\n 6.1598e-01, -6.8750e-01, 7.9808e-02, -2.0437e-01, 3.4504e-01,\r\n -6.7176e-01, -1.3692e-01, -2.7750e-01, -9.6740e-01, -3.6698e-01,\r\n 9.6934e-01, -2.5050e-01, -6.9297e-01, 4.8327e-01, -1.4613e-01,\r\n -5.1224e-01, 8.8387e-01, 6.9173e-01, 3.8395e-01, -1.7536e-01,\r\n 3.8873e-01, -4.3011e-03, 6.1876e-01, -8.9292e-01, 3.4243e-02,\r\n 4.5193e-01, -2.4782e-01, -4.7402e-01, -9.8375e-01, -3.1763e-01,\r\n 5.9109e-01, 9.9284e-01, 7.9634e-01, 2.4601e-01, 6.1729e-01,\r\n -1.8376e-01, 6.8750e-01, -9.7083e-01, 9.8624e-01, -2.0573e-01,\r\n 2.0418e-01, 4.1400e-01, 1.9102e-01, -9.1718e-01, -3.5273e-01,\r\n 8.9628e-01, -5.6812e-01, -8.9552e-01, -3.5567e-02, -4.9052e-01,\r\n -4.3559e-01, -6.2323e-01, 5.6863e-01, -2.6201e-01, -3.1324e-01,\r\n -1.2852e-02, 9.4585e-01, 9.8664e-01, 8.3363e-01, -2.4392e-01,\r\n 7.3786e-01, -9.4466e-01, -5.2720e-01, -1.6349e-02, 2.4207e-01,\r\n 3.6905e-02, 9.9638e-01, -5.8095e-01, -7.2046e-02, -9.4418e-01,\r\n -9.8921e-01, -1.0289e-01, -9.3301e-01, -5.3531e-02, -6.8719e-01,\r\n 5.3295e-01, 1.6390e-01, 2.3460e-01, 4.3260e-01, -9.9501e-01,\r\n -7.7318e-01, 2.6342e-01, -3.6949e-01, 4.0245e-01, -1.6657e-01,\r\n 4.5766e-01, 7.4537e-01, -5.8549e-01, 8.4632e-01, 9.3526e-01,\r\n -6.4963e-01, -7.8264e-01, 8.5868e-01, -2.9683e-01, 9.0246e-01,\r\n -6.5124e-01, 9.8896e-01, 8.6732e-01, 8.7014e-01, -9.5627e-01,\r\n -4.1195e-01, -9.1043e-01, -4.5438e-01, 5.7729e-02, -3.6862e-01,\r\n 5.7032e-01, 5.5757e-01, 3.0482e-01, 7.0850e-01, -6.6279e-01,\r\n 9.9909e-01, -5.0139e-01, -9.7001e-01, -2.2370e-01, -1.8440e-02,\r\n -9.9107e-01, 7.2208e-01, 2.4379e-01, 6.9083e-02, -3.2313e-01,\r\n -7.3217e-01, -9.7295e-01, 9.2268e-01, 5.0675e-02, 9.9215e-01,\r\n -8.0247e-02, -9.5682e-01, -4.1637e-01, -9.4549e-01, -2.9790e-01,\r\n -1.5625e-01, 2.4707e-01, -1.8468e-01, -9.7276e-01, 4.7428e-01,\r\n 5.6760e-01, 5.5919e-01, -1.9418e-01, 9.9932e-01, 1.0000e+00,\r\n 9.7844e-01, 9.3669e-01, 9.5284e-01, -9.9929e-01, -4.9083e-01,\r\n 9.9999e-01, -9.7835e-01, -1.0000e+00, -9.5292e-01, -6.5736e-01,\r\n 4.1425e-01, -1.0000e+00, -2.7896e-03, 7.0756e-02, -9.4186e-01,\r\n 2.7960e-01, 9.8389e-01, 9.9658e-01, -1.0000e+00, 8.8289e-01,\r\n 9.6828e-01, -6.2958e-01, 9.3367e-01, -3.7519e-01, 9.8027e-01,\r\n 4.2505e-01, 3.0766e-01, -3.2042e-01, 2.7469e-01, -7.8253e-01,\r\n -8.8309e-01, -1.5604e-01, -3.6222e-01, 9.9091e-01, 3.0116e-02,\r\n -7.8697e-01, -9.4496e-01, 2.2050e-01, -8.4521e-02, -4.8378e-01,\r\n -9.7952e-01, -1.3446e-01, 4.2209e-01, 7.8760e-01, 6.4992e-02,\r\n 2.0492e-01, -7.8143e-01, 2.2120e-01, -5.0228e-01, 3.7149e-01,\r\n 6.5244e-01, -9.4897e-01, -6.0978e-01, -4.8976e-03, -4.6856e-01,\r\n -2.8122e-01, -9.6984e-01, 9.8036e-01, -3.5220e-01, 7.4903e-01,\r\n 1.0000e+00, -1.0373e-01, -9.4037e-01, 6.2856e-01, 1.5745e-01,\r\n -1.1596e-01, 1.0000e+00, 7.2891e-01, -9.8543e-01, -5.3814e-01,\r\n 4.0543e-01, -4.9501e-01, -4.8527e-01, 9.9950e-01, -1.4058e-01,\r\n -2.3799e-01, -1.5841e-01, 9.8467e-01, -9.9180e-01, 9.7240e-01,\r\n -9.5292e-01, -9.8022e-01, 9.8012e-01, 9.5211e-01, -6.6387e-01,\r\n -7.2622e-01, 1.1509e-01, -3.1365e-01, 1.8487e-01, -9.7602e-01,\r\n 7.9482e-01, 5.2428e-01, -1.3540e-01, 9.1377e-01, -9.0275e-01,\r\n -5.2769e-01, 2.8301e-01, -4.9215e-01, 3.4866e-02, 8.3573e-01,\r\n 5.0270e-01, -2.4031e-01, -6.1194e-02, -2.5558e-01, -1.3530e-01,\r\n -9.8688e-01, 2.9877e-01, 1.0000e+00, -5.3199e-02, 4.4522e-01,\r\n -2.4564e-01, 3.8897e-02, -3.7170e-01, 3.6843e-01, 5.1087e-01,\r\n -1.9742e-01, -8.8481e-01, 4.5420e-01, -9.8222e-01, -9.8894e-01,\r\n 8.5417e-01, 1.4674e-01, -3.3154e-01, 9.9999e-01, 4.4333e-01,\r\n 7.1728e-02, 1.6790e-01, 9.6064e-01, 2.3267e-02, 7.2436e-01,\r\n 4.9905e-01, 9.8528e-01, -2.0286e-01, 5.2711e-01, 9.0711e-01,\r\n -5.6147e-01, -3.4452e-01, -6.1113e-01, -8.1268e-02, -9.2887e-01,\r\n 1.0119e-01, -9.7066e-01, 9.7404e-01, 8.2025e-01, 2.9760e-01,\r\n 1.9059e-01, 3.6089e-01, 1.0000e+00, -2.7256e-01, 6.5052e-01,\r\n -6.0092e-01, 9.0897e-01, -9.9819e-01, -9.1409e-01, -3.7810e-01,\r\n 1.2677e-02, -3.9492e-01, -3.0028e-01, 3.4323e-01, -9.7925e-01,\r\n 4.4501e-01, 3.7582e-01, -9.9622e-01, -9.9495e-01, 1.6366e-01,\r\n 9.2522e-01, -1.3063e-02, -9.5314e-01, -7.5003e-01, -6.5409e-01,\r\n 4.1526e-01, -7.6235e-02, -9.6046e-01, 3.2395e-01, -2.7184e-01,\r\n 4.7535e-01, -1.1767e-01, 5.6867e-01, 4.6844e-01, 8.3125e-01,\r\n -2.1505e-01, -2.6495e-01, -4.4479e-02, -8.5166e-01, 8.8927e-01,\r\n -8.9329e-01, -7.7919e-01, -1.5320e-01, 1.0000e+00, -4.3274e-01,\r\n 6.4268e-01, 7.7000e-01, 7.9197e-01, -5.4889e-02, 8.0927e-02,\r\n 7.9722e-01, 2.1034e-01, -1.9189e-01, -4.4749e-01, -8.0585e-01,\r\n -3.5409e-01, 7.0995e-01, 1.2411e-01, 2.0604e-01, 8.3328e-01,\r\n 7.4750e-01, 1.7900e-04, 7.6917e-02, -9.1725e-02, 9.9981e-01,\r\n -2.6801e-01, -8.3787e-02, -4.8642e-01, 1.1836e-01, -3.5603e-01,\r\n -5.8620e-01, 1.0000e+00, 2.2691e-01, 3.2801e-01, -9.9343e-01,\r\n -7.3298e-01, -9.5126e-01, 1.0000e+00, 8.4895e-01, -8.6216e-01,\r\n 6.9319e-01, 5.5441e-01, -1.1380e-02, 8.6958e-01, -1.2449e-01,\r\n -2.8602e-01, 1.8517e-01, 9.2221e-02, 9.6773e-01, -4.5911e-01,\r\n -9.7611e-01, -6.6894e-01, 3.7154e-01, -9.7862e-01, 9.9949e-01,\r\n -5.5391e-01, -2.0926e-01, -3.9404e-01, -2.3863e-02, 6.3624e-01,\r\n -1.0563e-01, -9.8927e-01, -1.4047e-01, 1.2247e-01, 9.7469e-01,\r\n 2.6847e-01, -6.0451e-01, -9.5354e-01, 4.5191e-01, 6.6822e-01,\r\n -7.2218e-01, -9.6438e-01, 9.7538e-01, -9.9165e-01, 5.6641e-01,\r\n 1.0000e+00, 2.2837e-01, -2.8539e-01, 1.6956e-01, -4.6714e-01,\r\n 2.5561e-01, -2.6744e-01, 7.4301e-01, -9.7890e-01, -2.7469e-01,\r\n -1.4162e-01, 2.7886e-01, -7.0853e-02, -5.8891e-02, 8.2879e-01,\r\n 1.9968e-01, -5.4085e-01, -6.8158e-01, 3.7584e-02, 3.5805e-01,\r\n 8.9092e-01, -1.7879e-01, -8.1491e-02, 5.0655e-02, -7.9140e-02,\r\n -9.5114e-01, -1.4923e-01, -3.5370e-01, -9.9994e-01, 7.4321e-01,\r\n -1.0000e+00, 2.1850e-01, -2.5182e-01, -2.2171e-01, 8.7817e-01,\r\n 2.9648e-01, 3.4926e-01, -8.2534e-01, -3.8831e-01, 7.6622e-01,\r\n 8.0938e-01, -2.1051e-01, -3.0882e-01, -7.6183e-01, 2.2523e-01,\r\n -1.4952e-02, 1.5150e-01, -2.1056e-01, 7.3482e-01, -1.5207e-01,\r\n 1.0000e+00, 1.0631e-01, -7.7462e-01, -9.8438e-01, 1.6242e-01,\r\n -1.6337e-01, 1.0000e+00, -9.4196e-01, -9.7149e-01, 3.9827e-01,\r\n -7.2371e-01, -8.6582e-01, 3.0937e-01, -6.4325e-02, -8.1062e-01,\r\n -8.8436e-01, 9.8219e-01, 9.3543e-01, -5.6058e-01, 4.5004e-01,\r\n -3.2933e-01, -5.5851e-01, -6.9835e-02, 6.0196e-01, 9.9111e-01,\r\n 4.1170e-01, 9.1721e-01, 5.9978e-01, -9.4103e-02, 9.7966e-01,\r\n 1.5322e-01, 5.3662e-01, 4.2338e-02, 1.0000e+00, 2.8920e-01,\r\n -9.3933e-01, 2.4383e-01, -9.8948e-01, -1.5036e-01, -9.7242e-01,\r\n 2.8053e-01, 1.1691e-01, 9.0178e-01, -2.1055e-01, 9.7547e-01,\r\n -5.0734e-01, -8.5119e-03, -5.2189e-01, 1.1963e-01, 4.0313e-01,\r\n -9.4529e-01, -9.8752e-01, -9.8975e-01, 4.5711e-01, -4.0753e-01,\r\n 5.8175e-02, 1.1543e-01, 8.6051e-02, 3.6199e-01, 4.3131e-01,\r\n -1.0000e+00, 9.5818e-01, 4.0499e-01, 6.9443e-01, 9.7521e-01,\r\n 6.7153e-01, 4.3386e-01, 2.2481e-01, -9.9118e-01, -9.9126e-01,\r\n -3.1248e-01, -1.4604e-01, 7.9951e-01, 6.1145e-01, 9.2726e-01,\r\n 4.0171e-01, -3.9375e-01, -2.0938e-01, -3.2651e-02, -4.1723e-01,\r\n -9.9582e-01, 4.5682e-01, -9.4401e-02, -9.8150e-01, 9.6766e-01,\r\n -5.5518e-01, -8.0481e-02, 4.4743e-01, -6.0429e-01, 9.7261e-01,\r\n 8.6633e-01, 3.7309e-01, 9.4917e-04, 4.6426e-01, 9.1590e-01,\r\n 9.6965e-01, 9.8799e-01, -4.6592e-01, 8.7146e-01, -3.1116e-01,\r\n 5.1496e-01, 6.7961e-01, -9.5609e-01, 1.3302e-03, 3.6581e-01,\r\n -2.3789e-01, 2.6341e-01, -1.2874e-01, -9.8464e-01, 4.8621e-01,\r\n -1.8921e-01, 6.1015e-01, -4.3986e-01, 2.1561e-01, -3.7115e-01,\r\n -1.5832e-02, -6.9704e-01, -7.3403e-01, 5.7310e-01, 5.0895e-01,\r\n 9.4111e-01, 6.9365e-01, 6.9171e-02, -7.3277e-01, -1.1294e-01,\r\n -4.0168e-01, -9.2587e-01, 9.6638e-01, 2.2207e-02, 1.5029e-01,\r\n 2.8954e-01, -8.5994e-02, 7.4631e-01, -1.5933e-01, -3.5710e-01,\r\n -1.6201e-01, -7.1149e-01, 9.0602e-01, -4.2873e-01, -4.6653e-01,\r\n -5.4765e-01, 7.4640e-01, 2.3966e-01, 9.9982e-01, -4.6795e-01,\r\n -6.4802e-01, -4.1201e-01, -3.4984e-01, 3.5475e-01, -5.4668e-01,\r\n -1.0000e+00, 3.6903e-01, -1.7324e-01, 4.3267e-01, -4.7206e-01,\r\n 6.3586e-01, -5.2151e-01, -9.9077e-01, -1.6597e-01, 2.6735e-01,\r\n 4.5069e-01, -4.3034e-01, -5.6321e-01, 5.7792e-01, 8.8123e-02,\r\n 9.4964e-01, 9.2798e-01, -3.3326e-01, 5.1963e-01, 6.0865e-01,\r\n -4.4019e-01, -6.8129e-01, 9.3489e-01]], grad_fn=<TanhBackward>))\r\n>>> len(output)\r\n3\r\n>>> output[0]\r\ntensor([[[ -6.3390, -6.3664, -6.4600, ..., -5.5354, -4.1787, -5.8384],\r\n [ -6.3550, -6.3077, -6.4661, ..., -5.3516, -4.1338, -4.0742],\r\n [ -6.7090, -6.6050, -6.6682, ..., -5.9591, -4.7142, -3.8219],\r\n [ -7.7608, -7.5956, -7.6634, ..., -6.8113, -5.7777, -4.1638],\r\n [ -8.6462, -8.5767, -8.6366, ..., -7.9503, -6.5382, -5.0959],\r\n [-12.8752, -12.3775, -12.2770, ..., -10.0880, -10.7659, -9.0092]]],\r\n grad_fn=<AddBackward0>)\r\n>>> output[0].shape\r\ntorch.Size([1, 6, 30522])\r\n>>> output[1]\r\ntensor([[[ 0.0929, -0.0264, -0.1224, ..., -0.2106, 0.1739, 0.1725],\r\n [ 0.4074, -0.0593, 0.5523, ..., -0.6791, 0.6556, -0.2946],\r\n [-0.2116, -0.6859, -0.4628, ..., 0.1528, 0.5977, -0.9102],\r\n [ 0.3992, -1.3208, -0.0801, ..., -0.3213, 0.2557, -0.5780],\r\n [-0.0757, -1.3394, 0.1816, ..., 0.0746, 0.4032, -0.7080],\r\n [ 0.5989, -0.2841, -0.3490, ..., 0.3042, -0.4368, -0.2097]]],\r\n grad_fn=<NativeLayerNormBackward>)\r\n>>> output[1].shape\r\ntorch.Size([1, 6, 768])\r\n>>> output[2]\r\ntensor([[-9.3097e-01, -3.3807e-01, -6.2162e-01, 8.4082e-01, 4.4154e-01,\r\n -1.5889e-01, 9.3273e-01, 2.2240e-01, -4.3249e-01, -9.9998e-01,\r\n -2.7810e-01, 8.9449e-01, 9.8638e-01, 6.4763e-02, 9.6649e-01,\r\n -7.7835e-01, -4.4046e-01, -5.9515e-01, 2.7585e-01, -7.4638e-01,\r\n 7.4700e-01, 9.9983e-01, 4.4468e-01, 2.8673e-01, 3.6586e-01,\r\n 9.7642e-01, -8.4343e-01, 9.6599e-01, 9.7235e-01, 7.2667e-01,\r\n -7.5785e-01, 9.2892e-02, -9.9089e-01, -1.7004e-01, -6.8200e-01,\r\n -9.9283e-01, 2.6244e-01, -7.9871e-01, 2.3397e-02, 4.6413e-02,\r\n -9.3371e-01, 2.7699e-01, 9.9995e-01, -3.2671e-01, 2.1108e-01,\r\n -2.0636e-01, -1.0000e+00, 1.9622e-01, -9.3330e-01, 6.8736e-01,\r\n 6.4731e-01, 5.3773e-01, 9.2759e-02, 4.1069e-01, 4.0360e-01,\r\n 1.9002e-01, -1.7049e-01, 7.5259e-03, -2.0453e-01, -5.7574e-01,\r\n -5.3062e-01, 3.9367e-01, -7.0627e-01, -9.2865e-01, 6.8820e-01,\r\n 3.2698e-01, -3.3506e-02, -1.2323e-01, -1.5304e-01, -1.8077e-01,\r\n 9.3398e-01, 2.6375e-01, 3.7505e-01, -8.9548e-01, 1.1777e-01,\r\n 2.2054e-01, -6.3351e-01, 1.0000e+00, -6.9228e-01, -9.8653e-01,\r\n 6.9799e-01, 4.0303e-01, 5.2453e-01, 2.3217e-01, -1.2151e-01,\r\n -1.0000e+00, 5.6760e-01, 2.9295e-02, -9.9318e-01, 8.3171e-02,\r\n 5.2939e-01, -2.3176e-01, -1.5694e-01, 4.9278e-01, -4.2614e-01,\r\n -3.8079e-01, -2.6060e-01, -6.9055e-01, -1.7180e-01, -1.9810e-01,\r\n -2.7986e-02, -7.2085e-02, -3.7635e-01, -3.7743e-01, 1.3508e-01,\r\n -4.3892e-01, -6.1321e-01, 1.7726e-01, -3.5434e-01, 6.4734e-01,\r\n 4.0373e-01, -2.8194e-01, 