Datasets:

---

DanielPFlorian

/

Transformers-Github-Issues

like 0

[]

[]

[ "Myle Ott from Facebook commented on the Twitter thread (https://twitter.com/myleott/status/1175750596630056961) confirming that there is an extra separator being used, so there should be details I did not understand well.\r\n\r\nI will revisit this issue when I understand it better.", "The `sep_token_extra` param is deprecated, as we have simpler ways to do this now (thanks to @LysandreJik). Closing this for now, feel free to re-open if needed." ]

[]

[]

[]

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

[]

[ "No short-term plan to implement this ourselves, but we'd welcome a PR (especially one involving the original authors for validation).", "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, thank you for the bug report. Indeed, this does seem problematic. I'm looking into it.", "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 } ]

[]

[ "Same as https://github.com/google-research/bert/issues/35", "And https://github.com/huggingface/transformers/issues/37", "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", "GPT2 can be used for rescoring https://arxiv.org/pdf/1910.11450.pdf", "[I tested](https://github.com/simonepri/lm-scorer/blob/master/tests/models/test_gpt2.py#L52-L239) GPT2 on different english error types (I used the one defined in the [ERRANT framework](https://www.aclweb.org/anthology/P17-1074/)) and it seems that is able to give a lower probability to the wrong version of a sentence (At least for simple examples)." ]

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

[]

[ "I am facing the same issue as there is no proper format available for defining the train and test dataset. \r\nAs usual, I use .csv file in a format of columns with (UID, Text, and Labels). But according to the Wiki.txt its more of arbitrary format.\r\n\r\nAny help would be appreciated.", "I'm having the same issue. I think it's counting the total length of the tokenized corpus not only the tokenized document length. I tried to run the wiki raw files as mentioned in the read me and still get this warning of total tokenized corpus length.\r\n\r\nI tried to the following formats with no success:\r\n1. sentence per line with a blank line in between docs\r\n2. document per line with a blank line in between docs\r\n\r\nUpdate:\r\nAfter looking at the code again it looks like even though this warning is showing the sequence length being longer than 512 it is still chunking the corpus into 512 tokens and training it that way. This raises the question of whether it is problematic to just separate the corpus based on token length alone especially that BERT for example is training on predicting next sentence. What happens to the probably recurring case of the data being chunked mid-way 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" ]

[]

[ { "login": "julien-c", "id": 326577, "node_id": "MDQ6VXNlcjMyNjU3Nw==", "avatar_url": "https://avatars.githubusercontent.com/u/326577?v=4", "gravatar_id": "", "url": "https://api.github.com/users/julien-c", "html_url": "https://github.com/julien-c", "followers_url": "https://api.github.com/users/julien-c/followers", "following_url": "https://api.github.com/users/julien-c/following{/other_user}", "gists_url": "https://api.github.com/users/julien-c/gists{/gist_id}", "starred_url": "https://api.github.com/users/julien-c/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/julien-c/subscriptions", "organizations_url": "https://api.github.com/users/julien-c/orgs", "repos_url": "https://api.github.com/users/julien-c/repos", "events_url": "https://api.github.com/users/julien-c/events{/privacy}", "received_events_url": "https://api.github.com/users/julien-c/received_events", "type": "User", "site_admin": false } ]

[ "I think you may be right and we've been meaning to fix this. cf recent discussion @LysandreJik @VictorSanh ", "thanks. Adding LM fine-tuning to fast-bert. Have added a workaround for now :)", "Also see https://github.com/pytorch/fairseq/issues/1187" ]

[ { "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", "In my environment, **it works as expected**!\r\n\r\n_Environment_:\r\n\r\n- **Python**: 3.6.9\r\n- **O.S.** : Linux-4.15.0-70-generic-x86_64-with-debian-buster-sid\r\n- **Transformers**: 2.1.1 (installed from source with `pip install git+https://github.com/huggingface/transformers.git`)\r\n- **Torch**: 1.3.1\r\n\r\n_Example code_:\r\n```\r\n>>> import transformers\r\n>>> from transformers import BertForQuestionAnswering\r\n>>> model = BertForQuestionAnswering.from_pretrained('bert-large-uncased-whole-word-masking-finetuned-squad')\r\n100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 341/341 [00:00<00:00, 152821.63B/s]\r\n100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1340675298/1340675298 [02:04<00:00, 10793596.06B/s]\r\n>>> ...\r\n```\r\n\r\nThe same correct behavior occurs with TensorFlow 2.0:\r\n```\r\n>>> import transformers\r\n>>> from transformers import TFBertForQuestionAnswering\r\n>>> model = TFBertForQuestionAnswering.from_pretrained('bert-large-uncased-whole-word-masking-finetuned-squad')\r\n████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1340675298/1340675298 [02:30<00:00, 8890808.45B/s]\r\n>>> ...\r\n```\r\n\r\nNow, you can close this issue!\r\n\r\n> ## Bug\r\n> Model I am using (Bert, XLNet....): Bert\r\n> \r\n> Language I am using the model on (English, Chinese....): English\r\n> \r\n> The problem arise when using:\r\n> \r\n> * [ ] the official example scripts: (give details): BertForQuestionAnswering\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): SQuaD\r\n> * [ ] 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. from pytorch_transformers import BertForQuestionAnswering\r\n> 2. model = BertForQuestionAnswering.from_pretrained('bert-large-uncased-whole-word-masking-finetuned-squad')\r\n> \r\n> ## Expected behavior\r\n> ## Environment\r\n> * OS:\r\n> * Python version:\r\n> * PyTorch version:\r\n> * PyTorch Transformers version (or branch):\r\n> * Using GPU ?\r\n> * Distributed of parallel setup ?\r\n> * Any other relevant information:\r\n> \r\n> ## Additional context", "This is usually because of \r\n- a network error or\r\n- not enough space on the disk\r\n\r\nwhile downloading the file. To make sure it isn't the first, you can try running the `from_pretrained` method with the `force_download` option set to `True`:\r\n\r\n```py\r\nmodel = BertForQuestionAnswering.from_pretrained('bert-large-uncased-whole-word-masking-finetuned-squad', force_download=True)\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 } ]

[]

[ "From what I can tell, Radam makes automatic warmup and LARS is good but requires more calculations per batch. Before implementing it here it's worth do testing to tell if it's a good idea.", "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", "any update?", "For anyone interested in testing, I've created a fork that uses Radam+LARS+LookAhead, https://github.com/i404788/transformers", "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/1301?src=pr&el=h1) Report\n> Merging [#1301](https://codecov.io/gh/huggingface/transformers/pull/1301?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/2dc8cb87341223e86220516951bb4ad84f880b4a?src=pr&el=desc) will **increase** coverage by `0.22%`.\n> The diff coverage is `96.35%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1301?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1301 +/- ##\n==========================================\n+ Coverage 84.69% 84.91% +0.22% \n==========================================\n Files 84 85 +1 \n Lines 12596 12840 +244 \n==========================================\n+ Hits 10668 10903 +235 \n- Misses 1928 1937 +9\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1301?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tests/modeling\\_bert\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1301/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX2JlcnRfdGVzdC5weQ==) | `97.47% <100%> (+1.09%)` | :arrow_up: |\n| [transformers/configuration\\_rbert.py](https://codecov.io/gh/huggingface/transformers/pull/1301/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2NvbmZpZ3VyYXRpb25fcmJlcnQucHk=) | `100% <100%> (ø)` | |\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/1301/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `74.54% <92.59%> (+3.32%)` | :arrow_up: |\n| [transformers/tests/modeling\\_roberta\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1301/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3JvYmVydGFfdGVzdC5weQ==) | `85% <93.67%> (+5.49%)` | :arrow_up: |\n| [transformers/modeling\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/1301/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2JlcnQucHk=) | `88.92% <96.15%> (+0.75%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1301?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/1301?src=pr&el=footer). Last update [2dc8cb8...4bcfa63](https://codecov.io/gh/huggingface/transformers/pull/1301?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Ok, I went through the PR.\r\n\r\nThis is a very nice work @RichJackson!\r\n\r\nOne thing we should simplify though is to not have a separate configuration for RBERT and roberta. I will update a bit the configuration classes so we can safely add new parameters in them and have them initialized to defaults values when loading from pretrained config.\r\n\r\nLet me do that now in this PR.", "Actually I can't push on your PR so I'll create a new one to update that." ]

[]

[]

[ "you should set padding and all the labels to be ignored to -1. In Bert/XLNet training, we usually only use 15% of the tokens as labels." ]

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

[]

[ "Give us a little more details about your `(latency, compute)` constraints.", "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/pytorch-transformers/pull/1298?src=pr&el=h1) Report\n> Merging [#1298](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1298?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/pytorch-transformers/commit/9f995b99d4c4067662c3bd4f1274315c0839deeb?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1298?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1298 +/- ##\n=======================================\n Coverage 80.77% 80.77% \n=======================================\n Files 57 57 \n Lines 8092 8092 \n=======================================\n Hits 6536 6536 \n Misses 1556 1556\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1298?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [pytorch\\_transformers/tokenization\\_xlnet.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1298/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdG9rZW5pemF0aW9uX3hsbmV0LnB5) | `89.18% <ø> (ø)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1298?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/pytorch-transformers/pull/1298?src=pr&el=footer). Last update [9f995b9...51decd5](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1298?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Thanks!" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1297?src=pr&el=h1) Report\n> Merging [#1297](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1297?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/pytorch-transformers/commit/0d1dad6d5323cf627cb8d7ddd428856ab8475f6b?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1297?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1297 +/- ##\n=======================================\n Coverage 80.77% 80.77% \n=======================================\n Files 57 57 \n Lines 8092 8092 \n=======================================\n Hits 6536 6536 \n Misses 1556 1556\n```\n\n\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1297?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/pytorch-transformers/pull/1297?src=pr&el=footer). Last update [0d1dad6...2a11412](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1297?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Hi @rajarsheem ! Thank you for your PR. With the example scripts, we are really reaching for **simple scripts that showcase how the library works** and how it interacts with different elements of the Pytorch codebase (ex: distributed learning, gradient clipping, ...).\r\n\r\nUsing a text file as input may be useful in some cases, however, I don't feel like it really gives a deeper understanding of the library, as it is just a different way to obtain a context string. I don't think it would be particularly worth it in terms of the added complexity/deeper understanding of the lib ratio.\r\n\r\nPlease don't let that discourage you from opening other PRs." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1296?src=pr&el=h1) Report\n> Merging [#1296](https://codecov.io/gh/huggingface/transformers/pull/1296?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/391db836ab7ed2ca61c51a7cf1b135b6ab92be58?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1296?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1296 +/- ##\n=======================================\n Coverage 84.72% 84.72% \n=======================================\n Files 84 84 \n Lines 12591 12591 \n=======================================\n Hits 10668 10668 \n Misses 1923 1923\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1296?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1296/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl91dGlscy5weQ==) | `91.48% <ø> (ø)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1296?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/1296?src=pr&el=footer). Last update [391db83...a951585](https://codecov.io/gh/huggingface/transformers/pull/1296?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Ok, great, thanks @danai-antoniou!" ]

[]

[]

[ "- \"token type embeddings\" are the BERT paper's segment embeddings \r\n- embeddings are inside the pretrained weights", "Ah that makes sense. So there are no \"separate\" word2vec-style pretrained embedding models for the different types of embeddings which one could load with `nn.Embedding().from_pretrained`. Rather, they are loaded in a bunch as a set of pretrained weights. Theoretically, though, one could extract the weights for each embedding, and extract the vocab from the tokenizer, and create a simple lookup (`token\\tvector`)?\r\n\r\nThanks for the reply and your work.", "Sure you could, but I suspect it wouldn’t work too well.\r\n\r\nYou could say that a large language model’s hidden states are the new way to do word/sentence embeddings (see Sebastian Ruder’s imageNet moment).", "Apologies if this is taking too much of your time, but I have a follow-up question. Why wouldn't it work too well? I understand that they are not typical word2vec word representations, since they have been trained together with the whole language model, but why would extracting the embeddings and using them in another task not work well? In other words, what makes the token embeddings of BERT fundamentally different from a typical word2vec model?", "I think you'll find this repo (and associated EMNLP 2019 paper) by @nriemers interesting:\r\n\r\nhttps://github.com/UKPLab/sentence-transformers (built on top of `transformers`)", "> * \"token type embeddings\" are the BERT paper's segment embeddings\r\n> * embeddings are inside the pretrained weights\r\n\r\nhi, could you tell where the code about BertEmbedding loaded with the pre-trained weights is?" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1294?src=pr&el=h1) Report\n> Merging [#1294](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1294?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/pytorch-transformers/commit/0d1dad6d5323cf627cb8d7ddd428856ab8475f6b?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1294?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1294 +/- ##\n=======================================\n Coverage 80.77% 80.77% \n=======================================\n Files 57 57 \n Lines 8092 8092 \n=======================================\n Hits 6536 6536 \n Misses 1556 1556\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1294?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [pytorch\\_transformers/configuration\\_openai.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1294/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvY29uZmlndXJhdGlvbl9vcGVuYWkucHk=) | `89.13% <ø> (ø)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1294?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/pytorch-transformers/pull/1294?src=pr&el=footer). Last update [0d1dad6...119610b](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1294?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Indeed, thanks Sam" ]

[]

[]

[ "I found maybe the current code is not consistent with the pip package pytorch_transformers, so when use the pip package it does't work, but when just run the code without the pip package, it can work, but you need change some path to make the code work correctly!", "Hi, I believe this was fixed with @VictorSanh's commit ae50ad9" ]

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

[]

[ "@snaik2016 I ran into the same issue and had to parallelize my code to make it faster. Also getting rid of the while loop and list splicing in the TextDataset class with a for loop made it much quicker.", "Please check #1830 . I made some tuning on a training part. But I guess it'll still take many days for 1.8M rows dataset (in fact, talking about the token count rather than rows count is more meaningful) .", "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 } ]

[]

[ "Which model did you use？", "> Which model did you use？\r\n\r\nxlnet", "Hi! Could you show the inputs you use to trace your model?", "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", "I also have the same problem...." ]