4.5104e-01, -9.7876e-01, 6.1044e-01,\r\n -2.3526e-01, -9.9035e-01, -5.1350e-01, -9.9280e-01, 6.8329e-01,\r\n -2.1623e-01, -1.4641e-01, 9.8273e-01, 3.7345e-01, 4.8171e-01,\r\n -5.6467e-03, -7.3005e-01, -1.0000e+00, -7.2252e-01, -5.1978e-01,\r\n 7.0765e-02, -1.5036e-01, -9.8355e-01, -9.7384e-01, 5.8453e-01,\r\n 9.6710e-01, 1.4193e-01, 9.9981e-01, -2.1194e-01, 9.6675e-01,\r\n 2.3627e-02, -4.1555e-01, 1.9872e-01, -4.0593e-01, 6.5180e-01,\r\n 6.1598e-01, -6.8750e-01, 7.9808e-02, -2.0437e-01, 3.4504e-01,\r\n -6.7176e-01, -1.3692e-01, -2.7750e-01, -9.6740e-01, -3.6698e-01,\r\n 9.6934e-01, -2.5050e-01, -6.9297e-01, 4.8327e-01, -1.4613e-01,\r\n -5.1224e-01, 8.8387e-01, 6.9173e-01, 3.8395e-01, -1.7536e-01,\r\n 3.8873e-01, -4.3011e-03, 6.1876e-01, -8.9292e-01, 3.4243e-02,\r\n 4.5193e-01, -2.4782e-01, -4.7402e-01, -9.8375e-01, -3.1763e-01,\r\n 5.9109e-01, 9.9284e-01, 7.9634e-01, 2.4601e-01, 6.1729e-01,\r\n -1.8376e-01, 6.8750e-01, -9.7083e-01, 9.8624e-01, -2.0573e-01,\r\n 2.0418e-01, 4.1400e-01, 1.9102e-01, -9.1718e-01, -3.5273e-01,\r\n 8.9628e-01, -5.6812e-01, -8.9552e-01, -3.5567e-02, -4.9052e-01,\r\n -4.3559e-01, -6.2323e-01, 5.6863e-01, -2.6201e-01, -3.1324e-01,\r\n -1.2852e-02, 9.4585e-01, 9.8664e-01, 8.3363e-01, -2.4392e-01,\r\n 7.3786e-01, -9.4466e-01, -5.2720e-01, -1.6349e-02, 2.4207e-01,\r\n 3.6905e-02, 9.9638e-01, -5.8095e-01, -7.2046e-02, -9.4418e-01,\r\n -9.8921e-01, -1.0289e-01, -9.3301e-01, -5.3531e-02, -6.8719e-01,\r\n 5.3295e-01, 1.6390e-01, 2.3460e-01, 4.3260e-01, -9.9501e-01,\r\n -7.7318e-01, 2.6342e-01, -3.6949e-01, 4.0245e-01, -1.6657e-01,\r\n 4.5766e-01, 7.4537e-01, -5.8549e-01, 8.4632e-01, 9.3526e-01,\r\n -6.4963e-01, -7.8264e-01, 8.5868e-01, -2.9683e-01, 9.0246e-01,\r\n -6.5124e-01, 9.8896e-01, 8.6732e-01, 8.7014e-01, -9.5627e-01,\r\n -4.1195e-01, -9.1043e-01, -4.5438e-01, 5.7729e-02, -3.6862e-01,\r\n 5.7032e-01, 5.5757e-01, 3.0482e-01, 7.0850e-01, -6.6279e-01,\r\n 9.9909e-01, -5.0139e-01, -9.7001e-01, -2.2370e-01, -1.8440e-02,\r\n -9.9107e-01, 7.2208e-01, 2.4379e-01, 6.9083e-02, -3.2313e-01,\r\n -7.3217e-01, -9.7295e-01, 9.2268e-01, 5.0675e-02, 9.9215e-01,\r\n -8.0247e-02, -9.5682e-01, -4.1637e-01, -9.4549e-01, -2.9790e-01,\r\n -1.5625e-01, 2.4707e-01, -1.8468e-01, -9.7276e-01, 4.7428e-01,\r\n 5.6760e-01, 5.5919e-01, -1.9418e-01, 9.9932e-01, 1.0000e+00,\r\n 9.7844e-01, 9.3669e-01, 9.5284e-01, -9.9929e-01, -4.9083e-01,\r\n 9.9999e-01, -9.7835e-01, -1.0000e+00, -9.5292e-01, -6.5736e-01,\r\n 4.1425e-01, -1.0000e+00, -2.7896e-03, 7.0756e-02, -9.4186e-01,\r\n 2.7960e-01, 9.8389e-01, 9.9658e-01, -1.0000e+00, 8.8289e-01,\r\n 9.6828e-01, -6.2958e-01, 9.3367e-01, -3.7519e-01, 9.8027e-01,\r\n 4.2505e-01, 3.0766e-01, -3.2042e-01, 2.7469e-01, -7.8253e-01,\r\n -8.8309e-01, -1.5604e-01, -3.6222e-01, 9.9091e-01, 3.0116e-02,\r\n -7.8697e-01, -9.4496e-01, 2.2050e-01, -8.4521e-02, -4.8378e-01,\r\n -9.7952e-01, -1.3446e-01, 4.2209e-01, 7.8760e-01, 6.4992e-02,\r\n 2.0492e-01, -7.8143e-01, 2.2120e-01, -5.0228e-01, 3.7149e-01,\r\n 6.5244e-01, -9.4897e-01, -6.0978e-01, -4.8976e-03, -4.6856e-01,\r\n -2.8122e-01, -9.6984e-01, 9.8036e-01, -3.5220e-01, 7.4903e-01,\r\n 1.0000e+00, -1.0373e-01, -9.4037e-01, 6.2856e-01, 1.5745e-01,\r\n -1.1596e-01, 1.0000e+00, 7.2891e-01, -9.8543e-01, -5.3814e-01,\r\n 4.0543e-01, -4.9501e-01, -4.8527e-01, 9.9950e-01, -1.4058e-01,\r\n -2.3799e-01, -1.5841e-01, 9.8467e-01, -9.9180e-01, 9.7240e-01,\r\n -9.5292e-01, -9.8022e-01, 9.8012e-01, 9.5211e-01, -6.6387e-01,\r\n -7.2622e-01, 1.1509e-01, -3.1365e-01, 1.8487e-01, -9.7602e-01,\r\n 7.9482e-01, 5.2428e-01, -1.3540e-01, 9.1377e-01, -9.0275e-01,\r\n -5.2769e-01, 2.8301e-01, -4.9215e-01, 3.4866e-02, 8.3573e-01,\r\n 5.0270e-01, -2.4031e-01, -6.1194e-02, -2.5558e-01, -1.3530e-01,\r\n -9.8688e-01, 2.9877e-01, 1.0000e+00, -5.3199e-02, 4.4522e-01,\r\n -2.4564e-01, 3.8897e-02, -3.7170e-01, 3.6843e-01, 5.1087e-01,\r\n -1.9742e-01, -8.8481e-01, 4.5420e-01, -9.8222e-01, -9.8894e-01,\r\n 8.5417e-01, 1.4674e-01, -3.3154e-01, 9.9999e-01, 4.4333e-01,\r\n 7.1728e-02, 1.6790e-01, 9.6064e-01, 2.3267e-02, 7.2436e-01,\r\n 4.9905e-01, 9.8528e-01, -2.0286e-01, 5.2711e-01, 9.0711e-01,\r\n -5.6147e-01, -3.4452e-01, -6.1113e-01, -8.1268e-02, -9.2887e-01,\r\n 1.0119e-01, -9.7066e-01, 9.7404e-01, 8.2025e-01, 2.9760e-01,\r\n 1.9059e-01, 3.6089e-01, 1.0000e+00, -2.7256e-01, 6.5052e-01,\r\n -6.0092e-01, 9.0897e-01, -9.9819e-01, -9.1409e-01, -3.7810e-01,\r\n 1.2677e-02, -3.9492e-01, -3.0028e-01, 3.4323e-01, -9.7925e-01,\r\n 4.4501e-01, 3.7582e-01, -9.9622e-01, -9.9495e-01, 1.6366e-01,\r\n 9.2522e-01, -1.3063e-02, -9.5314e-01, -7.5003e-01, -6.5409e-01,\r\n 4.1526e-01, -7.6235e-02, -9.6046e-01, 3.2395e-01, -2.7184e-01,\r\n 4.7535e-01, -1.1767e-01, 5.6867e-01, 4.6844e-01, 8.3125e-01,\r\n -2.1505e-01, -2.6495e-01, -4.4479e-02, -8.5166e-01, 8.8927e-01,\r\n -8.9329e-01, -7.7919e-01, -1.5320e-01, 1.0000e+00, -4.3274e-01,\r\n 6.4268e-01, 7.7000e-01, 7.9197e-01, -5.4889e-02, 8.0927e-02,\r\n 7.9722e-01, 2.1034e-01, -1.9189e-01, -4.4749e-01, -8.0585e-01,\r\n -3.5409e-01, 7.0995e-01, 1.2411e-01, 2.0604e-01, 8.3328e-01,\r\n 7.4750e-01, 1.7900e-04, 7.6917e-02, -9.1725e-02, 9.9981e-01,\r\n -2.6801e-01, -8.3787e-02, -4.8642e-01, 1.1836e-01, -3.5603e-01,\r\n -5.8620e-01, 1.0000e+00, 2.2691e-01, 3.2801e-01, -9.9343e-01,\r\n -7.3298e-01, -9.5126e-01, 1.0000e+00, 8.4895e-01, -8.6216e-01,\r\n 6.9319e-01, 5.5441e-01, -1.1380e-02, 8.6958e-01, -1.2449e-01,\r\n -2.8602e-01, 1.8517e-01, 9.2221e-02, 9.6773e-01, -4.5911e-01,\r\n -9.7611e-01, -6.6894e-01, 3.7154e-01, -9.7862e-01, 9.9949e-01,\r\n -5.5391e-01, -2.0926e-01, -3.9404e-01, -2.3863e-02, 6.3624e-01,\r\n -1.0563e-01, -9.8927e-01, -1.4047e-01, 1.2247e-01, 9.7469e-01,\r\n 2.6847e-01, -6.0451e-01, -9.5354e-01, 4.5191e-01, 6.6822e-01,\r\n -7.2218e-01, -9.6438e-01, 9.7538e-01, -9.9165e-01, 5.6641e-01,\r\n 1.0000e+00, 2.2837e-01, -2.8539e-01, 1.6956e-01, -4.6714e-01,\r\n 2.5561e-01, -2.6744e-01, 7.4301e-01, -9.7890e-01, -2.7469e-01,\r\n -1.4162e-01, 2.7886e-01, -7.0853e-02, -5.8891e-02, 8.2879e-01,\r\n 1.9968e-01, -5.4085e-01, -6.8158e-01, 3.7584e-02, 3.5805e-01,\r\n 8.9092e-01, -1.7879e-01, -8.1491e-02, 5.0655e-02, -7.9140e-02,\r\n -9.5114e-01, -1.4923e-01, -3.5370e-01, -9.9994e-01, 7.4321e-01,\r\n -1.0000e+00, 2.1850e-01, -2.5182e-01, -2.2171e-01, 8.7817e-01,\r\n 2.9648e-01, 3.4926e-01, -8.2534e-01, -3.8831e-01, 7.6622e-01,\r\n 8.0938e-01, -2.1051e-01, -3.0882e-01, -7.6183e-01, 2.2523e-01,\r\n -1.4952e-02, 1.5150e-01, -2.1056e-01, 7.3482e-01, -1.5207e-01,\r\n 1.0000e+00, 1.0631e-01, -7.7462e-01, -9.8438e-01, 1.6242e-01,\r\n -1.6337e-01, 1.0000e+00, -9.4196e-01, -9.7149e-01, 3.9827e-01,\r\n -7.2371e-01, -8.6582e-01, 3.0937e-01, -6.4325e-02, -8.1062e-01,\r\n -8.8436e-01, 9.8219e-01, 9.3543e-01, -5.6058e-01, 4.5004e-01,\r\n -3.2933e-01, -5.5851e-01, -6.9835e-02, 6.0196e-01, 9.9111e-01,\r\n 4.1170e-01, 9.1721e-01, 5.9978e-01, -9.4103e-02, 9.7966e-01,\r\n 1.5322e-01, 5.3662e-01, 4.2338e-02, 1.0000e+00, 2.8920e-01,\r\n -9.3933e-01, 2.4383e-01, -9.8948e-01, -1.5036e-01, -9.7242e-01,\r\n 2.8053e-01, 1.1691e-01, 9.0178e-01, -2.1055e-01, 9.7547e-01,\r\n -5.0734e-01, -8.5119e-03, -5.2189e-01, 1.1963e-01, 4.0313e-01,\r\n -9.4529e-01, -9.8752e-01, -9.8975e-01, 4.5711e-01, -4.0753e-01,\r\n 5.8175e-02, 1.1543e-01, 8.6051e-02, 3.6199e-01, 4.3131e-01,\r\n -1.0000e+00, 9.5818e-01, 4.0499e-01, 6.9443e-01, 9.7521e-01,\r\n 6.7153e-01, 4.3386e-01, 2.2481e-01, -9.9118e-01, -9.9126e-01,\r\n -3.1248e-01, -1.4604e-01, 7.9951e-01, 6.1145e-01, 9.2726e-01,\r\n 4.0171e-01, -3.9375e-01, -2.0938e-01, -3.2651e-02, -4.1723e-01,\r\n -9.9582e-01, 4.5682e-01, -9.4401e-02, -9.8150e-01, 9.6766e-01,\r\n -5.5518e-01, -8.0481e-02, 4.4743e-01, -6.0429e-01, 9.7261e-01,\r\n 8.6633e-01, 3.7309e-01, 9.4917e-04, 4.6426e-01, 9.1590e-01,\r\n 9.6965e-01, 9.8799e-01, -4.6592e-01, 8.7146e-01, -3.1116e-01,\r\n 5.1496e-01, 6.7961e-01, -9.5609e-01, 1.3302e-03, 3.6581e-01,\r\n -2.3789e-01, 2.6341e-01, -1.2874e-01, -9.8464e-01, 4.8621e-01,\r\n -1.8921e-01, 6.1015e-01, -4.3986e-01, 2.1561e-01, -3.7115e-01,\r\n -1.5832e-02, -6.9704e-01, -7.3403e-01, 5.7310e-01, 5.0895e-01,\r\n 9.4111e-01, 6.9365e-01, 6.9171e-02, -7.3277e-01, -1.1294e-01,\r\n -4.0168e-01, -9.2587e-01, 9.6638e-01, 2.2207e-02, 1.5029e-01,\r\n 2.8954e-01, -8.5994e-02, 7.4631e-01, -1.5933e-01, -3.5710e-01,\r\n -1.6201e-01, -7.1149e-01, 9.0602e-01, -4.2873e-01, -4.6653e-01,\r\n -5.4765e-01, 7.4640e-01, 2.3966e-01, 9.9982e-01, -4.6795e-01,\r\n -6.4802e-01, -4.1201e-01, -3.4984e-01, 3.5475e-01, -5.4668e-01,\r\n -1.0000e+00, 3.6903e-01, -1.7324e-01, 4.3267e-01, -4.7206e-01,\r\n 6.3586e-01, -5.2151e-01, -9.9077e-01, -1.6597e-01, 2.6735e-01,\r\n 4.5069e-01, -4.3034e-01, -5.6321e-01, 5.7792e-01, 8.8123e-02,\r\n 9.4964e-01, 9.2798e-01, -3.3326e-01, 5.1963e-01, 6.0865e-01,\r\n -4.4019e-01, -6.8129e-01, 9.3489e-01]], grad_fn=<TanhBackward>)\r\n>>> output[2].shape\r\ntorch.Size([1, 768])\r\n>>>\r\n```\r\n\r\n> @TheEdoardo93 it doesn't matter whether it is GPT2 or bert. Both has the same error :\r\n> I'm trying to play with GPT2 that's why I pasted my own code.\r\n> \r\n> Using BERT as an Encoder and Decoder\r\n> \r\n> ```\r\n> >>> model = PreTrainedEncoderDecoder.from_pretrained('bert-base-uncased','bert-base-uncased')\r\n> >>> ouput = model(torch.tensor( encoder_input_ids).unsqueeze(0))\r\n> Traceback (most recent call last):\r\n> File \"<stdin>\", line 1, in <module>\r\n> File \"/home/guest_1/.local/lib/python3.6/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> TypeError: forward() missing 1 required positional argument: 'decoder_input_ids'\r\n> ```", "I got the same issue", "Do you try to follow my suggestion reported above? In my environment, it **works as expected**.\r\nMy environment details:\r\n- **OS**: Ubuntu 16.04\r\n- **Python**: 3.6.9\r\n- **Transformers**: 2.2.2 (installed with `pip install transformers`)\r\n- **PyTorch**: 1.3.1\r\n- **TensorFlow**: 2.0\r\n\r\nIf not, can you post your **environment** and a list of **steps** to reproduce the bug?\r\n\r\n> I got the same issue", "> Said so, I've tested your code and I've revealed how to working with PreTrainedEncoderDecoder correctly without bugs. You can see my code below.\r\n\r\n@TheEdoardo93 that doesn't make sense you giving your encoders input as a decoder's input. \r\n\r\n", "> > Said so, I've tested your code and I've revealed how to working with PreTrainedEncoderDecoder correctly without bugs. You can see my code below.\r\n> \r\n> @TheEdoardo93 that doesn't make sense your giving your encoders input as a decoder's input.\r\n> Never-mind, I know what's the issue is so closing it.\r\n\r\nSorry, it was my mistake. Can you share with us what was the problem and how to solve it?", "@anandhperumal Could you please share how you solved the issue? did you path a ```<BOS>``` to decoder input?\r\nAppreciate that", "@anandhperumal can you let us know how you fixed the issue?", "You can pass a start token to the decoder, just like a Seq2Seq Arch.\r\n" ]