[ { "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/pytorch-transformers/pull/1289?src=pr&el=h1) Report\n> Merging [#1289](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1289?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/pytorch-transformers/commit/0d1dad6d5323cf627cb8d7ddd428856ab8475f6b?src=pr&el=desc) will **increase** coverage by `0.22%`.\n> The diff coverage is `39.39%`.\n\n[](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1289?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1289 +/- ##\n==========================================\n+ Coverage 80.77% 80.99% +0.22% \n==========================================\n Files 57 57 \n Lines 8092 8072 -20 \n==========================================\n+ Hits 6536 6538 +2 \n+ Misses 1556 1534 -22\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1289?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [pytorch\\_transformers/configuration\\_bert.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1289/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvY29uZmlndXJhdGlvbl9iZXJ0LnB5) | `88.57% <100%> (+1.47%)` | :arrow_up: |\n| [pytorch\\_transformers/modeling\\_bert.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1289/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvbW9kZWxpbmdfYmVydC5weQ==) | `85.17% <31.03%> (-3.16%)` | :arrow_down: |\n| [pytorch\\_transformers/modeling\\_xlnet.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1289/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvbW9kZWxpbmdfeGxuZXQucHk=) | `77.42% <0%> (+2.89%)` | :arrow_up: |\n| [pytorch\\_transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1289/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvbW9kZWxpbmdfcm9iZXJ0YS5weQ==) | `75.22% <0%> (+10.26%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1289?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/pytorch-transformers/pull/1289?src=pr&el=footer). Last update [0d1dad6...cd97cad](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1289?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "I think our goal here should be that you'd be able to extend the relevant classes without having modify to modify the lib's code.\r\n\r\nThoughts @LysandreJik @thomwolf?", "Yes, I agree with @julien-c.\r\n\r\nHaving adapters is a nice addition but we should have a mechanism that lets us extend the base code (of each model) instead of modifying it for each type of adapting mechanism.\r\n\r\nOne way we could do that is to have standardized pointers to a selection of relevant portions of the models (also asked by Matt Newman for some AllenNLP extension of the models).", "Would that be on the roadmap soon? I can resubmit my PR after there's a more generalized approach for extending model functionality.", "Thinking about it again, I think the approach you proposed is the right one if we want to integrate adapters in the library as a permanent option.\r\n\r\nBut I have two questions:\r\n- do we want to integrate adapters as a permanent option? My quick tests with adapters for the NAACL tutorial on transfer learning were giving mixed results. Do you have clear indications and benchmarks that they are useful in practice, @zphang?\r\n- if we do integrate them in the library we would want to have them for all the models.\r\n\r\nIf we don't want to integrate them as a permanent option then we would have to find a general way to extend the models easily so people can add stuff like this in a simple way. This is open for comments.", "In the meantime, I've actual moved to using an implementation that doesn't involve modifying the Transformers library. Here's an example (pardon the hacks):\r\n\r\nhttps://github.com/zphang/nlprunners/blob/9caaf94ea99102a9980012b934a4373dc4996108/nlpr/proj/adapters/modeling.py#L55-L135\r\n\r\nIt involves swapping out relevant portions of the model with modified/similar layers. Given the differences between the major transformer models, I think this would be the more sustainable and less intrusive (as long as the underlying transformer code doesn't change too often). \r\n\r\nPerformance-wise, my experience has been that adapters work roughly as advertised: consistently slightly less well than fine-tuning the whole model, but only the adapter layers + classifier head need to be tuned.", "I think the repository you are linking to is private Jason", "Oops! Here's a gist of the code: https://gist.github.com/zphang/8eb4717b6f74c82a8ca4637ae9236e21", "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", "Closing for now, we'll continue the work on this topic in the more general #2165 " ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1288?src=pr&el=h1) Report\n> Merging [#1288](https://codecov.io/gh/huggingface/transformers/pull/1288?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/2dc8cb87341223e86220516951bb4ad84f880b4a?src=pr&el=desc) will **decrease** coverage by `3.92%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1288?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1288 +/- ##\n==========================================\n- Coverage 84.69% 80.77% -3.93% \n==========================================\n Files 84 57 -27 \n Lines 12596 8092 -4504 \n==========================================\n- Hits 10668 6536 -4132 \n+ Misses 1928 1556 -372\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1288?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tests/conftest.py](https://codecov.io/gh/huggingface/transformers/pull/1288/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL2NvbmZ0ZXN0LnB5) | | |\n| [transformers/modeling\\_tf\\_transfo\\_xl\\_utilities.py](https://codecov.io/gh/huggingface/transformers/pull/1288/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3RyYW5zZm9feGxfdXRpbGl0aWVzLnB5) | | |\n| [transformers/data/processors/utils.py](https://codecov.io/gh/huggingface/transformers/pull/1288/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2RhdGEvcHJvY2Vzc29ycy91dGlscy5weQ==) | | |\n| [transformers/modeling\\_gpt2.py](https://codecov.io/gh/huggingface/transformers/pull/1288/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2dwdDIucHk=) | | |\n| [transformers/tests/modeling\\_tf\\_openai\\_gpt\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1288/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX29wZW5haV9ncHRfdGVzdC5weQ==) | | |\n| [transformers/modeling\\_tf\\_xlm.py](https://codecov.io/gh/huggingface/transformers/pull/1288/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3hsbS5weQ==) | | |\n| [transformers/optimization.py](https://codecov.io/gh/huggingface/transformers/pull/1288/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL29wdGltaXphdGlvbi5weQ==) | | |\n| [transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1288/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2ZpbGVfdXRpbHMucHk=) | | |\n| [transformers/tokenization\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/1288/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl9iZXJ0LnB5) | | |\n| [transformers/tests/tokenization\\_distilbert\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1288/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL3Rva2VuaXphdGlvbl9kaXN0aWxiZXJ0X3Rlc3QucHk=) | | |\n| ... and [131 more](https://codecov.io/gh/huggingface/transformers/pull/1288/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1288?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/1288?src=pr&el=footer). Last update [2dc8cb8...f7978f7](https://codecov.io/gh/huggingface/transformers/pull/1288?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Thanks, updated a bit to use `format`" ]

[]

[]

[ "Should be fixed on master and the new release (2.0)" ]

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

[]

[ "Yes, why not, do you want to submit a PR for that?", "> Yes, why not, do you want to submit a PR for that?\r\n\r\nThanks~\r\nBy the way, is there any code formatting requirement or a contribution docs for developers?\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", "seems like the first checkpoint's eval results does not have a prefix?" ]

[]

[]

[ "Also getting this effect when using the reproduction code on my system.", "It's not a bug. This is an artefact produced by BPE as explained here https://github.com/huggingface/pytorch-transformers/blob/d483cd8e469126bed081c59473bdf64ce74c8b36/pytorch_transformers/tokenization_gpt2.py#L106\r\n\r\nI think the solution is to process whitespaces after the tokeniser." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1284?src=pr&el=h1) Report\n> Merging [#1284](https://codecov.io/gh/huggingface/transformers/pull/1284?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/2dc8cb87341223e86220516951bb4ad84f880b4a?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/1284?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1284 +/- ##\n==========================================\n- Coverage 84.69% 84.68% -0.01% \n==========================================\n Files 84 84 \n Lines 12596 12598 +2 \n==========================================\n+ Hits 10668 10669 +1 \n- Misses 1928 1929 +1\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1284?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1284/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3V0aWxzLnB5) | `92.44% <66.66%> (-0.25%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1284?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/1284?src=pr&el=footer). Last update [2dc8cb8...c50783e](https://codecov.io/gh/huggingface/transformers/pull/1284?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "LGTM @slayton58 thanks" ]

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

[]

[ "You can just instanciate the models without the `.from_pretraining()` like so:\r\n```python\r\nconfig = BertConfig(**optionally your favorite parameters**)\r\nmodel = BertForPretraining(config)\r\n```\r\n\r\nI added a flag to `run_lm_finetuning.py` that gets checked in the `main()`. Maybe this snipped helps (note, I am only using this with Bert w/o next sentence prediction).\r\n\r\n```python\r\n# check if instead initialize freshly\r\nif args.do_fresh_init:\r\n config = config_class()\r\n tokenizer = tokenizer_class()\r\n if args.block_size <= 0:\r\n args.block_size = tokenizer.max_len # Our input block size will be the max possible for the model\r\n args.block_size = min(args.block_size, tokenizer.max_len)\r\n model = model_class(config=config)\r\nelse:\r\n config = config_class.from_pretrained(args.config_name if args.config_name else args.model_name_or_path)\r\n tokenizer = tokenizer_class.from_pretrained(args.tokenizer_name if args.tokenizer_name else args.model_name_or_path)\r\n if args.block_size <= 0:\r\n args.block_size = tokenizer.max_len # Our input block size will be the max possible for the model\r\n args.block_size = min(args.block_size, tokenizer.max_len)\r\n model = model_class.from_pretrained(args.model_name_or_path, from_tf=bool('.ckpt' in args.model_name_or_path), config=config)\r\nmodel.to(args.device)\r\n```", "Hi,\r\n\r\nthanks for the quick response.\r\nI am more interested in the XLNet and TransformerXL models. Would they have the same interface?\r\n\r\n", "I don’t know firsthand, but suppose so and it is fundamentally an easy problem to reinitialize weights randomly before any kind of training in pytorch :)\n\nGood luck,\nZacharias\nAm 18. Sep. 2019, 1:56 PM +0200 schrieb Stamenov <[email protected]>:\n> Hi,\n> thanks for the quick response.\n> I am more interested in the XLNet and TransformerXL models. Would they have the same interface?\n> —\n> You are receiving this because you commented.\n> Reply to this email directly, view it on GitHub, or mute the thread.\n", "I think XLNet requires a very specific training procedure, see #943 :+1: \r\n\r\n\"For XLNet, the implementation in this repo is missing some key functionality (the permutation generation function and an analogue of the dataset record generator) which you'd have to implement yourself.\"\r\n", "https://github.com/huggingface/pytorch-transformers/issues/1283#issuecomment-532598578\r\n\r\nHmm, tokenizers' constructors require a `vocab_file` parameter...", "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", "@Stamenov Did you figure out how to pretrain XLNet? I'm interested in that as well.", "No, I haven't. According to some recent tweet, huggingface could prioritize putting more effort into providing interfaces for self pre-training.", "You can now leave `--model_name_or_path` to None in `run_language_modeling.py` to train a model from scratch.\r\n\r\nSee also https://huggingface.co/blog/how-to-train" ]

[]

[]

[]

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

[]

[ "Hi Tuhin, you can use `examples/run_lm_finetuning.py` now. The scripts in `examples/lm_finetuning` are deprecated (removed on master now).", "Thomas i checked the wiki text 2 format and its confusing to me . Do we have to seperate documents by new lines ? My input file is a set of single sentence documents one per line . Do i need a new line after each sentence ?\r\n\r\nMy format is\r\nsent1\r\nsent2\r\nsent3\r\n\r\nDo i need to have\r\nsent1\r\n\r\nsent2\r\n\r\nsent3 ?\r\n\r\nI am currently running without a new line after each sentence and getting 09/19/2019 03:37:22 - WARNING - pytorch_transformers.tokenization_utils - Token indices sequence length is longer than the specified maximum sequence length for this model (3989110 >\r\n 512). Running this sequence through the model will result in indexing errors\r\n\r\n\r\n", "Also while running\r\ndef mask_tokens(inputs, tokenizer, args):\r\n\r\n- [ ] labels = inputs.clone()\r\n\r\n\r\n \r\n\r\n- [ ] # We sample a few tokens in each sequence for masked-LM training (with probability args.mlm_probability defaults to 0.15 in Bert/RoBERTa)\r\n- [ ] **masked_indices = torch.bernoulli(torch.full(labels.shape, args.mlm_probability)).bool()**\r\n- [ ] \r\n\r\nGetting error \r\nAttributeError: 'Tensor' object has no attribute 'bool'\r\nLine 110\r\n", "you need to update your pytorch version. I believe bool() function was intrcoduced in torch 1.1.0 or 1.2.0", "> I am currently running without a new line after each sentence and getting 09/19/2019 03:37:22 - WARNING - pytorch_transformers.tokenization_utils - Token indices sequence length is longer than the specified maximum sequence length for this model (3989110 >\r\n> 512). Running this sequence through the model will result in indexing errors\r\n\r\nI'm also getting this error. Is there more detail on the expected input format for the LM pretraining?", "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" ]

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

[]

[ "You can use `example/run_lm_finetuning` now, the scripts in the `example/lm_finetuning/` folder are deprecated (removed on master).", "What kind input format is good for `example/run_lm_finetuning.py`?", "Hi, there's an example using WikiText-2 in the [documentation](https://huggingface.co/pytorch-transformers/examples.html#language-model-fine-tuning). \r\n\r\nA file containing text is really all that's needed! You can change the way the file is used in `TextDataset` to better reflect the text you're fine-tuning the model to.", "Thanks, Lysandre.", "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" ]

[]

[]

[ "Indeed, I've checked and the 100 and 17 language models don't use language indices.\r\nJust supply `langs=None`.\r\nYou can see that in the official notebook from Facebook: https://github.com/facebookresearch/XLM/blob/master/generate-embeddings.ipynb", "I see. I missed this issue here which explains it pretty well. https://github.com/huggingface/pytorch-transformers/issues/1034\r\nThanks for the help!" ]

[ { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null }, { "id": 1565794707, "node_id": "MDU6TGFiZWwxNTY1Nzk0NzA3", "url": "https://api.github.com/repos/huggingface/transformers/labels/Write%20With%20Transformer", "name": "Write With Transformer", "color": "a84bf4", "default": false, "description": "" } ]

[]

[ "Hi @varkarrus, thank you for your feature request. There is a \"save & publish\" button on the top right-hand side, which saves your document on a specific URL. Does this fit your needs?", "Does it begin autosaving after you do that? If so, then probably!", "Nope, it does not currently auto-save 🙁\r\n\r\nOne quick fix I've thought about would be to pop a window alert if closing a tab that's got non-saved changes.", "Oh, even that alone would be a lifesaver!", "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" ]

[]

[]