[]

[]

[ "Is it related to #2204 ?\r\n**Segmentation fault** (usually) means that you tried to access memory that you do not have access to.", "> Is it related to #2204 ?\r\n\r\nYes, it looks like the same bug as this. I know how to solve it but do not know why.😂\r\n\r\n\r\n", "Solved in #2207 ", "@jstzwj I'm still getting this issue on \"transformers==2.2.2\" on PyPi, how to solve it?", "> @jstzwj I'm still getting this issue on \"transformers==2.2.2\" on PyPi, how to solve it?\r\n\r\n#2207 " ]

[]

[]

[ "#2207 ", "Solved by @LysandreJik yesterday (this was due to upstream dependency bug)" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2203?src=pr&el=h1) Report\n> Merging [#2203](https://codecov.io/gh/huggingface/transformers/pull/2203?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/94c99db34cf9074a212c36554fb925c513d70ab1?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2203?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2203 +/- ##\n=======================================\n Coverage 81.47% 81.47% \n=======================================\n Files 122 122 \n Lines 18342 18342 \n=======================================\n Hits 14945 14945 \n Misses 3397 3397\n```\n\n\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2203?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2203?src=pr&el=footer). Last update [94c99db...a297846](https://codecov.io/gh/huggingface/transformers/pull/2203?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "This is a good idea :) @thomwolf " ]

[]

[]

[ "Thanks for the bug report. Fixed on `master` in 9a399ead253e27792cbf0ef386cc39f9b7084f8f by reverting the output of #1778." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2201?src=pr&el=h1) Report\n> Merging [#2201](https://codecov.io/gh/huggingface/transformers/pull/2201?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/f061606277322a013ec2d96509d3077e865ae875?src=pr&el=desc) will **decrease** coverage by `0.07%`.\n> The diff coverage is `0%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2201?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2201 +/- ##\n==========================================\n- Coverage 80.32% 80.25% -0.08% \n==========================================\n Files 122 122 \n Lines 18342 18358 +16 \n==========================================\n Hits 14734 14734 \n- Misses 3608 3624 +16\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2201?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_encoder\\_decoder.py](https://codecov.io/gh/huggingface/transformers/pull/2201/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2VuY29kZXJfZGVjb2Rlci5weQ==) | `25.92% <0%> (-6.39%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2201?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2201?src=pr&el=footer). Last update [f061606...0f844f5](https://codecov.io/gh/huggingface/transformers/pull/2201?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "@julien-c @LysandreJik @thomwolf you can review my code for solving issue #2196 :)", "Cool, thanks @TheEdoardo93 ! Do you mind if I clean the commits as to not have a merge commit (rebasing on master instead)? I'll push directly on your fork if that's okay. You'll still be the author of the commits.", "@LysandreJik\r\nThe steps I’ve done for this PR are the following:\r\n- make my changes on the source code\r\n- git add/commit/push\r\n- after that, the automatic tests start to run but an error of “segmentation fault” occurs\r\n- i’ve changed a line only for “redo” the tests suite after the bug was solved, but my local branch was below “master”. I didn’t see this fact so I didn’t do a “git pull”. After that, I’ve done git add/commit/push and run the tests suite and now the tests suite is working without error\r\n\r\nSaid so, my changes only occurs in my first commit. The other ones are due to Transformers’ master changes. Have I answered to you? I hope yes.\r\n\r\nThanks for being the author of this PR yet!", "I rebased your code on the current master branch so that there's only the two commits. Thanks @TheEdoardo93.", "I feel like this is very verbose so let's maybe revisit when we drop Python 2 support." ]

[]

[]

[ "A few questions (the ones asked in the issue templates...):\r\n- which version/branch of `transformers` are you using?\r\n- which exact command line are you running?", "Hi,\r\nsorry, my bad.\r\n\r\nI am calling run_ner.py with\r\n```\r\n python .\\run_ner.py --data_dir ./ --model_type bert --labels .\\labels.txt --model_name_or_path bert_base_uncased --output_dir pytorch_ner --max_seq_length 75 --n\r\num_train_epochs 5 --per_gpu_train_batch_size 12 --save_steps 750 --seed 4711 --do_train --do_eval --do_predict --cache_dir s3_cache\r\n```\r\nI am using transformers 2.2.2", "The JSON file the Python script is trying to use is the configuration of BERT model (`config.json`)?\r\n\r\n> I am trying to run the run_ner.py example described here: (https://huggingface.co/transformers/examples.html#named-entity-recognition)\r\n> \r\n> When running the example I get the following exception:\r\n> \r\n> ```\r\n> I1217 09:36:10.744300 14416 file_utils.py:40] PyTorch version 1.3.1 available.\r\n> W1217 09:36:11.329299 14416 run_ner.py:422] Process rank: -1, device: cuda, n_gpu: 1, distributed training: False, 16-bits training: False\r\n> I1217 09:36:11.863193 14416 configuration_utils.py:160] loading configuration file https://s3.amazonaws.com/models.huggingface.co/bert/bert_base_uncased/config.json from cache at s3_cache\\67674340071d93960fbc3eb74cf1d0b51de232689bfc75d63e3f4ab1c9a052f9\r\n> Traceback (most recent call last):\r\n> File \".\\run_ner.py\", line 531, in <module>\r\n> main()\r\n> File \".\\run_ner.py\", line 441, in main\r\n> cache_dir=args.cache_dir if args.cache_dir else None)\r\n> File \"C:\\Program Files\\Python\\Python37\\lib\\site-packages\\transformers\\configuration_utils.py\", line 163, in from_pretrained\r\n> config = cls.from_json_file(resolved_config_file)\r\n> File \"C:\\Program Files\\Python\\Python37\\lib\\site-packages\\transformers\\configuration_utils.py\", line 196, in from_json_file\r\n> return cls.from_dict(json.loads(text))\r\n> File \"C:\\Program Files\\Python\\Python37\\lib\\json\\__init__.py\", line 348, in loads\r\n> return _default_decoder.decode(s)\r\n> File \"C:\\Program Files\\Python\\Python37\\lib\\json\\decoder.py\", line 337, in decode\r\n> obj, end = self.raw_decode(s, idx=_w(s, 0).end())\r\n> File \"C:\\Program Files\\Python\\Python37\\lib\\json\\decoder.py\", line 355, in raw_decode\r\n> raise JSONDecodeError(\"Expecting value\", s, err.value) from None\r\n> json.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0)\r\n> ```\r\n> \r\n> A JSON files does not seemed to be found. Can someone tell me which file this is?", "Do I have to retrieve the file from somewhere ? I assumed everything is retrieved automatically by the script?", "Reminds me of this isse: https://github.com/huggingface/transformers/issues/2154", "Well, the `bert_base_uncased` for the `model_name_or_path` option is not correct. Just use: `bert-base-uncased` :)", "Yes, the `bert_base_uncased` should be `bert-base-uncased`.\r\n\r\nI've improved these error messages on the master with #2164 by the way, it should be more explicit now that the configuration file can't be found.", "@Horsmann Additionally, if you want to use an uncased model, make sure that you pass `--do_lower_case` to the `run_ner.py` script!", "Thanks! It is working now. The exception looked like I am missing a file so I didn't checked for the model name." ]

[]

[]

[ "What do you mean \"for training\"? Training what? I imagine that you wish to do this for a downstream task. In that case you can just use BERT for feature extraction and use its features in another system. I suppose one could also fine-tune the model during training where you somewhere inject other features into the model, but you'll have to think carefully where and how you optimize the model (one optimizer for finetuning and training newly injected parameters, multiple optimizers, and so on).\r\n\r\nI'd say you could start with feature extraction and adding other features. ", "As correctly said by @BramVanroy , you can see some examples of this approach [here](https://jalammar.github.io/a-visual-guide-to-using-bert-for-the-first-time/).\r\n\r\n> ## Questions & Help\r\n> I don't know whether models like Bert can capture some traditional features or not, such as the number of nouns in a sentence, the number of words in a sentence, etc. But I want to add these features to Bert or xlnet for training. Can anyone help me?", "Hi @TheEdoardo93 , I read the approach you told me. But I still have some confusing, since the model could output a vector of 768 for every sentence. Where I concat my feature like \"the number of nouns in a sentence\"? Just the end line of the vector and normalized all of them to 1? Thank you.", "`BERT` gives you a feature vector of size 768 for each sentence. Besides this, you can add _N_ features to this vector in order to have at the end a vector of size 768 + _N_, in which the N features have been chosen by you, e.g. number of nouns in a sentence.\r\nSo, from 0 to 767 you have the features extracted with `BERT` model, and from 768 to 768+_N_-1 you have your **custom** features extracted by you.\r\n\r\nN.B: remember that it's very important in Machine Learning to have the range for all features the same (or quite the same), in order to not give more importance to some rather than others. For this task, you can use e.g. StandardScaler or MinMaxScaler from [Scikit-learn](https://scikit-learn.org/stable/) or a custom scaler implemented by you. \r\n\r\n\r\n> Hi @TheEdoardo93 , I read the approach you told me. But I still have some confusing, since the model could output a vector of 768 for every sentence. Where I concat my feature like \"the number of nouns in a sentence\"? Just the end line of the vector and normalized all of them to 1? Thank you.", "@TheEdoardo93 Thank you very much !" ]

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

[]

[ "Hi, if you're working on a clone/fork of this library, then you can freely change the following lines in `transformers/data/processors/glue.py` (lines 254-262 on the current master):\r\n\r\n```py\r\n def get_train_examples(self, data_dir):\r\n \"\"\"See base class.\"\"\"\r\n return self._create_examples(\r\n self._read_tsv(os.path.join(data_dir, \"train.tsv\")), \"train\")\r\n\r\n def get_dev_examples(self, data_dir):\r\n \"\"\"See base class.\"\"\"\r\n return self._create_examples(\r\n self._read_tsv(os.path.join(data_dir, \"dev.tsv\")), \"dev\")\r\n```\r\n\r\nto read the `test.tsv` instead of the `dev.tsv`.", "Hi @LysandreJik,\r\n\r\nThanks for your reply!\r\nHowever, the format of test set is different from the dev set. It has an extra `id` field and does not have ground truth labels. Thus, the simple change you suggested wouldn't work in this case. Besides, GLUE leaderboard also requires a specific output format for submission.\r\n\r\nI have noticed that in Huggingface's workshop paper [DistilBERT](https://arxiv.org/abs/1910.01108), you also use DEV set instead of TEST result from GLUE leaderboard. I guess it may be somehow related to the lack of a submission script. To facilitate further research, I hope you can make a script for GLUE submission. As far as I know, many researchers are using huggingface/transformers for their research. I really appreciate it if you can provide a script for the community.\r\n\r\nThanks!", "Hi @LysandreJik,\r\nI've just realized that you are one of the authors of DistilBERT so I suppose you know exactly what I mean. It can truly benefit the research community if you do this. Thanks.", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "hi did u get the script? ", "@yuchenlin My co-author wrote one himself. I added you on Wechat but you didn't respond. ", "Hey, I think this would be helpful for many in the community. Can you maybe share the script you wrote for this via a pull request?", "@Breakend Sure, I will verify the script and open a pull request.", "Hello @JetRunner , no news about this script?\r\nThank you in advance", "> Hello @JetRunner , no news about this script?\r\n> Thank you in advance\r\n\r\nSorry, I totally forgot about this stuff. I'll notify you as soon as we upload it.\r\n\r\ncc @michaelzhouwang", "> Hello @JetRunner , no news about this script?\r\n> Thank you in advance\r\n\r\nHi I have uploaded the script for prediction on GLUE benchmarks at:\r\nhttps://github.com/JetRunner/BERT-of-Theseus/tree/master/glue_script\r\nYou can first replace the glue.py in src/transformers/data/processor/ and then use run_prediction.py.", "> > Hello @JetRunner , no news about this script?\r\n> > Thank you in advance\r\n> \r\n> Hi I have uploaded the script for prediction on GLUE benchmarks at:\r\n> https://github.com/JetRunner/BERT-of-Theseus/tree/master/glue_script\r\n> You can first replace the glue.py in src/transformers/data/processor/ and then use run_prediction.py.\r\n\r\ncc @Breakend @jibay ", "@JetRunner @MichaelZhouwang Thank you for the quick answer, i will check asap :)" ]

[ { "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 because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "I resolved this issue you've raised correctly. I'll make a PR today to solve this bug! :)\r\nUPDATE: I've made the PR: you can look [here](https://github.com/huggingface/transformers/pull/2201).\r\n\r\n> ## Bug\r\n> Model I am using (Bert, XLNet....):\r\n> \r\n> Language I am using the model on (English, Chinese....):\r\n> \r\n> The problem arise when using:\r\n> \r\n> * [ ] the official example scripts: (give details)\r\n> * [x] my own modified scripts: (give details)\r\n> \r\n> The tasks I am working on is:\r\n> \r\n> * [ ] an official GLUE/SQUaD task: (give the name)\r\n> * [x] my own task or dataset: (give details)\r\n> \r\n> ## To Reproduce\r\n> Steps to reproduce the behavior:\r\n> \r\n> 1. Define a model and try to save the model then the error occurs. This is because the encoder and decoder model is save in different directories and the directories are not created.\r\n> In order to handle this, you need to check if the given path exists then the encoder folder exists if not create a folder above this [line](https://github.com/huggingface/transformers/blob/3f5ccb183e3cfa755dea2dd2afd9abbf1a0f93b8/transformers/modeling_encoder_decoder.py#L169)\r\n> \r\n> Code to reproduce:\r\n> \r\n> ```\r\n> from transformers import PreTrainedEncoderDecoder\r\n> model = PreTrainedEncoderDecoder.from_pretrained('bert-base-uncased', 'bert-base-uncased', cache_dir='../transformers/cache')\r\n> model.save_pretrained(final_model_output)\r\n> ```\r\n> \r\n> Error:\r\n> \r\n> ```\r\n> Traceback (most recent call last):\r\n> File \"<stdin>\", line 1, in <module>\r\n> File \"/home/guest_1/anaconda3/envs/my_env/lib/python3.6/site-packages/transformers/modeling_encoder_decoder.py\", line 167, in save_pretrained\r\n> self.encoder.save_pretrained(os.path.join(save_directory, \"encoder\"))\r\n> File \"/home/guest_1/anaconda3/envs/my_env/lib/python3.6/site-packages/transformers/modeling_utils.py\", line 239, in save_pretrained\r\n> assert os.path.isdir(save_directory), \"Saving path should be a directory where the model and configuration can be saved\"\r\n> AssertionError: Saving path should be a directory where the model and configuration can be saved\r\n> ```\r\n> \r\n> ## Expected behavior\r\n> ## Environment\r\n> ```\r\n> OS: ubuntu\r\n> Python version: 3.6\r\n> PyTorch version:1.3.0\r\n> PyTorch Transformers version (or branch):2.2.0\r\n> Using GPU ? Yes\r\n> Distributed of parallel setup ? No\r\n> Any other relevant information:\r\n> ```\r\n> \r\n> ## Additional context", "@TheEdoardo93 did you try doing inference ?" ]

[]

[]

[]

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

[]

[ "@thomwolf or @julien-c any idea what's going on with the seg fault in the `build_py3_torch_and_tf` [run](https://circleci.com/gh/huggingface/transformers/9811?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link)? was able to run and pass locally...\r\n\r\n ", "> @thomwolf or @julien-c any idea what's going on with the seg fault in the `build_py3_torch_and_tf` [run](https://circleci.com/gh/huggingface/transformers/9811?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link)? was able to run and pass locally...\r\n> \r\n\r\nI've encountered this problem too while adding my PR some days ago. It was broken something inside Transformers library, but @LysandreJik solved this bug two days ago in #2207. Therefore, you can re-launch your tests and you'll see that now it works as expected! :)\r\n\r\nUPDATE: you've to install transformers from source code `master` branch through `pip install git+https://github.com/huggingface/transformers.git`", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "Hi, you can add new vocabulary using the tokenizer's `add_tokens` method.", "Thanks for the quick response!" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2192?src=pr&el=h1) Report\n> Merging [#2192](https://codecov.io/gh/huggingface/transformers/pull/2192?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/d8034092153a6850052862f154a398b88b8ba4e5?src=pr&el=desc) will **increase** coverage by `1.12%`.\n> The diff coverage is `33.33%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2192?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2192 +/- ##\n==========================================\n+ Coverage 80.21% 81.33% +1.12% \n==========================================\n Files 120 120 \n Lines 18254 18261 +7 \n==========================================\n+ Hits 14642 14853 +211 \n+ Misses 3612 3408 -204\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2192?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_tf\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2192/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3V0aWxzLnB5) | `92.3% <0%> (ø)` | :arrow_up: |\n| [transformers/modeling\\_tf\\_pytorch\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2192/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3B5dG9yY2hfdXRpbHMucHk=) | `92.15% <37.5%> (+82.56%)` | :arrow_up: |\n| [transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/2192/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX29wZW5haS5weQ==) | `81.45% <0%> (+1.32%)` | :arrow_up: |\n| [transformers/modeling\\_ctrl.py](https://codecov.io/gh/huggingface/transformers/pull/2192/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2N0cmwucHk=) | `96.47% <0%> (+2.2%)` | :arrow_up: |\n| [transformers/modeling\\_xlnet.py](https://codecov.io/gh/huggingface/transformers/pull/2192/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3hsbmV0LnB5) | `74.54% <0%> (+2.32%)` | :arrow_up: |\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/2192/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `71.76% <0%> (+12.35%)` | :arrow_up: |\n| [transformers/tests/modeling\\_tf\\_common\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2192/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2NvbW1vbl90ZXN0LnB5) | `97.41% <0%> (+17.24%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2192?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2192?src=pr&el=footer). Last update [d803409...2dd30a3](https://codecov.io/gh/huggingface/transformers/pull/2192?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Great, this works!" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2191?src=pr&el=h1) Report\n> Merging [#2191](https://codecov.io/gh/huggingface/transformers/pull/2191?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/ceae85ad60da38cacb14eca49f752669a4fe31dc?src=pr&el=desc) will **decrease** coverage by `<.01%`.\n> The diff coverage is `100%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2191?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2191 +/- ##\n==========================================\n- Coverage 79.92% 79.92% -0.01% \n==========================================\n Files 131 131 \n Lines 19469 19470 +1 \n==========================================\n Hits 15561 15561 \n- Misses 3908 3909 +1\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2191?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2191/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl91dGlscy5weQ==) | `92.1% <100%> (+0.01%)` | :arrow_up: |\n| [transformers/tests/utils.py](https://codecov.io/gh/huggingface/transformers/pull/2191/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL3V0aWxzLnB5) | `90.62% <0%> (-3.13%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2191?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2191?src=pr&el=footer). Last update [ceae85a...cb6d54b](https://codecov.io/gh/huggingface/transformers/pull/2191?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Awesome!" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2190?src=pr&el=h1) Report\n> Merging [#2190](https://codecov.io/gh/huggingface/transformers/pull/2190?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/e92bcb7eb6c5b9b6ed313cc74abaab50b3dc674f?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2190?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2190 +/- ##\n=======================================\n Coverage 81.35% 81.35% \n=======================================\n Files 120 120 \n Lines 18254 18254 \n=======================================\n Hits 14851 14851 \n Misses 3403 3403\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2190?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tokenization\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/2190/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl9iZXJ0LnB5) | `96.38% <ø> (ø)` | :arrow_up: |\n| [transformers/configuration\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/2190/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2NvbmZpZ3VyYXRpb25fYmVydC5weQ==) | `87.09% <ø> (ø)` | :arrow_up: |\n| [transformers/modeling\\_tf\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/2190/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX2JlcnQucHk=) | `96.31% <ø> (ø)` | :arrow_up: |\n| [transformers/modeling\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/2190/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2JlcnQucHk=) | `87.75% <ø> (ø)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2190?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2190?src=pr&el=footer). Last update [e92bcb7...3c1aede](https://codecov.io/gh/huggingface/transformers/pull/2190?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "This looks really awesome, thanks for sharing!\r\n\r\n**Uskomatonta**!! (yes this is the only Finnish word I know)\r\n\r\nThe new recommended way of uploading the files is inside folders, that way you'll be able to \r\ndo `AutoModel.from_pretrained(\"TurkuNLP/bert-base-finnish-[un]cased-v1\")` out-of-the-box, without even having to modify the lib's code (though we may want to add \"bert-base-finnish-[un]cased-v1\" as a shortcut name anyways).\r\n\r\n(We just [updated the documentation](https://github.com/huggingface/transformers/commit/855ff0e91d8b3bd75a3b1c1316e2efd814373764#commitcomment-36452545) this morning so this is very new)\r\n\r\nDo you want to re-upload the files inside folders? Or I can do it on our side too.\r\n\r\nAlso the ArXiv link in the PR's post seems broken, is the paper not public yet?\r\n\r\nAnyways, thanks for sharing, this is awesome!", "> (We just updated the documentation this morning so this is very new)\r\n\r\nAh, just missed it then.\r\n\r\n> Do you want to re-upload the files inside folders? Or I can do it on our side too.\r\n\r\nWould be great if you could do it, thanks.\r\n\r\n> Also the ArXiv link in the PR's post seems broken, is the paper not public yet?\r\n\r\nThe paper is scheduled to be announced on arxiv at 8pm EST so I'll fix the link tomorrow (in 12 hours or so).", "Looks like a URL was incorrect in the original PR, fixed it.\r\n\r\nMerging now, thank you again!", "By the way, would you guys be interesting in beta-testing a new way of pre-training a tokenizer on a corpus, @haamis?\r\n\r\n@n1t0 is working on something that might be of interest to you.", "Also @haamis we're rolling out model and contributor pages: e.g. https://huggingface.co/TurkuNLP\r\n\r\nAnything you would like to see added to this page? How can it be most helpful? Thanks!", "> By the way, would you guys be interesting in beta-testing a new way of pre-training a tokenizer on a corpus, @haamis?\r\n> \r\n> @n1t0 is working on something that might be of interest to you.\r\n\r\nI saw there was a faster implementation of the tokenizer in the works, improved speed would be nice.\r\n\r\n> Also @haamis we're rolling out model and contributor pages: e.g. https://huggingface.co/TurkuNLP\r\n> \r\n> Anything you would like to see added to this page? How can it be most helpful? Thanks!\r\n\r\nThat looks good to me.\r\n\r\nSorry about the late reply." ]

[]

[]

[ "@stefan-it Following the merge of #1959, you should not have to duplicate the weights conversion script anymore. It should work out of the box, `fairseq.XLMRModel` being a subclass of `fairseq.RobertaModel`.", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2189?src=pr&el=h1) Report\n> Merging [#2189](https://codecov.io/gh/huggingface/transformers/pull/2189?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/2f1c745cded91b2f6cfed5b502ea5cbd7d6b9ac7?src=pr&el=desc) will **increase** coverage by `1.12%`.\n> The diff coverage is `45.45%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2189?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2189 +/- ##\n==========================================\n+ Coverage 80.2% 81.33% +1.12% \n==========================================\n Files 125 125 \n Lines 18444 18458 +14 \n==========================================\n+ Hits 14793 15012 +219 \n+ Misses 3651 3446 -205\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2189?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/configuration\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/2189/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2NvbmZpZ3VyYXRpb25fYmVydC5weQ==) | `100% <ø> (ø)` | :arrow_up: |\n| [transformers/modeling\\_xlm\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/2189/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3hsbV9yb2JlcnRhLnB5) | `100% <ø> (ø)` | :arrow_up: |\n| [transformers/modeling\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/2189/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2JlcnQucHk=) | `87.75% <ø> (ø)` | :arrow_up: |\n| [transformers/tokenization\\_distilbert.py](https://codecov.io/gh/huggingface/transformers/pull/2189/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl9kaXN0aWxiZXJ0LnB5) | `100% <ø> (ø)` | :arrow_up: |\n| [transformers/configuration\\_xlm\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/2189/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2NvbmZpZ3VyYXRpb25feGxtX3JvYmVydGEucHk=) | `100% <ø> (ø)` | :arrow_up: |\n| [transformers/modeling\\_tf\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/2189/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX2JlcnQucHk=) | `96.31% <ø> (ø)` | :arrow_up: |\n| [transformers/tokenization\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/2189/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl9iZXJ0LnB5) | `96.38% <ø> (ø)` | :arrow_up: |\n| [transformers/tokenization\\_xlm\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/2189/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl94bG1fcm9iZXJ0YS5weQ==) | `36.92% <0%> (-0.58%)` | :arrow_down: |\n| [transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2189/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2ZpbGVfdXRpbHMucHk=) | `71.42% <100%> (ø)` | :arrow_up: |\n| [transformers/modeling\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/2189/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2F1dG8ucHk=) | `36.36% <28.57%> (-0.53%)` | :arrow_down: |\n| ... and [9 more](https://codecov.io/gh/huggingface/transformers/pull/2189/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2189?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2189?src=pr&el=footer). Last update [2f1c745...3376adc](https://codecov.io/gh/huggingface/transformers/pull/2189?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "@julien-c Thanks for that hint! I tested the script (#1959) and it's working with XLM-R (training was sucessful). I'm going to test the base model now 😅", "This is really awesome @stefan-it!\r\nMerging now!", "@stefan-it Hi, how to save the XLM-Roberta model? I tried `torch.save` for the model but reported `*** TypeError: can't pickle SwigPyObject objects`" ]