[ "Thanks for adding this :+1: \r\n\r\nI've one suggestion for some improvement (rfc): can we make the `get_labels()` function a bit more configurable? E.g. reading the labels from a file `labels.txt` would be great, so I could use other datasets (e.g. GermEval, which has more labels) 🤔 What do you think 🤗", "That would be useful for fine-tuning on other datasets, I agree. I tried to keep it very close to the run_glue example for the beginning (where the labels are also fixed), and wait for feedback from the maintainers to know if these kind of extensions are wanted or not ;-)\r\nHow about adding a new CLI argument where the user can specify a path to a labels file, and we use the default CoNLL labels when no path was specified?", "@stecklin Would be great to have a cli argument for that (+ like your proposed suggestion) :heart: ", "This looks awesome, thanks a lot @stecklin and @stefan-it!\r\nHappy to review this when you folks think it's ready.\r\nAnd ping me if I can help otherwise.", "Results are pretty good. I wrote an additional prediction script that outputs a CoNLL compatible format, so that I could verify the results with the official CoNLL evaluation script. Here I fine-tuned a `bert-base-cased` model (5 epochs):\r\n\r\nDevelopment set:\r\n\r\n```bash\r\nprocessed 51362 tokens with 5942 phrases; found: 5997 phrases; correct: 5661.\r\naccuracy: 99.10%; precision: 94.40%; recall: 95.27%; FB1: 94.83\r\n LOC: precision: 96.74%; recall: 96.90%; FB1: 96.82 1840\r\n MISC: precision: 89.54%; recall: 91.00%; FB1: 90.26 937\r\n ORG: precision: 92.24%; recall: 92.24%; FB1: 92.24 1341\r\n PER: precision: 96.06%; recall: 97.99%; FB1: 97.02 1879\r\n```\r\n\r\nTest set:\r\n\r\n```bash\r\nprocessed 46435 tokens with 5648 phrases; found: 5712 phrases; correct: 5185.\r\naccuracy: 98.26%; precision: 90.77%; recall: 91.80%; FB1: 91.29\r\n LOC: precision: 92.12%; recall: 93.23%; FB1: 92.67 1688\r\n MISC: precision: 79.75%; recall: 82.48%; FB1: 81.09 726\r\n ORG: precision: 89.23%; recall: 90.31%; FB1: 89.77 1681\r\n PER: precision: 95.92%; recall: 95.92%; FB1: 95.92 1617\r\n```", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1275?src=pr&el=h1) Report\n> Merging [#1275](https://codecov.io/gh/huggingface/transformers/pull/1275?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/80889a0226b8f8022fd9ff65ed6bce71b60ba800?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1275?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1275 +/- ##\n=======================================\n Coverage 85.98% 85.98% \n=======================================\n Files 91 91 \n Lines 13579 13579 \n=======================================\n Hits 11676 11676 \n Misses 1903 1903\n```\n\n\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1275?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/1275?src=pr&el=footer). Last update [80889a0...c55badc](https://codecov.io/gh/huggingface/transformers/pull/1275?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "The labels can be configured now :heavy_check_mark: \r\n@stefan-it Do we want to add your prediction script as well? I think that would be very useful, after all NER prediction is not as straightforward as e.g. sequence classification prediction.", "@stecklin Thanks 😊 I'm currently using this script:\r\n\r\nhttps://gist.github.com/stefan-it/c39b63eb0043182010f2f61138751e0f\r\n\r\nIt mainly copies parts from the `evaluate` function. \r\n\r\nBut I think a more elegant way would be to fully re-use the evaluate function. The function currently returns the evaluation result, but maybe it could return a tuple of results and predicted tags?", "@stefan-it I followed your suggestion, the evaluate function now returns the results and the predictions. I added the argument `--do_predict` to predict on a test set.\r\n@thomwolf I think now would be a good moment for you to have a look. Let me know your feedback!", "The mentioned script compatible with new \"Transformers\" source code ? ", "This looks awesome, thank you for the script. I did something similar that worked but this code is totally better. Thanks @stecklin !!", "Ok I've reviewed the PR and it looks great, thanks a lot @stecklin and @stefan-it.\r\n\r\nI've rebased, switched from `pytorch-transformers` to `transformers` and added `seqeval` in the requirements.\r\n\r\nThe only missing element is to add a simple usage explanation in the examples readme file at `examples/README.md` which explain:\r\n- how to download the training/testing data,\r\n- an example of command-line to run the script, and\r\n- an example of results with this command line.\r\n\r\n@stefan-it do you want to share the command line you use for the above results?", " @thomwolf No problem, here's an example for GermEval 2014 (German NER):\r\n\r\n# Data (Download and pre-processing steps)\r\n\r\nData can be obtained from the [GermEval 2014](https://sites.google.com/site/germeval2014ner/data) shared task page.\r\n\r\nHere are the commands for downloading and pre-processing train, dev and test datasets. The original data format has four (tab-separated) columns, in a pre-processing step only the two relevant columns (token and outer span NER annotation) are extracted:\r\n\r\n```bash\r\ncurl -L 'https://sites.google.com/site/germeval2014ner/data/NER-de-train.tsv?attredirects=0&d=1' \\\r\n| grep -v \"^#\" | cut -f 2,3 | tr '\\t' ' ' > train.txt.tmp\r\ncurl -L 'https://sites.google.com/site/germeval2014ner/data/NER-de-dev.tsv?attredirects=0&d=1' \\\r\n| grep -v \"^#\" | cut -f 2,3 | tr '\\t' ' ' > dev.txt.tmp\r\ncurl -L 'https://sites.google.com/site/germeval2014ner/data/NER-de-test.tsv?attredirects=0&d=1' \\\r\n| grep -v \"^#\" | cut -f 2,3 | tr '\\t' ' ' > test.txt.tmp\r\n```\r\n\r\nThe GermEval 2014 dataset contains some strange \"control character\" tokens like `'\\x96', '\\u200e', '\\x95', '\\xad' or '\\x80'`. One problem with these tokens is, that `BertTokenizer` returns an empty token for them, resulting in misaligned `InputExample`s. I wrote a script that a) filters these tokens and b) splits longer sentences into smaller ones (once the max. subtoken length is reached).\r\n\r\n```bash\r\nwget \"https://raw.githubusercontent.com/stefan-it/fine-tuned-berts-seq/master/scripts/preprocess.py\"\r\n```\r\nLet's define some variables that we need for further pre-processing steps and training the model:\r\n\r\n```bash\r\nexport MAX_LENGTH=128\r\nexport BERT_MODEL=bert-base-multilingual-cased\r\n```\r\n\r\nRun the pre-processing script on training, dev and test datasets:\r\n\r\n```bash\r\npython3 preprocess.py train.txt.tmp $BERT_MODEL $MAX_LENGTH > train.txt\r\npython3 preprocess.py dev.txt.tmp $BERT_MODEL $MAX_LENGTH > dev.txt\r\npython3 preprocess.py test.txt.tmp $BERT_MODEL $MAX_LENGTH > test.txt\r\n```\r\n\r\nThe GermEval 2014 dataset has much more labels than CoNLL-2002/2003 datasets, so an own set of labels must be used:\r\n\r\n```bash\r\ncat train.txt dev.txt test.txt | cut -d \" \" -f 2 | grep -v \"^$\"| sort | uniq > labels.txt\r\n```\r\n\r\n# Training\r\n\r\nAdditional environment variables must be set:\r\n\r\n```bash\r\nexport OUTPUT_DIR=germeval-model\r\nexport BATCH_SIZE=32\r\nexport NUM_EPOCHS=3\r\nexport SAVE_STEPS=750\r\nexport SEED=1\r\n```\r\n\r\nTo start training, just run:\r\n\r\n```bash\r\npython3 run_ner.py --data_dir ./ \\\r\n--model_type bert \\\r\n--labels ./labels.txt \\\r\n--model_name_or_path $BERT_MODEL \\\r\n--output_dir $OUTPUT_DIR \\\r\n--max_seq_length $MAX_LENGTH \\\r\n--num_train_epochs $NUM_EPOCHS \\\r\n--per_gpu_train_batch_size $BATCH_SIZE \\\r\n--save_steps $SAVE_STEPS \\\r\n--seed $SEED \\\r\n--do_train \\\r\n--do_eval \\\r\n--do_predict\r\n```\r\n\r\nIf your GPU supports half-precision training, just add the `--fp16` flag. After training, the model will be both evaluated on development and test datasets.\r\n\r\n# Evaluation\r\n\r\nEvaluation on development dataset outputs the following for our example:\r\n\r\n```bash\r\n10/04/2019 00:42:06 - INFO - __main__ - ***** Eval results *****\r\n10/04/2019 00:42:06 - INFO - __main__ - f1 = 0.8623348017621146\r\n10/04/2019 00:42:06 - INFO - __main__ - loss = 0.07183869666975543\r\n10/04/2019 00:42:06 - INFO - __main__ - precision = 0.8467916366258111\r\n10/04/2019 00:42:06 - INFO - __main__ - recall = 0.8784592370979806\r\n```\r\n\r\nOn the test dataset the following results could be achieved:\r\n\r\n```bash\r\n10/04/2019 00:42:42 - INFO - __main__ - ***** Eval results *****\r\n10/04/2019 00:42:42 - INFO - __main__ - f1 = 0.8614389652384803\r\n10/04/2019 00:42:42 - INFO - __main__ - loss = 0.07064602487454782\r\n10/04/2019 00:42:42 - INFO - __main__ - precision = 0.8604651162790697\r\n10/04/2019 00:42:42 - INFO - __main__ - recall = 0.8624150210424085\r\n```\r\n\r\nPlease let me know if you have more questions 🤗", "Hi, great work, thanks for sharing!\r\n\r\nI think the argument `overwrite_cache` is not used in the code. I suspect there is a missing if check in the `load_and_cache_examples()` function.", "There was something strange with git on this branch (32 files changed...) so I had to do a rebase and force push on your PR @stecklin. Please do a `git reset --hard` to be up-to-date with the new clean state on the remote repo.\r\n\r\nNow it looks in order for merging with master.", "@stefan-it Not able to reproduce the above results. \r\n\r\nThe best I can for dev dataset get is this : \r\n\r\n11/27/2019 18:16:38 - INFO - __main__ - ***** Eval results *****\r\n11/27/2019 18:16:38 - INFO - __main__ - f1 = 0.12500000000000003\r\n11/27/2019 18:16:38 - INFO - __main__ - loss = 1.6597001552581787\r\n11/27/2019 18:16:38 - INFO - __main__ - precision = 0.2\r\n11/27/2019 18:16:38 - INFO - __main__ - recall = 0.09090909090909091\r\n\r\nAny pointers on what I am missing ?\r\n", "@oneraghavan What version/commit of `transformers` are you using? Do you use the GermEval dataset or another one? I'll check the example :)", "@stefan-it Thanks for quick response :) . I am using commit from oct 24. Has anything changed since then ? I am following the same steps said in examples/readme.md . Let me know if you want me to check with latest commit .", "Could you try to use the latest `master` version? I re-do the experiment on GermEval, here are the results:\r\n\r\nEvaluation on dev set:\r\n\r\n```bash\r\nf1 = 0.8702821546353977\r\nloss = 0.07410008722260086\r\nprecision = 0.8530890804597702\r\nrecall = 0.8881824981301422\r\n```\r\n\r\nEvaluation on test set:\r\n\r\n```bash\r\nf1 = 0.860249697946033\r\nloss = 0.07239935705435063\r\nprecision = 0.8561808561808562\r\nrecall = 0.8643573972159275\r\n```", "@stefan-it I tried with the latest master. Not able to reproduce. I am exactly following the instructions given in readme.md. The following are the results I am getting .\r\n\r\n11/28/2019 09:34:50 - INFO - __main__ - ***** Eval results *****\r\n11/28/2019 09:34:50 - INFO - __main__ - f1 = 0.12500000000000003\r\n11/28/2019 09:34:50 - INFO - __main__ - loss = 1.1732935905456543\r\n11/28/2019 09:34:50 - INFO - __main__ - precision = 0.2\r\n11/28/2019 09:34:50 - INFO - __main__ - recall = 0.09090909090909091\r\n\r\nCan you check if there are any other parameters that is being given in run_ner.py parameters . How much epoch are you training ? ", "In order to reproduce the conll score reported in BERT paper (92.4 bert-base and 92.8 bert-large) one trick is to apply a truecaser on article titles (all upper case sentences) as preprocessing step for conll train/dev/test. This can be simply done with the following method.\r\n\r\n```\r\n#https://github.com/daltonfury42/truecase\r\n#pip install truecase\r\nimport truecase\r\nimport re\r\n\r\n\r\n\r\n\r\n# original tokens\r\n#['FULL', 'FEES', '1.875', 'REOFFER', '99.32', 'SPREAD', '+20', 'BP']\r\n\r\ndef truecase_sentence(tokens):\r\n word_lst = [(w, idx) for idx, w in enumerate(tokens) if all(c.isalpha() for c in w)]\r\n lst = [w for w, _ in word_lst if re.match(r'\\b[A-Z\\.\\-]+\\b', w)]\r\n\r\n if len(lst) and len(lst) == len(word_lst):\r\n parts = truecase.get_true_case(' '.join(lst)).split()\r\n\r\n # the trucaser have its own tokenization ...\r\n # skip if the number of word dosen't match\r\n if len(parts) != len(word_lst): return tokens\r\n\r\n for (w, idx), nw in zip(word_lst, parts):\r\n tokens[idx] = nw\r\n\r\n# truecased tokens\r\n#['Full', 'fees', '1.875', 'Reoffer', '99.32', 'spread', '+20', 'BP']\r\n\r\n```\r\n\r\nAlso, i found useful to use : very small learning rate (5e-6) \\ large batch size (128) \\ high epoch num (>40).\r\n\r\nWith these configurations and preprocessing, I was able to reach 92.8 with bert-large." ]