[ { "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 because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "How do you set your seed? Personally I use the following, which sets... a lot of seeds but also useful variables for the backend.\r\n\r\n```python\r\n def set_seed(seed: Optional[int]):\r\n \"\"\" Set all seeds to make results reproducible (deterministic mode).\r\n When seed is None, disables deterministic mode. \"\"\"\r\n if seed is not None:\r\n torch.manual_seed(seed)\r\n torch.cuda.manual_seed_all(seed)\r\n torch.backends.cudnn.deterministic = True\r\n torch.backends.cudnn.benchmark = False\r\n np.random.seed(seed)\r\n random.seed(seed)\r\n os.environ['PYTHONHASHSEED'] = str(seed)\r\n```\r\n\r\nAlso make sure that you are setting your model to `.eval()`.", "Output is deterministic and reproducible on each GPU.\r\n\r\nI.e., on the AWS P3, I get Output A consistently whenever I run it with the same input\r\n\r\nOn AWS P2, I get Output B consistently whenever I run it with the same input.\r\n\r\n\r\nCould there be any chance that the default seeding is somehow based on the GPU identifier?\r\n\r\n\r\nThe other weird thing is that the divergence between Output A and Output B happens very far along in the output stream. They are the same for a long while, many hundreds of tokens, and then they diverge.\r\n\r\nI would think that a seed difference would cause them to diverge right away. ", "Hm, as far as I know different GPUs should still give consistent results with a fixed seed. Does it also happen with other generative models?", "gpu-specific numerical imprecision?", "Haven't tried other models, just GPT2.\r\n\r\nIt's also hard to provide a reproduction case, because this happens with a big seed (something like 500 words), and happens a few thousand words into the output.\r\n\r\nTo achieve this long output, which is longer than the 1024 tokens in the model vector, I'm reseeding over and over with the output, daisy chaining the last output to generate the next input. There is no nondeterminism in this reseeding process, though, and I've verified that the input to the sampling function is identical every step of the way, until that one word where one GPU finally makes a different choice.\r\n\r\nAnd this different choice is deterministic on each GPU. It always picks a different word at that spot.\r\n\r\nI can at least provide the output samples to show you what I mean.\r\n\r\nAlso, I'll leave this open for now in case someone can shed some light on this or has the same problem. But after quite a bit of testing, I've determined that it's nothing on my end that is causing the divergence. Probably some GPU-specific imprecision that I can't control. It's also not a problem, exactly, but just something unexpected that made me take a closer look at my own code.\r\n\r\nAnyway, search for the word \"hunting\" in each of these to see where the output diverges.\r\n\r\nAWS P2 output sample:\r\n\r\n> When Oedipa got home she cried, didn't know how much, for many long minutes, and then came up with that often enough, two people's wanton moments being too much. Wendell lay across the back seat of her car and was sitting up when she arrived home. She started holding him more carefully, so as not to bump his head on the window, and almost drifted off until she could think of little else. \"In other words, I won't remember my husband's name, Wendell,\" she told herself.\r\n> \r\n> \"Or maybe it won't be him, and I'll know who died.\" She opened her window and sat down with her arms on the steering wheel and turned the key. \"I won't be able to take it, won't be able to get there.\"\r\n> \r\n> She lost two hours, wasn't able to drive. She was sad. And maybe this wasn't the time to get angry. Oedipa had something of a position of privilege; she had just come through a dozen solid months with the murder of one man and a whole quarter of a century of acquaintance with that same man, when the only memory she could derive was of his unholy ghost that seemed to hide away as often as it was borne up to meet it. Was that at all real, her itchy sense that somebody was out there who wasn't quite supposed to be there, trailing slowly across the sun-kissed fields of lowlands and behind the straight and narrow lanes of what appeared to be an English village?\r\n> \r\n> What happened in the fall of 1966 and early 1967 wasn't the best of times, but if she had to go by only the state's case and two sworn affidavits, it was bad, she thought, with things festering under her. I think it's worse here than when I came here, she said herself, and then shifted her viewpoint; people are really bad here, people are all over the map, and yet she sees them all the same.\r\n> It could have all been better, she thought. \"He was even worse than he was before,\" she thought, \"and I'm the mother of his child.\" I really wish that he were here, she said, and felt a rumbling in her, remembering and disbelieving this last sentence, old hat.\r\n> \r\n> By the time she finished cooking breakfast the next morning, she felt a familiar course of fatigue (she knew this feeling, had felt it before and gone through it) but not quite because of anything she'd done. It was the same stuff of the weeks gone by. By three o'clock in the afternoon she was feverishly studying the latest installment in the The Winnipeg Free Press and wondering, if she was going to have another baby, what time they were going to have the baby. Later, after Wendell had gone to bed, and she had fallen into a restless, frenzied sleep, it became clear that those thoughts had been heading away from the child, toward her father. Her husband was supposed to come out and hug her one more time. And something strange happened: no hug.\r\n> \r\n> \r\n> Part 1\r\n> \r\n> Fifteen-thirty on that sunny October day in the early nineteen-seventies in Lake of the Woods, an hour's drive from Indiana, was a normal day. It was a typical county-issue late afternoon: a burst of snow, mostly covering the ground at eight or nine o'clock; an overweight man riding a first-class brown ticket train, in cotton dress and gold plate badge, who was carrying a sapphire metallic briefcase into St. Martinville, Oklahoma City; he stood before the crescent-shaped office of the Mulberry County Auditor; one of those large stainless steel doors opened and a ruggedly handsome man walked out in his tan, easy-looking suit, without his outer appearance a warning sign to any observer. The man was Wendell Sams, chief of police of Mulberry County, and his name was Earl Sams.\r\n> \r\n> Earl Sams had been a cop for nineteen years. He'd been born on this farm in 1917 and made it into adulthood with farm-yard kinbaku and wide experience of the milieu of farmers' wives, country festivals, \"cutesy songs and melodies and songs of the land,\" hunting a pig in a Louisiana cotton field, a hiker frolicking with a deer in the Ozark hills, living in two houses together, raising and maintaining eighty-seven kids, three cars, two planes, and a private railroad and a utility truck. (\"It wasn't very good farming and it wasn't very good trucking, but it was only ten miles from downtown Atlanta,\" Earl Sams had once said.) Then there was the acreage; old-school equipment, all sorts of \"pedestrian carts,\" tailgates and fountains of soft drinks, canned goods, canned food, that dreaded kimchi corn.\r\n> \r\n> When Earl Sams came along, the town was failing slowly. the factory and mill district around St. Martinville had been neglected, falling on hard times after the corn companies had shut their doors. It had two hospitals and a sheriff's office, but the infant would have to wait, it would have to wait. Nobody wanted to move into the county to take advantage of the increased area that Sams had planned to buy with the money he had just received in his current job. The road was lined with ranch houses and pulled up to many of them by the local back roads, where in the summer, with the grass growing deep and fast, they could have gravel runways that stretched over miles of dirt. There were a couple of old county airfields which now had strip lights and power lines just to the north. The county used to have a train depot in the 1920s, but the local farmer's crew did not like to travel from Kansas City to Hickman or Chapel Hill and start their shift, so in order to stay in business, the depot had been razed, leaving nothing but a fence and a ditto house.\r\n> \r\n\r\nAWS P3 output sample:\r\n\r\n> \r\n> When Oedipa got home she cried, didn't know how much, for many long minutes, and then came up with that often enough, two people's wanton moments being too much. Wendell lay across the back seat of her car and was sitting up when she arrived home. She started holding him more carefully, so as not to bump his head on the window, and almost drifted off until she could think of little else. \"In other words, I won't remember my husband's name, Wendell,\" she told herself.\r\n> \r\n> \"Or maybe it won't be him, and I'll know who died.\" She opened her window and sat down with her arms on the steering wheel and turned the key. \"I won't be able to take it, won't be able to get there.\"\r\n> \r\n> She lost two hours, wasn't able to drive. She was sad. And maybe this wasn't the time to get angry. Oedipa had something of a position of privilege; she had just come through a dozen solid months with the murder of one man and a whole quarter of a century of acquaintance with that same man, when the only memory she could derive was of his unholy ghost that seemed to hide away as often as it was borne up to meet it. Was that at all real, her itchy sense that somebody was out there who wasn't quite supposed to be there, trailing slowly across the sun-kissed fields of lowlands and behind the straight and narrow lanes of what appeared to be an English village?\r\n> \r\n> What happened in the fall of 1966 and early 1967 wasn't the best of times, but if she had to go by only the state's case and two sworn affidavits, it was bad, she thought, with things festering under her. I think it's worse here than when I came here, she said herself, and then shifted her viewpoint; people are really bad here, people are all over the map, and yet she sees them all the same.\r\n> It could have all been better, she thought. \"He was even worse than he was before,\" she thought, \"and I'm the mother of his child.\" I really wish that he were here, she said, and felt a rumbling in her, remembering and disbelieving this last sentence, old hat.\r\n> \r\n> By the time she finished cooking breakfast the next morning, she felt a familiar course of fatigue (she knew this feeling, had felt it before and gone through it) but not quite because of anything she'd done. It was the same stuff of the weeks gone by. By three o'clock in the afternoon she was feverishly studying the latest installment in the The Winnipeg Free Press and wondering, if she was going to have another baby, what time they were going to have the baby. Later, after Wendell had gone to bed, and she had fallen into a restless, frenzied sleep, it became clear that those thoughts had been heading away from the child, toward her father. Her husband was supposed to come out and hug her one more time. And something strange happened: no hug.\r\n> \r\n> \r\n> Part 1\r\n> \r\n> Fifteen-thirty on that sunny October day in the early nineteen-seventies in Lake of the Woods, an hour's drive from Indiana, was a normal day. It was a typical county-issue late afternoon: a burst of snow, mostly covering the ground at eight or nine o'clock; an overweight man riding a first-class brown ticket train, in cotton dress and gold plate badge, who was carrying a sapphire metallic briefcase into St. Martinville, Oklahoma City; he stood before the crescent-shaped office of the Mulberry County Auditor; one of those large stainless steel doors opened and a ruggedly handsome man walked out in his tan, easy-looking suit, without his outer appearance a warning sign to any observer. The man was Wendell Sams, chief of police of Mulberry County, and his name was Earl Sams.\r\n> \r\n> Earl Sams had been a cop for nineteen years. He'd been born on this farm in 1917 and made it into adulthood with farm-yard kinbaku and wide experience of the milieu of farmers' wives, country festivals, \"cutesy songs and melodies and songs of the land,\" hunting the dawn and the sunset, among a horticultural and industrial past that, through mites and lies, was dressed up as simple piety, tradition, or familial environment. He was in his fifties, so that he looked little of his twenty-nine years and how well he matched the image of old-time independence he had been given. He'd been a lieutenant in the Corps of Coroners of Pulaski County, Arkansas, and then-run and more or less ad-hoc detective lieutenant on Davenport Road, for a short time in Lake County. The Mulberry County office seemed to be an unchanged place, although new signs had gone up informing the new chief that there would be a transfer. And yet Earl Sams liked it here, liked the cabin, the lake and the horses, and when the announcement was made that it would be his desk, he could just have stayed.\r\n> \r\n> His first words as chief were flat, indifferent. \"To whom it may concern, sir,\" he said, \"my name is Earl Sams.\"\r\n> \r\n> He had a wife, Rosalie, and they had four children; four boys and two girls. The boys were all little, and a couple of them were to be would-be sons-in-law. The eldest of them was fourteen and half; he was usually called Denton, for something and something else; and one of the girls was sixteen, a girl who had held her own in the boys' club, although now he could't really remember any one there. The boys had always liked him, laughed at him when he played catch with them, had found him amusing to be around. He liked them all, too, although when they were adults he didn't bother to find out their names, their names, their ages or their ages' dates. They were all small and stocky, tall but skinny, and bright. They were working people, but they kept his life in the background and seemed neither to get for him the unnecessary attention nor to get annoyed when he took time out to buy the place a new color and put up new signs. He bought the land himself, with luck, in an estate in Manitoba or Kansas or Tennessee, and he kept busy with old times, with horses and the crew, mending camp and clearing the woods and digging the drift and sharpening the bow, or tracking down grubs and other insects. The crew consisted of nothing but horses: two black from Arkansas, two Indian, one old pit bull. They tended the corn and lifted pigs and other sheep, took care of the cows, mopped up the mud, raised the chickens and pulled in the fertilizer. The younger brother was kind, handy with tools, and the father was a man with a truck and a radio. He used it and offered as much service as he could. He was tall, lean and clean-cut, with the deeply cool eyes of an oil-field man or two and a shrewdly-willed attitude about his job. As Chief Sams he was even a little hard to like, not unlike the older residents of Mulberry, but it wasn't his fault. Old men make mistakes, and it was no fault of theirs.\r\n\r\n", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "I'm sorry that you're angry with Transformers library and its authors, but I'm not share your opinion. This framework is well documented, developed and updated (the most important part of each library).\r\n\r\nHowever, if you want to watch and/or train the model for the summarization task, you can refer [here](https://github.com/nlpyang/PreSumm), as said in the [README.md](https://github.com/huggingface/transformers/tree/master/examples/summarization). \r\n\r\nI share with you that, for completeness, it could be useful for many people a Python script that allows to train a summarization model with a custom dataset.", "Hi\r\nI really believe you would better off fully remove this folder, previously training part was also included but was not complete, after weeks of waiting you decided to fully remove it? why is this?\r\n\r\nPlease reconsider your decision of including such codes into the repository, have you ever asked yourself what is the point of evaluating on the already trained model, while not allowing user to actually train such models? \r\nIf the user need to revert back to presum repo, then let the user also evaluate there, there is no point of including the codes which is not complete, this is a bad choice and hurt the dignity of repo in the long run. ", "@thomwolf \r\nI put Thomas in cc, I really believe adding such incomplete codes is not proper, and hurt the dignity of this repository in the long run.", "@juliahane ... I gotta say that while I understand your frustration and what you are requesting, your attitude completely sucks and is unlikely to solicit a response from the huggingface team.\r\n\r\nThere is no way in hell you can rationalize requesting the team to do away from the inference code based on the pre-trained model just because you can't fine-tune it for your own dataset. The code is complete insofar as what it intends to do.\r\n\r\nNow ... I'd love to have this be finetunable and would love to see what the HF team produces. In the meantime, I stepped through their code and figured out what you need to do in `modeling_bertabs.py` to make this just so. I'm glad to share with you, but in the words of Base Commander Colonel Nathan Jessup, \"You gotta ask nicely.\"", "Hi\nThanks for your response. If i was not sounding nice i apologize for it.\nUnfortunately i do believe in every single word of what i said. Adding\nsummarization just for evaluation does not help then let people also revert\nback to presum for it. I really dont get the point here of adding loading\nand calling pretrained models. Unfortunately i really believe such attitude\nfrom your team in long run hurt hugging face name for sure. People see your\nrepo as the greatest repo for deep learning but if you start adding codes\nlike this which does not train and pointless from my view it does change\npeoples mind. I am sorry this is the truth. Adding summarization code\nwithout allowing user to train is pointless. Also i expect you to be more\nwelcoming towards complaints. There is really no point of loading\npretrained models from another repo and let user to call it. This is a\nlegitimate complain and your attitude of you gaurding gainst it completely\nsucks.\n\nOn Mon, Dec 16, 2019, 11:19 PM ohmeow <[email protected]> wrote:\n\n> @juliahane <https://github.com/juliahane> ... I gotta say that while I\n> understand your frustration and what you are requesting, your attitude\n> completely sucks and is unlikely to solicit a response from the huggingface\n> team.\n>\n> There is no way in hell you can rationalize requesting the team to do away\n> from the inference code based on the pre-trained model just because you\n> can't fine-tune it for your own dataset. The code is complete insofar as\n> what it intends to do.\n>\n> Now ... I'd love to have this be finetunable and would love to see what\n> the HF team produces. In the meantime, I stepped through their code and\n> figured out what you need to do in modeling_bertabs.py to make this just\n> so. I'm glad to share with you, but in the words of Base Commander Colonel\n> Nathan Jessup, \"You gotta ask nicely.\"\n>\n> —\n> You are receiving this because you were mentioned.\n> Reply to this email directly, view it on GitHub\n> <https://github.com/huggingface/transformers/issues/2186?email_source=notifications&email_token=AM3GZMZZO3EXCNUOZA3DNGTQY75IPA5CNFSM4J3B6S32YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEHAJ3TA#issuecomment-566271436>,\n> or unsubscribe\n> <https://github.com/notifications/unsubscribe-auth/AM3GZM4KS2VL5QBN5KX72GLQY75IPANCNFSM4J3B6S3Q>\n> .\n>\n", "Also i would like to add this point. Previously huggingface added\nsummarization codes that has evaluation part but it was not implemented and\nthe code was failing is several parts basically huggingface uploaded fully\nnot tested code. I respect good code. Everyone should respect writing good\ncode which is tested. Later you removed that training part and leave\nevaluation part. Still not a complete code which really serve user no\nfunctionality than calling already trained models. Both acts of adding code\nwhich breaks in at least 10 parts not at all complete in anysense like\nadding flags and then not writing conditions.... Is really something which\nhurt your name in the long run. Resulting in people losing trust in\nhuggingface.\n\nOn Mon, Dec 16, 2019, 11:41 PM julia hane <[email protected]> wrote:\n\n> Hi\n> Thanks for your response. If i was not sounding nice i apologize for it.\n> Unfortunately i do believe in every single word of what i said. Adding\n> summarization just for evaluation does not help then let people also revert\n> back to presum for it. I really dont get the point here of adding loading\n> and calling pretrained models. Unfortunately i really believe such attitude\n> from your team in long run hurt hugging face name for sure. People see your\n> repo as the greatest repo for deep learning but if you start adding codes\n> like this which does not train and pointless from my view it does change\n> peoples mind. I am sorry this is the truth. Adding summarization code\n> without allowing user to train is pointless. Also i expect you to be more\n> welcoming towards complaints. There is really no point of loading\n> pretrained models from another repo and let user to call it. This is a\n> legitimate complain and your attitude of you gaurding gainst it completely\n> sucks.\n>\n> On Mon, Dec 16, 2019, 11:19 PM ohmeow <[email protected]> wrote:\n>\n>> @juliahane <https://github.com/juliahane> ... I gotta say that while I\n>> understand your frustration and what you are requesting, your attitude\n>> completely sucks and is unlikely to solicit a response from the huggingface\n>> team.\n>>\n>> There is no way in hell you can rationalize requesting the team to do\n>> away from the inference code based on the pre-trained model just because\n>> you can't fine-tune it for your own dataset. The code is complete insofar\n>> as what it intends to do.\n>>\n>> Now ... I'd love to have this be finetunable and would love to see what\n>> the HF team produces. In the meantime, I stepped through their code and\n>> figured out what you need to do in modeling_bertabs.py to make this just\n>> so. I'm glad to share with you, but in the words of Base Commander Colonel\n>> Nathan Jessup, \"You gotta ask nicely.\"\n>>\n>> —\n>> You are receiving this because you were mentioned.\n>> Reply to this email directly, view it on GitHub\n>> <https://github.com/huggingface/transformers/issues/2186?email_source=notifications&email_token=AM3GZMZZO3EXCNUOZA3DNGTQY75IPA5CNFSM4J3B6S32YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEHAJ3TA#issuecomment-566271436>,\n>> or unsubscribe\n>> <https://github.com/notifications/unsubscribe-auth/AM3GZM4KS2VL5QBN5KX72GLQY75IPANCNFSM4J3B6S3Q>\n>> .\n>>\n>\n", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "Bert, XLnet all work fine for me", "Have you ever read in the Issues section, e.g. #1852, #1848, #1849 and #1805? They suggest different solutions for your problem, e.g. changing the input sequence limit to 128.\r\n\r\n> ## Bug\r\n> Model I am using (Bert, XLNet....):\r\n> Roberta\r\n> Language I am using the model on (English, Chinese....):\r\n> English\r\n> The problem arise when using:\r\n> \r\n> * [ ] the official example scripts: (give details)\r\n> * [x] my own modified scripts: (give details)\r\n> \r\n> ```\r\n> class QuestModel(nn.Module):\r\n> def __init__(self, n_classes=30):\r\n> super(QuestModel, self).__init__()\r\n> self.model_name = 'QuestModel'\r\n> self.bert_model = models[MODEL_NAME].from_pretrained(MODEL_NAME)\r\n> self.fc = nn.Linear(LINEAR_LAYER[MODEL_NAME], n_classes)\r\n> self.dropout = nn.Dropout(p=0.2)\r\n> \r\n> def forward(self, ids, seg_ids):\r\n> attention_mask = (ids > 0).float()\r\n> layers, pool_out = self.bert_model(input_ids=ids, token_type_ids=seg_ids, attention_mask=attention_mask)\r\n> \r\n> out = self.dropout(pool_out)\r\n> logit = self.fc(out)\r\n> return logit\r\n> \r\n> ......\r\n> for i, (ids, seg_ids, labels) in enumerate(data_loader):\r\n> ids, seg_ids, labels = ids.cuda(), seg_ids.cuda(), labels.cuda()\r\n> outputs = data_parallel(model, (ids, seg_ids))\r\n> scores = torch.sigmoid(outputs)\r\n> loss = loss_fn(outputs, labels)\r\n> # loss = custom_loss(pred, y_batch.to(device))\r\n> preds.append(outputs.cpu().numpy())\r\n> original.append(labels.cpu().numpy())\r\n> \r\n> avg_loss += loss.item() / len(data_loader)\r\n> ......\r\n> ```\r\n> \r\n> The tasks I am working on is:\r\n> \r\n> * [ ] an official GLUE/SQUaD task: (give the name)\r\n> * [x] my own task or dataset: (give details)\r\n> Google quest QA label\r\n> \r\n> ## To Reproduce\r\n> Steps to reproduce the behavior:\r\n> \r\n> 1. I have a script that worked fine when I use bert from transformers package\r\n> 2. Then I change tokenizer and model to roberta\r\n> 3. Always got the following errors\r\n> \r\n> ```\r\n> /opt/conda/conda-bld/pytorch_1565272279342/work/aten/src/THC/THCTensorIndex.cu:361: void indexSelectLargeIndex(TensorInfo<T, IndexType>, TensorInfo<T, IndexType>, TensorInfo<long, IndexType>, int, int, IndexType, IndexType, long) [with T = float, IndexType = unsigned int, DstDim = 2, SrcDim = 2, IdxDim = -2, IndexIsMajor = true]: block: [130,0,0], thread: [95,0,0] Assertion `srcIndex < srcSelectDimSize` failed.\r\n> Traceback (most recent call last):\r\n> File \"/mnt/home/dunan/Learn/Kaggle/google_quest/kaggle_google_quest/train.py\", line 267, in <module>\r\n> train(args)\r\n> File \"/mnt/home/dunan/Learn/Kaggle/google_quest/kaggle_google_quest/train.py\", line 138, in train\r\n> score, val_loss = predict(model, val_loader)\r\n> File \"/mnt/home/dunan/Learn/Kaggle/google_quest/kaggle_google_quest/train.py\", line 252, in predict\r\n> preds.append(outputs.cpu().numpy())\r\n> RuntimeError: CUDA error: device-side assert triggered\r\n> ```\r\n> \r\n> ## Environment\r\n> * OS: CentOS\r\n> * Python version: 3.6\r\n> * PyTorch version: 1.2\r\n> * PyTorch Transformers version (or branch): 2.2.2\r\n> * Using GPU ? Yes\r\n> * Distributed of parallel setup ? All happened when I using 1, 2, 4 Gpu\r\n> * Any other relevant information:\r\n> \r\n> ## Additional context", "Currently, roberta model has a problem when handling multiple sentences.\r\nJust removing token_type_ids from input or a bit modification will work.\r\n(https://github.com/huggingface/transformers/issues/1538)", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "I'm **not** able to import T5Tokenizer in my environment: I received an **ImportError** exception.\r\nI'm using Python 3.6.9, OS Ubuntu 16.04, Transformers 2.2.2 (installed now with `pip install transformers`), PyTorch 1.3.1 and TensorFlow 2.0. What am I missing?\r\nThe stack trace is the following:\r\n```\r\n>>> import transformers\r\n>>> transformers.__version__\r\n>>> '2.2.2'\r\n>>> from transformers import T5Tokenizer\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\nImportError: cannot import name 'T5Tokenizer'\r\n```\r\n\r\nI can't import `T5Model` as well.\r\n\r\n> ## Bug\r\n> Model I am using (Bert, XLNet....): T5\r\n> \r\n> ## To Reproduce\r\n> Steps to reproduce the behavior:\r\n> \r\n> 1. Load T5Tokenizer\r\n> 2. Try getting the CLS or SEP token: `tokenizer.sep_token` or `tokenizer.cls_token`\r\n> 3. An error will be raised \"Using cls_token, but it is not set yet.\"\r\n> \r\n> Running the latest commit on the master branch.\r\n> \r\n> I imagine that the T5 implementation is not complete yet, but I thought I'd point it out anyway. I figured that T5 expects the same input as BERT, as stated in the source:\r\n> \r\n> https://github.com/huggingface/transformers/blob/e92bcb7eb6c5b9b6ed313cc74abaab50b3dc674f/transformers/modeling_t5.py#L651-L658\r\n> \r\n> As an aside, I saw this incorrect mention of XLNetTokenizer in the T5Tokenizer. Probably overlooked while working on adding _yet another model_ to transformers. You guys are crazy!\r\n> \r\n> https://github.com/huggingface/transformers/blob/e92bcb7eb6c5b9b6ed313cc74abaab50b3dc674f/transformers/tokenization_t5.py#L112-L120", "@TheEdoardo93 T5 is _not_ part of the 2.2.2 release. In other words, it is not part of a PyPi release yet. You'll need to uninstall transformers and install from this repo.", "I've installed Transformers from source with `pip install git+https://github.com/huggingface/transformers.git` and now it works as expected! Thank you\r\n\r\n> @TheEdoardo93 T5 is _not_ part of the 2.2.2 release. In other words, it is not part of a PyPi release yet. You'll need to uninstall transformers and install from this repo.", "Yes, thanks for trying it already @BramVanroy.\r\n\r\nCurrently, T5 is still missing a few features to be easily usable.\r\nI'll add them over the coming days/weeks.\r\n\r\nHere is what we still need and we plan to add:\r\n- a clean sequence generation API (T5 is designed to be used in a text generation setting, you should read the paper if you haven't already by the way, it's a great paper) <= working on this right now at #1840 \r\n- a clean way to do model parallelism to spread the model on several GPUs. For instance, the biggest checkpoint is 42 GB (!) so you'll need a few GPU only to load the model <= Working on this after the seq generation (first draft at #2165 )\r\n- a script to pre-process GLUE/SQUAD to set them in text generation setting <= Later", "I just skimmed through https://github.com/huggingface/transformers/pull/1840 and it's great to see how you're focusing on the user-experience @thomwolf. I think that that's a very important thing, together with well structured and written documentation. The field is changing rapidly, and especially for people who are working more on the linguistics side of things or who are just getting into transformers, a clear API is a must to get started. Thanks a lot for that (and your and the team's work in general, of course)! \r\n\r\nConcerning https://github.com/huggingface/transformers/pull/2165, why don't you just require people to rent TPUv3-2048 with 32**TB** of memory? No biggey. I kid of course! Nice to see that efforts are underway to bring these big models to consumers as well. This work completely goes over my head, but I am curious to see what the end result is. Would that mean that we could load that 42GB checkpoint on our 4x V100 (16GB each)?\r\n\r\nPerhaps this issue should stay open and be closed when T5 is finished?", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "These are logits, i.e. unnormalized scores for each possible token at the masked token position. You can convert them in (normalized) probabilities by taking their softmax. I don't think you can really assign any unit to these scores, in particular, because they are not normalized so you can add any constant value to all these scores (as long as it's the same value for all tokens in the vocabulary) and still get the same probabilities after applying a softmax.\r\n\r\nWe could return a softmax out of the model but if you only want to compute the argmax for instance (the most likely token), you can directly use these outputs so we don't want to force additional compute if some people don't need it.\r\n\r\nDuring training we don't use these output but the cross-entropy loss. The cross-entropy loss is obtained by first computing the logarithm of the softmax of these scores (log-probabilities) and then the negative log-likelihood of the target labels under this distribution. This is actually computed in one step by `torch.nn.CrossEntropyLoss` and returned as the loss of the model which is the first element of the tuple when you supply `mlm_labels` to a `XXXForMaskedLM` model.", "Hey @thomwolf, thanks for taking the time to help me better understand the internals of the language models! Coincidentally, I was reading through this excellent article by Chip Huyen (@chiphuyen) (https://thegradient.pub/understanding-evaluation-metrics-for-language-models/). \r\n\r\nBut if I understand you correctly, perplexity is not used in practice as a metric (I assume that it can't be evaluated anyway). Instead, CrossEntropyLoss is used, dealing with the MLM problem as a classification problem over C classes where C is the size of the vocabulary, correct? The labels would then be (presumably, internally) one-hot encoded vocabulary where there is only one `1` which is the expected token? \r\n\r\nFor some reason I always thought that MLM involved perplexity or multi-label classification, i.e. where mask could have multiple correct tokens. I'm glad to now get a better understanding, so thanks again for your time.", "Yeah this article by @chiphuyen is really great, I keep sharing it. I hope she writes more NLP articles in the future 😄 ", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[ { "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 because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "Have you resolved this issue? I observed a much worse result." ]

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

[]