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

[]

[ "I think this is an interesting addition and I like the way the PR is structured in general.\r\n\r\nBefore I dive in, could you lay down the status of the PR in terms of supported models, python version (we will still keep python 2 support for now), know issues and TO-DOs?", "The fully supported models are BERT, GPT2, XLNet, and RoBERTa. For these models tokenize_with_offsets always produces the same tokens as tokenize and the subword tokens are well-aligned to the original text, typically as well as possible.\r\n\r\nXLM has a known issue with patterns like '... inc. reported ...'\r\ntokxlm.tokenize_with_offsets('inc. reported')\r\n['inc</w>', '.</w>', 'reported</w>']\r\ntokxlm.tokenize('inc. reported')\r\n['inc.</w>', 'reported</w>']\r\nBecause tokenize_with_offsets passes whitespace separated 'chunks' (utils_squad style) the xlm tokenizer doesn't get to look ahead to see that a lowercase word follows the period.\r\n\r\nTransfoXL just needs the latest commit (not in this PR), while OpenAIGPT has an issue with '\\n</w>' tokens, which tokenize_with_offsets never produces.\r\n\r\nEverything works in Python 2 except GPT2 (and therefore RoBERTa) as well as XLNet. I think this is the case even without this PR. I'm not experienced with Python 2 though, so maybe I am just missing something.\r\n\r\n\r\n", "# [Codecov](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274?src=pr&el=h1) Report\n> Merging [#1274](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/pytorch-transformers/commit/45de034bf899af678d844351ff21ea0444815ddb?src=pr&el=desc) will **decrease** coverage by `0.97%`.\n> The diff coverage is `59.04%`.\n\n[](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1274 +/- ##\n==========================================\n- Coverage 81.16% 80.19% -0.98% \n==========================================\n Files 57 60 +3 \n Lines 8039 8405 +366 \n==========================================\n+ Hits 6525 6740 +215 \n- Misses 1514 1665 +151\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [...tests/regression\\_test\\_tokenization\\_with\\_offsets.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdGVzdHMvcmVncmVzc2lvbl90ZXN0X3Rva2VuaXphdGlvbl93aXRoX29mZnNldHMucHk=) | `0% <0%> (ø)` | |\n| [pytorch\\_transformers/tokenization\\_bert.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdG9rZW5pemF0aW9uX2JlcnQucHk=) | `95.79% <100%> (+0.12%)` | :arrow_up: |\n| [pytorch\\_transformers/tokenization\\_gpt2.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdG9rZW5pemF0aW9uX2dwdDIucHk=) | `96.87% <100%> (+0.18%)` | :arrow_up: |\n| [...ytorch\\_transformers/tests/tokenization\\_xlm\\_test.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdGVzdHMvdG9rZW5pemF0aW9uX3hsbV90ZXN0LnB5) | `97.87% <100%> (+0.14%)` | :arrow_up: |\n| [...ch\\_transformers/tests/tokenization\\_offsets\\_test.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdGVzdHMvdG9rZW5pemF0aW9uX29mZnNldHNfdGVzdC5weQ==) | `100% <100%> (ø)` | |\n| [...transformers/tests/tokenization\\_transfo\\_xl\\_test.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdGVzdHMvdG9rZW5pemF0aW9uX3RyYW5zZm9feGxfdGVzdC5weQ==) | `97.36% <100%> (+0.3%)` | :arrow_up: |\n| [...ch\\_transformers/tests/tokenization\\_roberta\\_test.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdGVzdHMvdG9rZW5pemF0aW9uX3JvYmVydGFfdGVzdC5weQ==) | `92.85% <100%> (+0.4%)` | :arrow_up: |\n| [pytorch\\_transformers/tokenization\\_transfo\\_xl.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdG9rZW5pemF0aW9uX3RyYW5zZm9feGwucHk=) | `34.54% <100%> (+0.36%)` | :arrow_up: |\n| [pytorch\\_transformers/tokenization\\_xlm.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdG9rZW5pemF0aW9uX3hsbS5weQ==) | `82.98% <100%> (+0.28%)` | :arrow_up: |\n| [pytorch\\_transformers/tokenization\\_openai.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdG9rZW5pemF0aW9uX29wZW5haS5weQ==) | `82.4% <100%> (+0.58%)` | :arrow_up: |\n| ... and [11 more](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274?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/pytorch-transformers/pull/1274?src=pr&el=footer). Last update [45de034...328e698](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1274?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Would love to see this feature available - what can I do to help get it merged in?", "We are currently working on a larger project around this and should come to this PR pretty soon (next week I hope).", "@michaelrglass How does this handle the destructive normalisation that occurs in eg BertTokenizer? Specifically, logic like [this](https://github.com/huggingface/transformers/blob/master/transformers/tokenization_bert.py#L330) means that the normalisation isn't length preserving, and it may not be possible to find the (normalised) token in the original input text.", "> @michaelrglass How does this handle the destructive normalisation that occurs in eg BertTokenizer? Specifically, logic like [this](https://github.com/huggingface/transformers/blob/master/transformers/tokenization_bert.py#L330) means that the normalisation isn't length preserving, and it may not be possible to find the (normalised) token in the original input text.\r\n\r\nIMO it's too error-prone to run a destructive tokenizer and then try to align the sequences after the fact. I wrote https://github.com/microsoft/bistring for this exact kind of problem. Otherwise it's very tricky to align substrings of the modified text with the original text, as both NFD and filtering out nonspacing marks can shift the positions of chars significantly.", "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" ]

[ { "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" ]

[]

[]

[ "Maybe the learning rate is too high?", "I change a GPU node and this situation doesn't appear. I will change the learning rate and see results. Thanks a lot." ]

[]

[]

[ "Are you initializing from a pretrained model? If no, than this is normal behaviour: your weights are randomly initialized. If yes, make sure your model is in evaluation mode (```model.eval()```), this disables dropout and other random modules.", "Thank you for quick response @srslynow . how to initialize weights and biases fora pre-trained model ? I thought weights and biases freezes after training.\r\n\r\nyou were right , I missed the model.eval() that's the reason I was getting slightly different embedding on each run becz of dropout layer.\r\n\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 } ]

[]

[ "You should give a look at NVIDIA's apex library and PyTorch `model.half()` method.", "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", "@thomwolf Do you know if the gpt2 model needs to be pre-trained with apex support in order to use NVIDIA's apex library (e.g. O1 mode) at inference time? Was mixed precision used during the training of the gpt2 model? Is there a way I can verify that?", "Do we have models trained with mixed precision enabled for gpt2? I can't find them in huggingface's repo. " ]

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

[]