[ "Do you mean lines 66-67 and 153 in the [modeling_albert.py](https://github.com/huggingface/transformers/blob/master/transformers/modeling_albert.py) script?\r\n\r\n> ## Feature\r\n> The conda package in conda forge channel (v2.1.1) is not the latest released version (v2.2.2) so the ALBERT model is missing from the package.\r\n> \r\n> ## Motivation\r\n> In conda environment we need the latest packages containing the ALBERT model.\r\n> \r\n> ## Additional context\r\n> Also, for the ALBERT model in its code it has printing comments for each 12 layers. It shouldn't be there I guess in a production ready version.", "@TheEdoardo93 in the last pip package in line 289 there was `print(\"Layer index\", layer_index)` that is removed right now.", "The last pip package (2.2.2) should not be outputting the layers. Could you please tell me what is the output in your console when running the following snippet?\r\n\r\n```py\r\nfrom transformers import AlbertModel, __version__\r\nimport torch\r\n\r\nprint(__version__)\r\nmodel = AlbertModel.from_pretrained(\"albert-base-v1\")\r\noutput = model(torch.tensor([[1,2,3]]))\r\n```", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "And yes, the above code does fix it.\r\n\r\nExample:\r\n--stop_token=\"wake\"\r\n\r\ntext = ' to the same conclusion: that her husband, the guy she had chosen for her favorite, had been'\r\n\r\ntrimmed text = ' to the same conclusion: that her husband, the guy she had chosen for her favorite, had been'\r\n\r\n\r\n\r\ntext = ' murdered the previous spring. Oedipa closed her eyes and tried to wake up.'\r\n\r\ntrimmed text = ' murdered the previous spring. Oedipa closed her eyes and tried to '", "Thank you for raising this issue @jasonrohrer, there indeed was an error with the stop token. It's been fixed in 18a879f." ]

[]

[]

[ "Hello @graykode \r\nYou can use whichever teacher you want, however in the method we propose, you need to make sure that the vocabularies match (knowledge distillation loss is applied to the distributions over the vocabulary).\r\nVictor", "Thanks for your advice! I will close this issue" ]

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

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2178?src=pr&el=h1) Report\n> Merging [#2178](https://codecov.io/gh/huggingface/transformers/pull/2178?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/26e04e51ef0774e681784d7be900c1119d46c52e?src=pr&el=desc) will **decrease** coverage by `<.01%`.\n> The diff coverage is `66.66%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2178?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2178 +/- ##\n==========================================\n- Coverage 73.73% 73.73% -0.01% \n==========================================\n Files 87 87 \n Lines 14921 14919 -2 \n==========================================\n- Hits 11002 11000 -2 \n Misses 3919 3919\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2178?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/modeling\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2178/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ191dGlscy5weQ==) | `63.91% <0%> (ø)` | :arrow_up: |\n| [src/transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2178/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHMucHk=) | `92.35% <100%> (-0.03%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2178?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2178?src=pr&el=footer). Last update [26e04e5...7870a49](https://codecov.io/gh/huggingface/transformers/pull/2178?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "@thomwolf \r\nThis pull request is ready for review", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2177?src=pr&el=h1) Report\n> Merging [#2177](https://codecov.io/gh/huggingface/transformers/pull/2177?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/e92bcb7eb6c5b9b6ed313cc74abaab50b3dc674f?src=pr&el=desc) will **increase** coverage by `<.01%`.\n> The diff coverage is `100%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2177?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2177 +/- ##\n==========================================\n+ Coverage 81.35% 81.36% +<.01% \n==========================================\n Files 120 120 \n Lines 18254 18256 +2 \n==========================================\n+ Hits 14851 14854 +3 \n+ Misses 3403 3402 -1\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2177?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2177/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl91dGlscy5weQ==) | `90.74% <100%> (+0.24%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2177?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2177?src=pr&el=footer). Last update [e92bcb7...cc01351](https://codecov.io/gh/huggingface/transformers/pull/2177?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "I've seen there was a similar PR https://github.com/huggingface/transformers/pull/1881 but focusing on special tokens. The current one is a bit more generic IMHO. You'll tell me what you think about it.", "Thank you for the detailed report, this is very cool. It looks good to me, great work @mandubian!", "Awesome, thanks a lot @mandubian " ]

[]

[]

[ "Can you please post the required version numbers? (Should be in the issue template)", "the version numbers is 2.2.0", "- Is it answering every question? for V2 you might want this flag passed\r\nhttps://github.com/huggingface/transformers/blob/master/examples/run_squad.py#L409\r\n\r\n- Your model is cased?\r\nhttps://github.com/huggingface/transformers/blob/master/examples/run_squad.py#L428\r\n\r\n--- \r\nEdit: I should say I haven't actually done this. I just know this flag exists. ", "@WenTingTseng If you want to use SQuAD v2.0, you have to pass the `--version_2_with_negative` argument to `run_squad.py`, otherwise the model supposes every question has at least an answer. Without that flag, you are basically not \"learning\" the no-answerable part.", "ok, problem is resolved\r\nThanks a lot" ]

[]

[]

[ "ok" ]

[ { "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 because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "having the same question... why add two seps instead of one? " ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2173?src=pr&el=h1) Report\n> Merging [#2173](https://codecov.io/gh/huggingface/transformers/pull/2173?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/7bd11dda6f43656cf0a3891b7f61a67196d233b4?src=pr&el=desc) will **decrease** coverage by `1.35%`.\n> The diff coverage is `9.09%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2173?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2173 +/- ##\n==========================================\n- Coverage 80.79% 79.43% -1.36% \n==========================================\n Files 113 113 \n Lines 17013 17067 +54 \n==========================================\n- Hits 13745 13558 -187 \n- Misses 3268 3509 +241\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2173?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/data/metrics/squad\\_metrics.py](https://codecov.io/gh/huggingface/transformers/pull/2173/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2RhdGEvbWV0cmljcy9zcXVhZF9tZXRyaWNzLnB5) | `0% <0%> (ø)` | :arrow_up: |\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/2173/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `53.2% <21.21%> (-18.57%)` | :arrow_down: |\n| [transformers/data/processors/squad.py](https://codecov.io/gh/huggingface/transformers/pull/2173/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2RhdGEvcHJvY2Vzc29ycy9zcXVhZC5weQ==) | `14.75% <5.5%> (+0.56%)` | :arrow_up: |\n| [transformers/modeling\\_tf\\_pytorch\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2173/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3B5dG9yY2hfdXRpbHMucHk=) | `9.72% <0%> (-85.42%)` | :arrow_down: |\n| [transformers/tests/modeling\\_tf\\_common\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2173/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2NvbW1vbl90ZXN0LnB5) | `80.51% <0%> (-16.42%)` | :arrow_down: |\n| [transformers/modeling\\_xlnet.py](https://codecov.io/gh/huggingface/transformers/pull/2173/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3hsbmV0LnB5) | `72.21% <0%> (-2.33%)` | :arrow_down: |\n| [transformers/modeling\\_ctrl.py](https://codecov.io/gh/huggingface/transformers/pull/2173/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2N0cmwucHk=) | `94.27% <0%> (-2.21%)` | :arrow_down: |\n| [transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/2173/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX29wZW5haS5weQ==) | `80.13% <0%> (-1.33%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2173?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2173?src=pr&el=footer). Last update [7bd11dd...805c21a](https://codecov.io/gh/huggingface/transformers/pull/2173?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Really nice job!\r\nHere are my results of RoBERTa-large on SQuAD using this PR:\r\n`Results: {'exact': 84.52792049187232, 'f1': 88.0216698977779, 'total': 11873, 'HasAns_exact': 80.90418353576248, 'HasAns_f1': 87.9017015344667, 'HasAns_total': 5928, 'NoAns_exact': 88.1412952060555, 'NoAns_f1': 88.1412952060555, 'NoAns_total': 5945, 'best_exact': 84.52792049187232, 'best_exact_thresh': 0.0, 'best_f1': 88.02166989777776, 'best_f1_thresh': 0.0}`\r\nThe hyper-parameters are as follows:\r\n`python ./examples/run_squad.py \\\r\n --model_type roberta \\\r\n --model_name_or_path roberta-large \\\r\n --do_train \\\r\n --do_eval \\\r\n --do_lower_case \\\r\n --train_file data/squad2/train-v2.0.json \\\r\n --predict_file data/squad2/dev-v2.0.json \\\r\n --learning_rate 2e-5 \\\r\n --num_train_epochs 2 \\\r\n --max_seq_length 384 \\\r\n --doc_stride 128 \\\r\n --output_dir ./models_roberta/large_squad2 \\\r\n --per_gpu_eval_batch_size=6 \\\r\n --per_gpu_train_batch_size=6 \\\r\n --save_steps 10000 --warmup_steps=500 --weight_decay=0.01 --overwrite_cache --overwrite_output_dir --threads 24 --version_2_with_negative`\r\n", "Really nice, thanks a lot @erenup " ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2172?src=pr&el=h1) Report\n> Merging [#2172](https://codecov.io/gh/huggingface/transformers/pull/2172?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/d46147294852694d1dc701c72b9053ff2e726265?src=pr&el=desc) will **increase** coverage by `0.48%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2172?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2172 +/- ##\n==========================================\n+ Coverage 80.32% 80.81% +0.48% \n==========================================\n Files 114 113 -1 \n Lines 17102 16999 -103 \n==========================================\n Hits 13737 13737 \n+ Misses 3365 3262 -103\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2172?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/data/processors/squad.py](https://codecov.io/gh/huggingface/transformers/pull/2172/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2RhdGEvcHJvY2Vzc29ycy9zcXVhZC5weQ==) | `14.49% <0%> (+0.3%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2172?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2172?src=pr&el=footer). Last update [d461472...fb92209](https://codecov.io/gh/huggingface/transformers/pull/2172?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

[ { "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 because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "As it's only a cosmetic change and for the sake of not breaking backward compat over cosmetic issues I'm reluctant to change this." ]

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

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2170?src=pr&el=h1) Report\n> Merging [#2170](https://codecov.io/gh/huggingface/transformers/pull/2170?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/c8ed1c82c8a42ef700d4129d227fa356385c1d60?src=pr&el=desc) will **decrease** coverage by `<.01%`.\n> The diff coverage is `0%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2170?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2170 +/- ##\n==========================================\n- Coverage 80.35% 80.34% -0.01% \n==========================================\n Files 114 114 \n Lines 17095 17097 +2 \n==========================================\n Hits 13736 13736 \n- Misses 3359 3361 +2\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2170?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/2170/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2JlcnQucHk=) | `87.43% <0%> (-0.32%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2170?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2170?src=pr&el=footer). Last update [c8ed1c8...e3c65da](https://codecov.io/gh/huggingface/transformers/pull/2170?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "Hi, there is a fundamental difference between PyTorch and TensorFlow in that the losses for PyTorch can be computed both inside the model forward method as well as outside, whereas it is only outside for TensorFlow.\r\n\r\nThis makes a difference when comparing the torch script and keras fit, as our GPT-2 implementation automatically computes the loss when giving the `labels` argument to the model, which would not be the case for TensorFlow. It computes the loss by shifting the examples by one index and comparing the model's token prediction to the true token value.\r\n\r\nIn order to train your GPT-2 (or DistilGPT-2) model on language modeling, you would have to create a dataset with:\r\n\r\n- the examples\r\n- the labels: these are the examples but shifted by one index so that the model may compare its prediction of the following token compared to the true token.\r\n\r\nLet me know if you have additional questions.", "@LysandreJik Thank you!\r\n\r\nAn additional question:\r\nI have a text file with one sentence per line, and an end of sentence token at the end of each line. \r\n\r\nIt seems I should concatenate the text into one long string and sweep a \"window\" of a specified size over the text like so:\r\n```\r\nself.examples = []\r\n with open(file_path, encoding=\"utf-8\") as f:\r\n text = f.read()\r\n\r\n tokenized_text = tokenizer.convert_tokens_to_ids(tokenizer.tokenize(text))\r\n\r\n for i in range(0, len(tokenized_text)-block_size+1, block_size): # Truncate in block of block_size\r\n self.examples.append(tokenizer.build_inputs_with_special_tokens(tokenized_text[i:i+block_size]))\r\n```\r\nand so in order to get examples I would need to specify a length of tokens for training and my inputs would be `examples_chunk[:-1]` and my labels would be `examples_chunk[1:]` ?", "Yes, you could also use `tokenizer.encode(text)`, rather than `tokenize` > `convert_tokens_to_ids` > `build_inputs_with_special_tokens`.", "@LysandreJik \r\n\r\nUnfortunately I am still getting NaN loss and no training? Here is the code, which I assume is still not correct somehow but I cannot seem to figure out why. \r\n\r\n```\r\nwith open(file_path, encoding=\"utf-8\") as f:\r\n text = f.read()\r\n \r\ntokenized_text = tokenizer.encode(text)\r\n\r\nexamples = []\r\nblock_size = 100\r\nfor i in range(0, len(tokenized_text)-block_size+1, block_size): # Truncate in block of block_size\r\n examples.append(tokenized_text[i:i+block_size])\r\n\r\ninputs, labels = [], []\r\nfor ex in examples:\r\n inputs.append(ex[:-1])\r\n labels.append(ex[1:])\r\n\r\ndataset= tf.data.Dataset.from_tensor_slices((inputs,labels))\r\n\r\nBATCH_SIZE = 16\r\nBUFFER_SIZE = 10000\r\n\r\ndataset = dataset.shuffle(BUFFER_SIZE).batch(BATCH_SIZE, drop_remainder=True)\r\n\r\noptimizer = tf.keras.optimizers.Adam(learning_rate=3e-5, epsilon=1e-08, clipnorm=1.0)\r\nloss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)\r\nmetric = tf.keras.metrics.SparseCategoricalAccuracy('accuracy')\r\nmodel.compile(optimizer=optimizer, loss=[loss, None, None, None, None, None, None], metrics=[metric])\r\nmodel.fit(dataset, epochs=3)\r\n```", "What is your input file? I'm running your code and I do get a decreasing loss alongside an increasing accuracy. Here's a [gist](https://gist.github.com/LysandreJik/c958925768eb6a9a72609ea99561d1cb) with the self-contained code (the text is in the file), let me know if running this still outputs a NaN loss.", "Thanks so much for your help. That code works just as you described with your text file, and also works fine when I use my own text file. \r\n\r\nI discovered that the issue is coming from adding special tokens to the tokenizer. \r\n\r\nMy text file is made up of one sentence per line such as:\r\n\r\n```\r\n<start> this is an example sentence from my text file <end>\r\n<start> this is line two of my file <end>\r\n```\r\n\r\nWhen I don't change the bos_token and eos_token, I get decreasing loss and increasing accuracy. \r\nAdding the following code is what results in a NaN loss:\r\n```\r\n#change eos and bos tokens\r\nspecial_tokens_dict = {'bos_token':\"<start>\", 'eos_token':\"<end>\"}\r\ntokenizer.add_special_tokens(special_tokens_dict)\r\n```\r\n\r\nAny idea why this could be? Thank you again for the help. \r\n\r\nEDIT: included the code with the problem block \r\n\r\n```\r\nwith open(file_path, encoding=\"utf-8\") as f:\r\n text = f.read()\r\n\r\n#change eos and bos tokens\r\nspecial_tokens_dict = {'bos_token':\"<start>\", 'eos_token':\"<end>\"}\r\ntokenizer.add_special_tokens(special_tokens_dict)\r\n\r\ntokenized_text = tokenizer.encode(text)\r\n\r\nexamples = []\r\nblock_size = 100\r\nfor i in range(0, len(tokenized_text)-block_size+1, block_size): # Truncate in block of block_size\r\n examples.append(tokenized_text[i:i+block_size])\r\n\r\ninputs, labels = [], []\r\nfor ex in examples:\r\n inputs.append(ex[:-1])\r\n labels.append(ex[1:])\r\n\r\ndataset= tf.data.Dataset.from_tensor_slices((inputs,labels))\r\n\r\nBATCH_SIZE = 16\r\nBUFFER_SIZE = 10000\r\n\r\ndataset = dataset.shuffle(BUFFER_SIZE).batch(BATCH_SIZE, drop_remainder=True)\r\n\r\noptimizer = tf.keras.optimizers.Adam(learning_rate=3e-5, epsilon=1e-08, clipnorm=1.0)\r\nloss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)\r\nmetric = tf.keras.metrics.SparseCategoricalAccuracy('accuracy')\r\nmodel.compile(optimizer=optimizer, loss=[loss, None, None, None, None, None, None], metrics=[metric])\r\nmodel.fit(dataset, epochs=3)\r\n```\r\n\r\nIn following the documentation, it appears that any time I try and run the command :\r\n`model.resize_token_embeddings(len(tokenizer))` I get a `NotImplementedError`\r\n\r\nIf, however, I assign the bos and eos tokens when I first create the tokenizer:\r\n`tokenizer = GPT2Tokenizer.from_pretrained(\"distilgpt2\", bos_token='<start>', eos_token='<end>')`\r\n\r\nTraining results in decreasing loss and increasing accuracy. I realized this thread has drifted quite a bit, so I would be happy to close this and start another tokenizer thread. ", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "In [this](https://github.com/pytorch/pytorch/issues/24018) thread on PyTorch's GitHub, they said that this bug has been fixed. In more details, _\"this bug was solved in cublas 10.2.0.186. The latest public version of cublas is 10.2.1.243 that was released with CUDA 10.1 Update 2.\"_", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[ { "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 } ]

[ "@LysandreJik - Note changes above.", "I believe the issue stems from the fact that the model cannot be evaluated unless it has been trained. A workaround is to specify the `model_name_or_path` to be the same as the `output_dir` so that it loads that when evaluating, but it isn't the best user experience.\r\n\r\nI'm thinking of forcing the evaluation to load the model from `model_name_or_path` rather than from `output_dir` when there is no `do_train` argument is specified. What do you think?", "That's exactly the situation, and exactly the behavior that I believe would work.\r\nThe training takes many hours (~20?) and I only want to do it once. Then I cache that in a central location and all of my many predict runs use that. ", "Let me know if the latest commit (c8ed1c8) fixes this issue.", "Well... no and yes. The \"no\" is that I had trouble with dependencies when running that commit ( issues with missing \"past\" module). The \"yes\" is that I was able to slot in the fix to the 2.1.1 version and the fix worked.\r\n\r\nNot sure what my problems are with running both from master and from c8ed1c8. But at least there's a way forward. For unattended clone and operate, I'd either need the patch to be applied to 2.1.1, or I'll need some guidance about dependency issues with c8ed1c8. \r\n\r\nThanks!", "When you install from master (`pip install git+https://github.com/huggingface/transformers`) you run into issues? Do you mind copying the error message along with your software versions?\r\n\r\nWe'll push a patch (2.2.2) later today.", "Works great. Don't know what was going wrong with my prior attempts. \r\n\r\nThanks!!\r\n", "Glad to hear that. Feel free to re-open if you have similar issues." ]

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

[]