[ "Hey, so not exactly a direct answer to your question, but bert outright doesn't do amazing on sentence similarity. This repo here should help with your question and their paper does a great job at explaining how their method works https://github.com/UKPLab/sentence-transformers. I think you will find better results with that. Hope it helps.", "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, I experienced this also in several experiments. BERT large is extremely sensitive to the random seed. Try some other seeds and you will likely get a performance at least on oar with the base model.\r\n\r\nI haven't it studied further why the large model is so sensitive to the random seed, but it appears that the gradient for some step destroys the model, from which on you only get bad scores. Might be some exploding gradient or some Nan issues.\r\n\r\nBest, \r\nNils Reimers ", "I find this issue as well -- no convergence with the large model.\r\n\r\nPotentially related:\r\n\r\nhttps://github.com/huggingface/transformers/issues/753\r\nhttps://github.com/huggingface/transformers/issues/92", "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 @YosiMass,\r\nThe simplest/more direct way to do transfer learning is indeed the 3rd solution.\r\nIf you use `run_glue.py`, the modification you made is correct. You also have to be careful since DistilBERT doesn't take `token_type_embeddings` as input --> [here](https://github.com/huggingface/pytorch-transformers/blob/master/examples/run_glue.py#L131) and [here](https://github.com/huggingface/pytorch-transformers/blob/master/examples/run_glue.py#L221).\r\nI'll add these modifications in a few days directly to these scripts so that it's seamless to use DistilBERT with run_squad or run_glue.", "Thanks @VictorSanh. \r\n\r\nYes, I handled the missing token_types as follows. In run_glue.train and run_glue.evaluate I changed From \r\n``` \r\ninputs = {'input_ids': batch[0],\r\n 'attention_mask': batch[1],\r\n 'token_type_ids': batch[2] if args.model_type in ['bert', 'xlnet'] else None, # XLM and RoBERTa don't use segment_ids\r\n 'labels': batch[3]}\r\n```\r\n\r\nTo\r\n```\r\nif args.model_type not in ['distilbert']:\r\n inputs = {'input_ids': batch[0],\r\n 'attention_mask': batch[1],\r\n 'token_type_ids': batch[2] if args.model_type in ['bert', 'xlnet'] else None, # XLM and RoBERTa don't use segment_ids\r\n 'labels': batch[3]}\r\nelse:\r\n inputs = {'input_ids': batch[0],\r\n 'attention_mask': batch[1],\r\n 'labels': batch[3]}\r\n```", "> handled the missing token_types as follows. In run_glue.train and run_glue.evaluate I changed From\r\n\r\nIt looks good to me!" ]

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

[]

[ "Yes!\r\n\r\nYou have to fine-tune BertForMultipleChoice to be able to use it.\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", "Does Bert add extra hidden layers that are randomly initialized on top of the pre trained network when using BertForMultipleChoice? Would these added hidden layers be the only hidden layers that are adjusted during the learning process?", "> Does Bert add extra hidden layers that are randomly initialized on top of the pre trained network when using BertForMultipleChoice? Would these added hidden layers be the only hidden layers that are adjusted during the learning process?\r\n\r\nOverall, the parameters of pretrained mode starts from saved states, and the parameters of the classifier are initialize randomly as it can be checked from transformers package. During the fine tuning all of the parameters are fine tuned." ]

[]

[]

[ "Ok should be fixed now on master with e768f23" ]

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

[]

[ "I am glad to see someone is working on this and really appreciate your work.\r\nCurrently I'm using LCS the original xlnet also uses to align the token and raw input to extract the answer highlighting.\r\nThis is really painful as it's slow and may fail at some corner case.\r\nCan't wait to see your pulled feature merged!", "Commenting in #1274 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" ]

[]

[]

[ "Closing own non-issue. Needed to download missing files listed in the four applicable tokenizer scripts. Working 100%. On to fine-tuning." ]

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

[]

[ "That's how it's done in the respective original implementations.", "it is very interesting. All other tokens will not be considered. \r\n\r\nDo you know whether other architectures has been tried out? Initially, when I had taken the transformer-model I created another output architecture. It is very interesting, whether this \"end architecture\" affects the performance significantly\r\n\r\nThanks a lot!", "I found this project doc [1], there are four variants of the end architecture, their performance is almost equal. \r\n\r\nI will be glad if you can provide other papers regarding this issue. \r\n\r\n[1] - http://web.stanford.edu/class/cs224n/reports/custom/15785631.pdf", "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,\r\n`tokenizer.decode()` expect a sequence of ids as indicated in the doc/docstring: https://huggingface.co/pytorch-transformers/main_classes/tokenizer.html#pytorch_transformers.PreTrainedTokenizer.decode", "well, as far as i see, the tokenizer can accept both - sequence or single index. For example:\r\n`tokenizer = XLNetTokenizer.from_pretrained('xlnet-base-cased')`\r\n`tokenizer.decode(3)`\r\nproduce `'<cls>'` as well as `tokenizer.decode([3])`.\r\n\r\nAnd the problem happens in both cases:\r\n`tokenizer.decode([4])` produces empty string. And, what is even more strange, if index 4 is in the sequence, the result is a list instead of a string. For example:\r\n\r\n`tokenizer.decode([35,109])` gives `'I new'`, but `tokenizer.decode([35,109,4])` generates list instead of a string `['I new']`", "ok, it looks like 4 is an index of a separation token and it turns one sequence into sequence of sequences each of which will be decoded into string. So if index 4 presents in the sequence, the result will be list of strings. This behavior is quite unintuitive. If it was created purposely, should be properly documented. I've got very strange bug in my code that was hard to track, since index 4 was predicted very rarely and out of 500 predictions 1 was buggy. ", "Thanks for the bug report. Indeed this was an unwanted behavior. Fixed on master now." ]

[ { "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" ]

[]

[]

[ "LGTM but let's have @julien-c or @LysandreJik confirm", "lgtm too", "It would be appreciated if you review this PR! @LysandreJik ", "@ikuyamada Out of curiosity, in which cases did you need to specify this `padding_idx`? It shouldn't have any impact on the inference so are you training a model from scratch? (with @LysandreJik)", "Merged in https://github.com/huggingface/transformers/commit/a6a6d9e6382961dc92a1a08d1bab05a52dc815f9", "big drawback is that we initialize the embeddings multiple times.\r\n\r\n@ikuyamada do you have an idea to improve this?", "Thank you for merging this PR!\r\n\r\n> Out of curiosity, in which cases did you need to specify this padding_idx? It shouldn't have any impact on the inference so are you training a model from scratch?\r\n\r\nIn my understanding, the embedding corresponding to `padding_idx` is not updated while training (pre-training or fine-tuning). Because the embedding of the token `<s>` may play some roles for computing contextualized embeddings for other tokens, and the output embedding of the `<s>` token is used for computing a feature vector for some fine-tuning tasks, I think the embedding should be updated while training.\r\n\r\n> big drawback is that we initialize the embeddings multiple times.\r\n> @ikuyamada do you have an idea to improve this?\r\n\r\nWe can avoid this by removing [the constructor call of the `BertEmbeddings`](https://github.com/huggingface/transformers/blob/master/transformers/modeling_roberta.py#L44) and simply initialize the `token_type_embeddings`, `LayerNorm`, and `dropout` in the constructor of the `RobertaEmbeddings`.\r\nIf you prefer this implementation, I will create a PR again! :)\r\n@julien-c ", "> In my understanding, the embedding corresponding to padding_idx is not updated while training (pre-training or fine-tuning). Because the embedding of the token `<s>` may play some roles for computing contextualized embeddings for other tokens, and the output embedding of the `<s>` token is used for computing a feature vector for some fine-tuning tasks, I think the embedding should be updated while training.\r\n\r\nYes you are correct.\r\n\r\n> If you prefer this implementation, I will create a PR again! :)\r\n\r\nThat would be great, thank you." ]

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

[]

[ "Hi, \r\n\r\nFound a solution to my problem to some extent. I had cloned the latest apex repo and testing out with PyTorch-pretrained-Bert which is causing the problem to take more time for execution. I took the older apex repo and tested the code and working as earlier (45 min per epoch)\r\n\r\nBut when I was using the pytorch-transformers with latest apex repo it is taking 6 hours for one epoch. \r\n\r\nIs that usual to take 6 hours for one epoch in colab ??\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" ]

[]

[]

[ "What didn't work for you with the fairseq implementation?\r\n\r\nIt seems pretty self-contained: https://github.com/pytorch/fairseq/blob/master/fairseq/optim/adafactor.py#L65-L213", "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", "FYI @sshleifer -- I was wrong -- able to train T5-large even batch==1 with FP32, no gradient check-pointing and ADAM. Given that T5 team strongly recommends AdaFactor -- giving it a try, other pieces perhaps being more difficult..." ]

[]

[]

[ "Additional info: simple_lm_finetuning.py works with pytorch-transformers version 1.1.0, but not with version 1.2.0.", "Maybe related to the change in order of parameters to BertModel's forward method ?\r\nSee #1246", "Hi, the lm finetuning examples are now replaced by `examples/run_lm_finetuning`" ]

[ { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null }, { "id": 1565794707, "node_id": "MDU6TGFiZWwxNTY1Nzk0NzA3", "url": "https://api.github.com/repos/huggingface/transformers/labels/Write%20With%20Transformer", "name": "Write With Transformer", "color": "a84bf4", "default": false, "description": "" } ]