[ "We have no scripts for pre-training, but we do have scripts for fine-tuning (which seems to be what you want to do). Take a look at [run_lm_finetuning.py](https://github.com/huggingface/transformers/blob/master/examples/run_lm_finetuning.py) for more information. \r\n\r\nWe don't have examples that do NSP however, as it was proven with RoBERTa to not be particularly useful for training. You'll have to code it yourself or find an implementation somewhere if you want to train on that loss.", "@LysandreJik I got it. thank you !", "@JiangYanting If you want to do LM finetuning incl. NSP, you might wanna have a look at [FARM](https://github.com/deepset-ai/FARM). \r\nThere's an example script [here](https://github.com/deepset-ai/FARM/blob/master/examples/lm_finetuning.py ). \r\n\r\nFrom our experience it depends a lot on the domain whether NSP makes sense. In some industry applications, we made good experience with also adding other auxiliary tasks in this phase of model training (e.g. an additional classification task for available tags of documents / sentences).\r\n\r\n", "@tholor Wow, that's so cool ! I would have a try after I take a final exam^_^. thank you very much !", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[ { "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 because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2164?src=pr&el=h1) Report\n> Merging [#2164](https://codecov.io/gh/huggingface/transformers/pull/2164?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/e92bcb7eb6c5b9b6ed313cc74abaab50b3dc674f?src=pr&el=desc) will **decrease** coverage by `0.99%`.\n> The diff coverage is `90.53%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2164?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2164 +/- ##\n========================================\n- Coverage 81.35% 80.35% -1% \n========================================\n Files 120 122 +2 \n Lines 18254 18335 +81 \n========================================\n- Hits 14851 14734 -117 \n- Misses 3403 3601 +198\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2164?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tests/modeling\\_tf\\_openai\\_gpt\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2164/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX29wZW5haV9ncHRfdGVzdC5weQ==) | `94.73% <ø> (ø)` | :arrow_up: |\n| [transformers/tests/modeling\\_tf\\_bert\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2164/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2JlcnRfdGVzdC5weQ==) | `96.22% <ø> (ø)` | :arrow_up: |\n| [transformers/tests/modeling\\_tf\\_roberta\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2164/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX3JvYmVydGFfdGVzdC5weQ==) | `75.2% <ø> (ø)` | :arrow_up: |\n| [transformers/tests/modeling\\_ctrl\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2164/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX2N0cmxfdGVzdC5weQ==) | `93.57% <ø> (ø)` | :arrow_up: |\n| [transformers/tests/modeling\\_common\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2164/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX2NvbW1vbl90ZXN0LnB5) | `75.64% <ø> (ø)` | :arrow_up: |\n| [transformers/tests/modeling\\_tf\\_t5\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2164/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX3Q1X3Rlc3QucHk=) | `92.77% <ø> (ø)` | :arrow_up: |\n| [transformers/tests/modeling\\_bert\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2164/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX2JlcnRfdGVzdC5weQ==) | `97.07% <ø> (ø)` | :arrow_up: |\n| [transformers/tests/modeling\\_gpt2\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2164/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX2dwdDJfdGVzdC5weQ==) | `94.16% <ø> (ø)` | :arrow_up: |\n| [transformers/tests/modeling\\_tf\\_ctrl\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2164/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2N0cmxfdGVzdC5weQ==) | `94.05% <ø> (ø)` | :arrow_up: |\n| [transformers/tests/modeling\\_tf\\_distilbert\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2164/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2Rpc3RpbGJlcnRfdGVzdC5weQ==) | `99.08% <ø> (ø)` | :arrow_up: |\n| ... and [50 more](https://codecov.io/gh/huggingface/transformers/pull/2164/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2164?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2164?src=pr&el=footer). Last update [e92bcb7...1bbdbac](https://codecov.io/gh/huggingface/transformers/pull/2164?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "awesome", "Ok merging" ]

[ { "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 because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "We are still settling on the proper API for the pytorch version, so it will probably be awhile (months) before we make a tensorflow version. Feel free to take a stab at it, of course!", "Thank you for the reply :) I guess I will use PyTorch then.\r\nOn 11 Feb 2020, 16:28 +0100, Sam Shleifer <[email protected]>, wrote:\r\n\r\nWe are still settling on the proper API for the pytorch version, so it will probably be awhile (months) before we make a tensorflow version. Feel free to take a stab at it, of course!\r\n\r\n—\r\nYou are receiving this because you authored the thread.\r\nReply to this email directly, view it on GitHub<https://github.com/huggingface/transformers/issues/2163?email_source=notifications&email_token=ADZLZXGTOQOXZ37FREVFBZLRCK723A5CNFSM4J2MJBQKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOELM262Q#issuecomment-584691562>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ADZLZXDXDM3BSWXJF5D375TRCK723ANCNFSM4J2MJBQA>.\r\n" ]

[]

[]

[ "Hello, can you try with the patch that was released today (2.2.2) and let me know if it works for you?", "By updating the Transformers library from 2.2.1 to 2.2.2, **it works as expected without the bug** highlighted by @madrugado.\r\n\r\nMy environment is the following:\r\n- **Python** 3.6.9\r\n- **OS**: Ubuntu 16.04\r\n- **Transformers**: 2.2.2 (installed from PyPi with `pip install transformers`)\r\n- **PyTorch**: 1.3.1.\r\n- **TensorFlow**: 2.0\r\n\r\nThe stack trace is the following: \r\n```\r\nPython 3.6.9 |Anaconda, Inc.| (default, Jul 30 2019, 19:07:31) \r\n[GCC 7.3.0] on linux\r\nType \"help\", \"copyright\", \"credits\" or \"license\" for more information.\r\n>>> import transformers\r\n/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:541: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\r\n _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\r\n/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:542: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\r\n _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\r\n/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:543: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\r\n _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\r\n/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:544: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\r\n _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\r\n/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:545: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\r\n _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\r\n/home/<user>/anaconda3/envs/huggingface/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:550: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\r\n np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\r\n>>> from transformers import BertTokenizer\r\n>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')\r\n>>> text='Hello, my name is Edward'\r\n>>> temp = tokenizer.encode(text, add_special_tokens=True, max_length=50, pad_to_max_length=True)\r\n>>> temp\r\n[101, 7592, 1010, 2026, 2171, 2003, 3487, 102, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]\r\n>>> \r\n```\r\n\r\n> Hello, can you try with the patch that was released today (2.2.2) and let me know if it works for you?", "I also confirm that with 2.2.2 version everything is working fine. Thanks!", "There is no clear documentation on `pad_to_max_length` param I had hard time finding this. It would be great if it is added to docs, or if it is present can you point me to that page. Thanks" ]

[]

[]

[ "Is now solved by this [PR](https://github.com/huggingface/transformers/pull/2494)\r\nThanks a lot!", "Yes, thanks for the contribution @perdix!" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2160?src=pr&el=h1) Report\n> Merging [#2160](https://codecov.io/gh/huggingface/transformers/pull/2160?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/f061606277322a013ec2d96509d3077e865ae875?src=pr&el=desc) will **increase** coverage by `0.04%`.\n> The diff coverage is `49.69%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2160?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2160 +/- ##\n==========================================\n+ Coverage 80.32% 80.37% +0.04% \n==========================================\n Files 122 127 +5 \n Lines 18342 19000 +658 \n==========================================\n+ Hits 14734 15272 +538 \n- Misses 3608 3728 +120\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2160?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2160/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3V0aWxzLnB5) | `88.91% <15.38%> (-2.55%)` | :arrow_down: |\n| [transformers/configuration\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/2160/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2NvbmZpZ3VyYXRpb25fYXV0by5weQ==) | `44.68% <33.33%> (-0.78%)` | :arrow_down: |\n| [transformers/tokenization\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/2160/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl9hdXRvLnB5) | `59.18% <33.33%> (-1.69%)` | :arrow_down: |\n| [transformers/modeling\\_unilm.py](https://codecov.io/gh/huggingface/transformers/pull/2160/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3VuaWxtLnB5) | `36.95% <36.95%> (ø)` | |\n| [transformers/modeling\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/2160/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2F1dG8ucHk=) | `38.79% <53.84%> (+1.89%)` | :arrow_up: |\n| [transformers/tests/tokenization\\_unilm\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2160/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL3Rva2VuaXphdGlvbl91bmlsbV90ZXN0LnB5) | `55% <55%> (ø)` | |\n| [transformers/configuration\\_unilm.py](https://codecov.io/gh/huggingface/transformers/pull/2160/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2NvbmZpZ3VyYXRpb25fdW5pbG0ucHk=) | `87.09% <87.09%> (ø)` | |\n| [transformers/tests/modeling\\_unilm\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/2160/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3VuaWxtX3Rlc3QucHk=) | `93.75% <93.75%> (ø)` | |\n| [transformers/tokenization\\_unilm.py](https://codecov.io/gh/huggingface/transformers/pull/2160/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl91bmlsbS5weQ==) | `94.73% <94.73%> (ø)` | |\n| ... and [10 more](https://codecov.io/gh/huggingface/transformers/pull/2160/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2160?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2160?src=pr&el=footer). Last update [f061606...bbacc86](https://codecov.io/gh/huggingface/transformers/pull/2160?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Thank you for the PR! I edited your post to add the guideline for adding a new model; we'll check the boxes as we go. I'll have a look at the code and come back to you quickly!", "> # Typical workflow for including a model\r\n> * [ ] add your models and tokenizer to `pipeline.py`\r\n\r\n@rlouf Sorry, I didn't find the `pipeline.py` file.", "@sshleifer Thanks for the comments! We will merge them into the code. @addf400 ", "Let's restart with a new pull request @addf400 @donglixp ", "Is anyone still working on this? @addf400 @donglixp @JetRunner \r\nalso @thomwolf from #1530", "I'm also looking forward to applying the UniLM model via Huggingface Transformers!\r\n@donglixp @JetRunner @thomwolf ", "It seems that this pull request has lasted for a year but still not finished? Is someone still working on it? ", "Has this PR for UniLM model been added to Huggingface Transformers?\r\n@donglixp @JetRunner @thomwolf @sshleifer", "Hey @ontocord , I think it the \"minilm\" model should work out-of-the-box:\r\n\r\nhttps://github.com/huggingface/transformers/issues/5777\r\n\r\nNot sure if you're looking for this model :thinking: \r\n\r\nI haven't tried it yet, but the recent Microsoft papers (on language modeling) are looking really promising!", "Thanks @stefan-it. I don't think MiniLM and UniLM are the same thing, altough it all falls under one project. The MS papers are promising!", "I'm also looking forward to applying the unilm model via Huggingface Transformers!", "2022 year, still not merged the unilm model into the master branch.", "I'm still looking forward to applying the unilm model via Huggingface Transformers! 👻👻\r\n\r\n", "I'm still looking forward to applying the unilm model via Huggingface Transformers too!" ]

[ { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null }, { "id": 1845609017, "node_id": "MDU6TGFiZWwxODQ1NjA5MDE3", "url": "https://api.github.com/repos/huggingface/transformers/labels/seq2seq", "name": "seq2seq", "color": "fef2c0", "default": false, "description": "" } ]

[]

[ "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "I'm struggling with this also :( did you make any progress diagnosing the problem?", "No, I did not...I decided to go with a different model", "hello.. guys.. any answers to this? Why is there such a low score? I looked at the summaries, and they seem to be good, but i have no comparison benchmark. However, the rouge scores are much lower than paper. how so? ", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "active it", "@AI678 The BertSum is unfortunately not maintained anymore. If you're looking to do summarization, please check out the [seq2seq](https://github.com/huggingface/transformers/tree/master/examples/seq2seq) scripts." ]

[]

[]

[ "Hi, this is for the initialization. We don't have any scripts that show how to pretrain GPT-2 (therefore no need for initialization), only scripts to fine-tune it from a checkpoint.", "thanks for your reply\r\n" ]

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

[]

[ "Hi, they are named as such because that's a clean way to make sure the model on the S3 is the same as the model in the cache. The name is created from the `etag` of the file hosted on the S3.\r\n\r\nIf you want to save it with a given name, you can save it as such:\r\n\r\n```py\r\nfrom transformers import BertModel\r\n\r\nmodel = BertModel.from_pretrained(\"bert-base-cased\")\r\nmodel.save_pretrained(\"cased_L-12_H-768_A-12\")\r\n```", "@LysandreJik, following up the question above, and your answer, I ran this command first:\r\n\r\n```\r\nfrom transformers import RobertaModel\r\nmodel = RobertaModel.from_pretrained(\"roberta-large\")\r\nmodel.save_pretrained(\"./roberta-large-355M\")\r\n```\r\nI guess, we expect config.json, vocab, and all the other necessary files to be saved in `roberta-large-355M` directory.\r\n\r\nThen I ran:\r\n\r\n```\r\npython ./examples/run_glue.py --model_type roberta --model_name_or_path ./roberta-large-355M --task_name MRPC --do_train --do_eval --do_lower_case --data_dir $GLUE_DIR/$TASK_NAME --max_seq_length 128 --per_gpu_train_batch_size 32 --learning_rate 2e-5 --num_train_epochs 2.0 --output_dir ./results/mrpc/\r\n```\r\n\r\nand I am getting:\r\n\r\n```\r\nOSError: Model name './roberta-large-355M' was not found in tokenizers model name list (roberta-base, roberta-large, roberta-large-mnli, distilroberta-base, roberta-base-openai-detector, roberta-large-openai-detector). We assumed './roberta-large-355M' was a path or url to a directory containing vocabulary files named ['vocab.json', 'merges.txt'] but couldn't find such vocabulary files at this path or url\r\n```\r\nI checked the `roberta-large-355M` and there are only: `config.json` `pytorch_model.bin`, but files named ['vocab.json', 'merges.txt'] are missing.\r\n\r\nsame issue with the XLNET:\r\n```\r\n../workspace/transformers/xlnet_base# ls\r\nconfig.json pytorch_model.bin\r\n```\r\nWhat am I missing here? Why are all the files not downloaded properly?\r\n\r\nThanks.\r\n\r\n", "You also have to save the tokenizer into the same directory:\r\n```python\r\ntokenizer.save_pretrained(\"./roberta-large-355M\")\r\n```\r\n\r\nLet me know if this solves your issue.", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "OSError: Model name 'roberta-base' was not found in tokenizers model name list (roberta-base, roberta-large, roberta-large-mnli, distilroberta-base, roberta-base-openai-detector, roberta-large-openai-detector). We assumed 'roberta-base' was a path, a model identifier, or url to a directory containing vocabulary files named ['vocab.json', 'merges.txt'] but couldn't find such vocabulary files at this path or url.\r\nI got the error above even after saving the tokenizer, config, and model in the same directory", "the problem for me is , when i load the model turning wifi off or switch off internet connection it fail to run but when i turn internet connection it run again. how can i run it off line. \r\ni also set enviornment variable like this .\r\nimport os\r\nos.environ['HF_DATASETS_OFFLINE']='1'\r\nos.environ['TRANSFORMERS_OFFLINE']='1'\r\ngenerator = pipeline('text-generation', model='EleutherAI/gpt-neo-1.3B')\r\ngenerator(text, do_sample=True, min_length=5)\r\n\r\nresult\r\n \"Connection error, and we cannot find the requested files in the cached path.\"\r\nValueError: Connection error, and we cannot find the requested files in the cached path. Please try again or make sure your Internet connection is on.\r\n", "import os\r\nfrom transformers import pipeline\r\n#HF_DATASETS_OFFLINE = 1\r\n#TRANSFORMERS_OFFLINE = 1\r\n#os.environ[HF_DATASETS_OFFLINE = 1,TRANSFORMERS_OFFLINE = 1]\r\nos.environ[\"HF_DATASETS_OFFLINE\"] = \"1\"\r\nos.environ[\"TRANSFORMERS_OFFLINE\"] = \"1\"\r\ncache_dir='/Users/hossain/Desktop/gpt2/gpt-neo-1.3/model/'\r\ngenerator = pipeline('text-generation', model='EleutherAI/gpt-neo-1.3B')\r\n\r\ntext = 'i am fine. what about you?'\r\ngenerator(text, do_sample=True, min_length=5)\r\nresult: through an error\r\n\"Connection error, and we cannot find the requested files in the cached path.\"\r\nValueError: Connection error, and we cannot find the requested files in the cached path. Please try again or make sure your Internet connection is on.", "i have dig down into the sentence_transformers lib to see which folder contain the file after downloaded. And came up with this script to see where sentence_transformers keep its files.\r\n\r\n```python\r\nimport os\r\n\r\ntorch_home = os.path.expanduser(\r\n os.getenv(\"TORCH_HOME\",\r\n os.path.join(os.getenv(\"XDG_CACHE_HOME\",\r\n \"~/.cache\"), 'torch')))\r\n\r\nprint(torch_home)\r\n\r\n```\r\n\r\ni hope it helps", "> i have dig down into the sentence_transformers lib to see which folder contain the file after downloaded. And came up with this script to see where sentence_transformers keep its files.\r\n> \r\n> ```python\r\n> import os\r\n> \r\n> torch_home = os.path.expanduser(\r\n> os.getenv(\"TORCH_HOME\",\r\n> os.path.join(os.getenv(\"XDG_CACHE_HOME\",\r\n> \"~/.cache\"), 'torch')))\r\n> \r\n> print(torch_home)\r\n> ```\r\n> \r\n> i hope it helps\r\n\r\nthanks. the code works on windows too" ]

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

[]

[ "Hello @shreydesai,\r\nYou should have a look at [run_squad_w_distillation.py](https://github.com/huggingface/transformers/blob/master/examples/distillation/run_squad_w_distillation.py) which is the script used in the experiment you are mentioning.\r\nVictor", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "Hello! Indeed this is a known issue with version 2.2.1. You can either revert to 2.2.0 or install from source (`pip install git+https://github.com/huggingface/transformers`) until we push a new version (2.2.2) which should happen before the end of the week.", "I confirm that reverting to 2.2.0 solves the problem", "Bumping to 2.2.2 (released today) should solve the problem too!", "I can confirm that 2.2.2 fixes the issue. This question can be closed.", "Thanks! This is a great library!", "It really is! Can you close this question? That way the overview of open issues is a lot clearer. Thanks.", "Hi, I still get the same issue with version `2.5.1` (installed from source).\r\nThe `<MASK>` token seems to be split into it's individual characters when an input string is encoded. \r\n\r\nI trained the `roberta` model from scratch on my own dataset as described in https://huggingface.co/blog/how-to-train .\r\nI ran the following lines to test my trained model on the masked token prediction task.\r\n```\r\nconfig = RobertaConfig.from_json_file(\"drive/My Drive/doerrberto-small-v1/config.json\")\r\nmodel = RobertaForMaskedLM(config)\r\nstate_dict = torch.load(\"drive/My Drive/doerrberto-small-v1/pytorch_model.bin\")\r\nmodel.load_state_dict(state_dict)\r\ntokenizer = RobertaTokenizer(\"drive/My Drive/doerrberto-small-v1/vocab.json\", \"drive/My Drive/doerrberto-small-v1/merges.txt\")\r\n\r\nfill_mask = pipeline(\r\n \"fill-mask\",\r\n model=model,\r\n tokenizer=tokenizer\r\n )\r\n\r\nresult = fill_mask(sentence)\r\n```\r\n\r\nThis was when I encountered the `ValueError: only one element tensors can be converted to Python scalars` error. I then confirmed that this error was generated due to incorrect encoding of `<MASK>` token.\r\n\r\nAny help will be appreciated. Thanks!", "@aksub99 `RobertaTokenizer`'s mask token is actually `<mask>` not `<MASK>`\r\n\r\nYou can also just use `tokenizer.mask_token`", "@julien-c Thanks for pointing that out, but I had used `tokenizer.mask_token` while testing. Sorry for the typo in my previous comment.\r\nThat still gave me the same errors.\r\nThis is my complete testing code snippet and it's output.\r\nCode:\r\n```\r\nimport torch\r\nfrom transformers import RobertaConfig, RobertaForMaskedLM, pipeline, RobertaTokenizer\r\n\r\nconfig = RobertaConfig.from_json_file(\"drive/My Drive/doerrberto-small-v1/config.json\")\r\nmodel = RobertaForMaskedLM(config)\r\nstate_dict = torch.load(\"drive/My Drive/doerrberto-small-v1/pytorch_model.bin\")\r\nmodel.load_state_dict(state_dict)\r\ntokenizer = RobertaTokenizer(\"drive/My Drive/doerrberto-small-v1/vocab.json\", \"drive/My Drive/doerrberto-small-v1/merges.txt\")\r\n\r\nfill_mask = pipeline(\r\n \"fill-mask\",\r\n model=model,\r\n tokenizer=tokenizer\r\n )\r\n\r\n\r\nsentence = \"I {} you\".format(tokenizer.mask_token)\r\nprint(sentence)\r\n\r\ntoken_ids = tokenizer.encode(sentence, return_tensors='pt')\r\nprint(token_ids.squeeze())\r\n\r\nprint(tokenizer.mask_token_id)\r\n\r\n```\r\nOutput:\r\n```\r\nI <mask> you\r\ntensor([ 0, 387, 225, 32, 81, 3229, 34, 377, 2])\r\n4\r\n```\r\nClearly, the `<mask>` is being split into it's individual characters.", "Does the `doerrberto-small-v1` vocabulary file contain a mask token? Can you do `tokenizer.encode(tokenizer.mask_token)`, and does it return the `tokenizer.mask_token_id` in-between model-specific tokens?", "@LysandreJik Yes, the `doerrberto-small-v1` vocabulary file does contain a mask token and is associated with an ID of 4. \r\n`tokenizer.encode(tokenizer.mask_token)` gives out `[0, 225, 32, 81, 3229, 34, 2]` which means that the mask token is again being split up.\r\nSorry, could you explain what you mean by \"in-between model-specific tokens\"? ", "Seems broken in GPT2 ?\r\n\r\n\r\n```python\r\ntok = transformers.AutoTokenizer.from_pretrained(\"gpt2\")\r\ntok.cls_token = \"<|cls|>\"\r\nsample = \"My name is Barrack Obama<|cls|>I like pizza\"\r\nprint(tok.tokenize(sample))\r\n```\r\n>`['My', 'Ġname', 'Ġis', 'ĠBarr', 'ack', 'ĠObama', '<', '|', 'cl', 's', '|', '>', 'I', 'Ġlike', 'Ġpizza']`\r\n", "It does show up in the special tokens:\r\n```python\r\n>>> print(tok)\r\nPreTrainedTokenizerFast(name_or_path='gpt2', vocab_size=50257, model_max_len=1024, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'bos_token': '<|endoftext|>', 'eos_token': '<|endoftext|>', 'unk_token': '<|endoftext|>', 'cls_token': '<|cls|>'})\r\n```", "@LysandreJik maybe", "Using `<|endoftext|>` does work however. It's just when you add new special tokens that the tokenizer doesn't use them.\r\n\r\n```python\r\n>>> tok.tokenize(\"An attempt with eot<|endoftext|>Will it work\")\r\n['An', 'Ġattempt', 'Ġwith', 'Ġe', 'ot', '<|endoftext|>', 'Will', 'Ġit', 'Ġwork']\r\n```" ]

[]

[]

[ "Hello! Could you please tell me which model you are trying to download? I've just tried the following command and it succeeded without any issues:\r\n```py\r\nAlbertForQuestionAnswering.from_pretrained(\"albert-base-v2\", force_download=True)\r\n```\r\n\r\nI put the `force_download` flag to True to make sure I was downloading the files from the S3. Is there any way you could try this on your side?", "In my environment (Python 3.6.9, OS Ubuntu, Transformers 2.2.1 (installed from _PyPi_), PyTorch 1.3.1 and TensorFlow 2.0), **I'm not able to reproduce your bug**, so I'm able to download and use any ALBERT model I want. I've tried the same code line that in your case generates the error, e.g.\r\n```\r\n> from transformers import AlbertForQuestionAnswering\r\n> model = AlbertForQuestionAnswering.from_pretrained(X)\r\n``` \r\nwhere X is one of [_'albert-base-v1', 'albert-large-v1', 'albert-xlarge-v1', 'albert-xxlarge-v1', 'albert-base-v2', 'albert-large-v2', albert-xlarge-v2', 'albert-xxlarge-v2'_]\r\n\r\nYou can specify `force_download=True` when you're loading a specific version of AlBERT model, e.g.\r\n```\r\n> from transformers import AlbertForQuestionAnswering\r\n> model = AlbertForQuestionAnswering.from_pretrained('albert-base-v1', force_download=True)\r\n```\r\n\r\nN.B: at the moment, there is a known bug when using v2 AlBERT models, as said when you import this version in Transformers:\r\n> There is currently an upstream reproducibility issue with ALBERT v2 models. Please see https://github.com/google-research/google-research/issues/119 for more information.\r\n\r\n> ## Questions & Help\r\n> \r\n> \r\n> \r\n> Hi! There is some problem while downloading any of the pre-trained AlBERT models, however, there weren't any problems a few days ago. Could you please tell me where can I download the AlBERT TensorFlow checkpoints (albert_model.ckpt) for running convert_albert_original_tf_checkpoint_to_pytorch.py script? Unfortunately, I wasn't able to find any for resolving this as in #2110.\r\n> I'd really appreciate any help in resolving the issue. Thanks a bunch in advance!", "I've tried with all of the models with and without `force_download=True`\r\n\r\n\r\n\r\nUnfortunately, I have this bug now, however, it was OK yesterday. Besides, have the same issue using Jupyter notebook after restarting the kernel. Before that worked as expected.\r\n\r\nThanks for your concern and fast reply!", "Do you get the same errors with other models, like BERT?", "I've just tried again using completely new basic script and it worked, don't know what is that. I just thought it's the same as in another issue. \r\nBut anyway, thanks a lot, guys!", "Glad you could fix it." ]

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

[]

[ "Could you please post the full stack trace as well as the part of the code you use for fine-tuning?", "See here: https://gist.github.com/ohmeow/f2cc6ea0a9d0e4a5fa227942edcfa723\n\nI think it has something to do with how I'm preparing the target tokens but\nI'm not sure what the appropriate fix is. Looked at the BertSum source\ncode on github but it was confusing. Either way, the shape of the\ndecoder_ids is (batch size, max_seq_len) ... but the model chops off the\nlast column before passing the ids off to the decoder. My gut feeling is\nthat this is to account for the need to shift the ids right by 1 for the\ngold labels but not sure ... and that means the input should be\n(batch_size, max_seq_len+1).\n\nAny thoughts on what I should do or what I'm missing?\n\nThanks\n\nOn Thu, Dec 12, 2019 at 3:15 AM Rémi Louf <[email protected]> wrote:\n\n> Could you please post the full stack trace as well as the part of the code\n> you use for fine-tuning?\n>\n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub\n> <https://github.com/huggingface/transformers/issues/2153?email_source=notifications&email_token=AAADNMBPVGGALZOJ3SCV543QYIMN3A5CNFSM4JZZWIK2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEGWKWLY#issuecomment-564964143>,\n> or unsubscribe\n> <https://github.com/notifications/unsubscribe-auth/AAADNMCRVCPLMOPCLI2L4S3QYIMN3ANCNFSM4JZZWIKQ>\n> .\n>\n", "I think I may have solved the `decoder_input_ids` issue with the fix to my code:\r\n\r\n```\r\ndef fit_to_block_size(sequence, block_size, pad_token_id, sep_token_id, is_summary:bool=False):\r\n \"\"\" Adapt the source and target sequences' lengths to the block size.\r\n If the sequence is shorter than the block size we pad it with -1 ids\r\n which correspond to padding tokens.\r\n \"\"\"\r\n \r\n if len(sequence) > block_size:\r\n if (is_summary):\r\n sequence = sequence[:block_size]+ [symbols['EOS']]\r\n else:\r\n # ensure inclusion of whole sentences if possible\r\n sent_sep_idxs = [ idx for idx, t in enumerate(sequence) if t == sep_token_id and idx < block_size ]\r\n \r\n last_sent_sep_idx = min(max(sent_sep_idxs)+1 if (len(sent_sep_idxs) > 0) else block_size, block_size)\r\n \r\n sequence = sequence[:last_sent_sep_idx]\r\n \r\n if len(sequence) < block_size:\r\n sequence.extend([pad_token_id] * (block_size - len(sequence)))\r\n \r\n if (is_summary): \r\n sequence += [pad_token_id]\r\n \r\n return sequence\r\n```\r\nHowever, I'm now running into an error when the \"context\" attention is calculated in the `TransformerDecoderLayer` ...\r\n\r\n```\r\n~/development/_training/ml/nlp-playground/tritonlyticsai/text/modeling_bertabs.py in forward(self, key, value, query, mask, layer_cache, type, predefined_graph_1)\r\n 601 \r\n 602 if mask is not None:\r\n--> 603 mask = mask.unsqueeze(1).expand_as(scores)\r\n 604 scores = scores.masked_fill(mask, -1e18)\r\n 605 \r\n\r\nRuntimeError: The expanded size of the tensor (1) must match the existing size (512) at non-singleton dimension 3. Target sizes: [512, 8, 8, 1]. Tensor sizes: [8, 1, 512, 512]\r\n```\r\nThe passed in mask is built by the model code based on the dimensions of the source and target input ids ... which look right to me.", "@ohmeow Have you been able to fine-tune the BertAbs on your dataset? I would appreciate if your can share you experience.", "This is still a work in progress ... but the below should help you get\nstarted on fine-tuning the pretrained model.\n\nLook here: https://gist.github.com/ohmeow/7aa294e2959c1315fe7dfdf8091f2d87\n\nYou'll notice that I also copied a few of the HF .py files into my own\npackage (ohmeow.text). I did this two be able to step through the code,\ntroubleshoot, and also because a modification has to be made to\nmodeling_bertabs.py.\n\n#pdb.set_trace()\nencoder_hidden_states = encoder_output #encoder_output[0] --WTG--\ndec_state = self.decoder.init_decoder_state(\nencoder_input_ids, encoder_hidden_states\n)\ndecoder_outputs, _ = self.decoder(\ndecoder_input_ids[:, :-1], encoder_hidden_states, dec_state\n)\n#return decoder_outputs #--WTG--\nreturn self.generator(decoder_outputs)\n\n\nthe commented out sections are what was originally there in the HF code.\n\nOn Mon, Dec 30, 2019 at 6:17 PM Ehsan Hemmati <[email protected]>\nwrote:\n\n> @ohmeow <https://github.com/ohmeow> Have you been able to fine-tune the\n> BertAbs on your dataset? I would appreciate if your can share you\n> experience.\n>\n> —\n> You are receiving this because you were mentioned.\n> Reply to this email directly, view it on GitHub\n> <https://github.com/huggingface/transformers/issues/2153?email_source=notifications&email_token=AAADNMD6OHZO67VQURDKBJLQ3KTSRA5CNFSM4JZZWIK2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEH3TI6Q#issuecomment-569848954>,\n> or unsubscribe\n> <https://github.com/notifications/unsubscribe-auth/AAADNMGJ6MSCUTEYGHNB4BTQ3KTSRANCNFSM4JZZWIKQ>\n> .\n>\n", "@ohmeow Thanks for sharing this.\r\nJust what is the HF files you mentioned?", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "Hi, could you provide an example that was sped up by replacing that variable? When tokenizing 55k tokens 10 times without creating a variable for all_special_tokens I get the result in 3.88s whereas when creating a variable I get the result in 3.87s. This doesn't seem like such a big difference!", "In my case, I tested it on the feature_conversion for SQuAD. \r\n\r\nI measure the time to convert a subset of the SQuAD dataset to features:\r\n\r\n```\r\n start_time = time.time()\r\n\r\n tok_times = []\r\n for i in range(10):\r\n start_time = time.time()\r\n\r\n convert_examples_to_features(\r\n examples=eval_examples,\r\n tokenizer=tokenizer,\r\n max_seq_length=FLAGS.max_seq_length,\r\n doc_stride=FLAGS.doc_stride,\r\n max_query_length=FLAGS.max_query_length,\r\n is_training=False,\r\n output_fn=append_feature)\r\n\r\n delta_time = time.time() - start_time\r\n print('Run {}: Time for tokenization: {}'.format(i, delta_time))\r\n tok_times.append(delta_time)\r\n\r\n print('Avg time for tokenization: {}'.format(np.mean(tok_times)))\r\n```\r\n\r\nThe original implementation yields (in seconds):\r\n```\r\nRun 0: Time for tokenization: 1.8680036067962646\r\nRun 1: Time for tokenization: 1.8013951778411865\r\nRun 2: Time for tokenization: 1.7933814525604248\r\nRun 3: Time for tokenization: 1.7968308925628662\r\nRun 4: Time for tokenization: 1.8006742000579834\r\nRun 5: Time for tokenization: 1.7927491664886475\r\nRun 6: Time for tokenization: 1.8060340881347656\r\nRun 7: Time for tokenization: 1.7863578796386719\r\nRun 8: Time for tokenization: 1.807504415512085\r\nRun 9: Time for tokenization: 1.7879209518432617\r\nAvg time for tokenization: 1.8040851831436158\r\n```\r\n\r\nWhen initializing a variable instead and referencing it in tokenization:\r\n```\r\nRun 0: Time for tokenization: 0.7765586376190186\r\nRun 1: Time for tokenization: 0.6800308227539062\r\nRun 2: Time for tokenization: 0.6858618259429932\r\nRun 3: Time for tokenization: 0.6877231597900391\r\nRun 4: Time for tokenization: 0.6820297241210938\r\nRun 5: Time for tokenization: 0.6838114261627197\r\nRun 6: Time for tokenization: 0.6909258365631104\r\nRun 7: Time for tokenization: 0.6799609661102295\r\nRun 8: Time for tokenization: 0.6868128776550293\r\nRun 9: Time for tokenization: 0.679542064666748\r\nAvg time for tokenization: 0.6933257341384887\r\n```\r\n\r\nI basically just initialize a new variable in init of `tokenization_utils.py`:\r\n `self.all_special_tokens_init = self.all_special_tokens`\r\nAnd then I reference this variable in `split_on_tokens()` instead of the call to the property function `all_special_tokens`.", "Indeed, I do get a massive speedup when initializing a variable and using `squad_convert_examples_to_features`. Thank you for letting us know! I'll update this later today.", "Should have been fixed with f24a228", "No problem :) Thanks for the fix. I will close this then." ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "Ok, I've checked the errors in CI. Those are linked to the fact that in older python version, PosixPath is not converted automatically to String and `os.path.isdir/isfile` crash because it expects a string or int. So my patch works perfectly in latest version of python (like 3.7) but not older (like 3.5) which is quite ugly.\r\nSolutions are:\r\n- anywhere there is `isdir/isfile`, force `str(path)` in call which is ugly but will work except if you want to manage other types than strings in path\r\n- officially ask people to convert their path to strings when calling `from_pretrained` which is shame because it shows the impossibility to be backward compatible completely but it won't introduce `str()` everywhere in the code.\r\n\r\nWDYT?\r\n", "Hey @mandubian, thanks for offering that fix. I think that in the next release we'll remove this warning about albert models v2, which will solve the problem with PosixPaths.", "@LysandreJik perfect! It will still fail with python 3.5 on `isdir/isfile` but can we do anything for that? I'm not sure... that's the history of Python ;)", "The line was removed for version 2.3.0 so there's no need for that anymore. Thanks @mandubian :)." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2148?src=pr&el=h1) Report\n> Merging [#2148](https://codecov.io/gh/huggingface/transformers/pull/2148?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/030faccb8d45be9bdd2b4b80ff26f36dc41f622a?src=pr&el=desc) will **decrease** coverage by `0.02%`.\n> The diff coverage is `30.76%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/2148?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #2148 +/- ##\n==========================================\n- Coverage 80.07% 80.05% -0.03% \n==========================================\n Files 112 112 \n Lines 16866 16868 +2 \n==========================================\n- Hits 13505 13503 -2 \n- Misses 3361 3365 +4\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/2148?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/2148/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl91dGlscy5weQ==) | `90.22% <30.76%> (-0.8%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/2148?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/2148?src=pr&el=footer). Last update [030facc...3d57c51](https://codecov.io/gh/huggingface/transformers/pull/2148?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Great! Nice catch @LysandreJik " ]

[]

[]

[ "You can find the s3 URL of models here for distilbert: https://github.com/huggingface/transformers/blob/master/transformers/configuration_distilbert.py\r\n\r\nIf you build the docker on your machine, first download model files on your machine.\r\nThen just add those files to your container through Dockerfile.\r\n\r\nIf you want your Docker build to download from s3, you can install `aws-cli` in Dockerfile and run `aws s3 cli`. But it will make it slower.\r\n\r\nNaturally a model in your docker will make it a bit fatter.", "I think the path has changed slightly. I found the file in \"src\" folder under master. https://github.com/huggingface/transformers/blob/master/src/transformers/configuration_distilbert.py ", "About downloading model files, in configuration_distilbert.py, I only found https://s3.amazonaws.com/models.huggingface.co/bert/distilbert-base-uncased-distilled-squad-config.json file path. It just gives config. It is not a weights file / pickle file. Please suggest the path of files which I can download and make part of the local folder. Thanks.", "Links to pre-trained models are available in the beginning of each `modeling_xxx.py` file, e.g. for [BERT](https://github.com/huggingface/transformers/blob/master/src/transformers/modeling_bert.py#L34-L56). Put this with a configuration in your folder and you can load them locally.\r\n\r\nYou could also use the `save_pretrained` method to automatically create a folder that can be used with `from_pretrained`." ]