[]

[ "Commenting to say I have also noticed this. Also I would assume the small amount of tokens being generated per autocomplete lately is because of compute concerns, not time concerns. It is a bit limiting.", "Also having this problem!", "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 } ]

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

[ "Seems like the run_squad script is in bad shape now. It just doesn't work.", "same question.. also running this script with XLNet on Squad, is ~10 F1 points below BERT-Large-WWM. The difference in preprocessing as pointed out above could be one of the reasons.", "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/pytorch-transformers/pull/1252?src=pr&el=h1) Report\n> Merging [#1252](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1252?src=pr&el=desc) into [glue-example](https://codecov.io/gh/huggingface/pytorch-transformers/commit/5583711822f79d8b3b7e7ba2560748cc0cf5654f?src=pr&el=desc) will **increase** coverage by `0.08%`.\n> The diff coverage is `90.9%`.\n\n[](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1252?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## glue-example #1252 +/- ##\n===============================================\n+ Coverage 81.32% 81.4% +0.08% \n===============================================\n Files 57 57 \n Lines 8074 8104 +30 \n===============================================\n+ Hits 6566 6597 +31 \n+ Misses 1508 1507 -1\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1252?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [...h\\_transformers/tests/tokenization\\_tests\\_commons.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1252/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdGVzdHMvdG9rZW5pemF0aW9uX3Rlc3RzX2NvbW1vbnMucHk=) | `100% <100%> (ø)` | :arrow_up: |\n| [pytorch\\_transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1252/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdG9rZW5pemF0aW9uX3V0aWxzLnB5) | `90.02% <80%> (+0.57%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1252?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/pytorch-transformers/pull/1252?src=pr&el=footer). Last update [5583711...a804892](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1252?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

[]

[]

[ "This is a smaller model compared to the original, and is thus better suitable for usage on embedded devices / devices without large gpu's. Their blog posts explains this: https://medium.com/huggingface/distilbert-8cf3380435b5.", "Hello @tomohideshibata,\r\nTrue, we could use the same code base for BERT and DistilBERT.\r\nFor now, I prefer keeping them separate mainly for clarity since the architectures are slightly different:\r\n- No token_type_embeddings in DistilBERT\r\n- No Sequence Pooler in DistilBERT\r\nHandling these two in BertModel would unnecessarily (slightly) complexify the code and I'd like to keep it clean.\r\nAnother caveat: I use torch's `nn.LayerNorm` in DistilBERT while BERT uses a custom `BertLayerNorm`. There might be slightly different edge cases (I have to check).\r\nBut overall, you can totally implement a single class for BERT and DistilBERT on your side. I would suggest to have a look at [this script](https://github.com/huggingface/pytorch-transformers/blob/master/examples/distillation/scripts/extract_for_distil.py) to have the mapping between the names.", "@VictorSanh Thanks for your comments. I will try to make a conversion script from DistilBERT to BERT weights." ]

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

[]

[ "This is mostly a new head for Bert, right?\r\nIf so, yes I think it could be a nice addition.\r\nIs the tokenizer different as well?", "That's correct. The new head is [here](https://github.com/azdatascience/pytorch-transformers/blob/rbert/pytorch_transformers/modeling_bert.py#L832) (Think that's in the right place). A new [tokenizer](https://github.com/azdatascience/pytorch-transformers/blob/rbert/pytorch_transformers/tokenization_rbert.py) is required as well, as it needs to insert some special characters surrounding the entities of interest. ", "Ok, I think we can accept a PR for that if you want to submit one.\r\n\r\nTwo notes on that:\r\n- the tokenizer can inherit from `BertTokenizer` and you can probably mostly override the methods called `add_special_tokens_single_sentence` and `add_special_tokens_sentences_pair` to insert the special characters.\r\n- the model and tokenizer should have tests (check how we test the other models, it's pretty simple) and docstring.\r\n- adding an example similar to `run_glue` would be nice also I think or just a usage example in the docstring.", "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/pytorch-transformers/pull/1249?src=pr&el=h1) Report\n> Merging [#1249](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1249?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/pytorch-transformers/commit/364920e216c16d73c782a61a4cf6652e541fbe18?src=pr&el=desc) will **decrease** coverage by `0.21%`.\n> The diff coverage is `61.11%`.\n\n[](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1249?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1249 +/- ##\n==========================================\n- Coverage 81.23% 81.02% -0.21% \n==========================================\n Files 57 57 \n Lines 8029 8035 +6 \n==========================================\n- Hits 6522 6510 -12 \n- Misses 1507 1525 +18\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1249?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [pytorch\\_transformers/modeling\\_transfo\\_xl.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1249/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvbW9kZWxpbmdfdHJhbnNmb194bC5weQ==) | `54.43% <60%> (-0.18%)` | :arrow_down: |\n| [pytorch\\_transformers/modeling\\_utils.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1249/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvbW9kZWxpbmdfdXRpbHMucHk=) | `90.03% <66.66%> (-0.25%)` | :arrow_down: |\n| [pytorch\\_transformers/file\\_utils.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1249/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvZmlsZV91dGlscy5weQ==) | `66.66% <0%> (-4.85%)` | :arrow_down: |\n| [...orch\\_transformers/tests/tokenization\\_utils\\_test.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1249/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdGVzdHMvdG9rZW5pemF0aW9uX3V0aWxzX3Rlc3QucHk=) | `92% <0%> (-4%)` | :arrow_down: |\n| [...h\\_transformers/tests/tokenization\\_tests\\_commons.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1249/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdGVzdHMvdG9rZW5pemF0aW9uX3Rlc3RzX2NvbW1vbnMucHk=) | `97.43% <0%> (-2.57%)` | :arrow_down: |\n| [pytorch\\_transformers/tokenization\\_gpt2.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1249/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdG9rZW5pemF0aW9uX2dwdDIucHk=) | `95.86% <0%> (-0.83%)` | :arrow_down: |\n| [pytorch\\_transformers/tokenization\\_transfo\\_xl.py](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1249/diff?src=pr&el=tree#diff-cHl0b3JjaF90cmFuc2Zvcm1lcnMvdG9rZW5pemF0aW9uX3RyYW5zZm9feGwucHk=) | `33.89% <0%> (-0.29%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1249?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/pytorch-transformers/pull/1249?src=pr&el=footer). Last update [364920e...8bdee1c](https://codecov.io/gh/huggingface/pytorch-transformers/pull/1249?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "LGTM, thanks @ziliwang" ]

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

[]

[ "Hi Tuhin, adding GPT/GPT2 is a tiny bit more complicated since these models need to have special tokens added to their vocabulary prior to fine-tuning (just a few more lines of code though).\r\n\r\nDo you want to try to make a PR with these other models?", "Okay I will add bert as the model type i was just not sure if that d be the right thing to do , hence asked :)", "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,\r\nWe only add models when there are pretrained weights released.\r\nThis doesn't seem to be the case for KnowBert, or maybe I missed them?", "Yes, looking into if they are releasing pretrained weights, I incorrectly assumed they were. ", "Wow! Really looking forward to this.\r\nI really feel that models that combine text and unstructured data are not getting enough attention.\r\nThis is so relevant for creating great AI products because I dare to say that in most real life applications you do not deal with text only data. Metadata is crucial!", "The authors released the model. Any update on integrating it into huggingface?", "We'd welcome a community or author-contributed implementation! \r\n\r\n(Also might look into integrating it ourselves at some point, but bandwidth is low)\r\n\r\n[Update: [link to the implem + weights](https://github.com/allenai/kb)]", "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 Community, Any update on integrating KnowBert into huggingface?" ]

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

[]

[ "Indeed. What do you mean by \"examples\"? The docstrings examples?", "Most of examples folder.\r\nIn particular run_swag.py and lm finetuning scripts.", "just came here to say the same. ", "Indeed, I've fixed and cleaned up the examples in 8334993 (the lm finetuning examples are now replaced by `run_lm_finetuning`). Also indicated more clearly which examples are not actively maintained and tested.", "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" ]

[]

[]

[ "sorry I found the problem: the pip source code and the zip code downloaded are different, especially for the \"BertForSequenceClassification\" class" ]

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

[]

[ "Hi, by unconditional generation do you mean generating a sequence from no context? If so, if using GPT-2, you can set your initial context to be: `<|endoftext|>`. This will generate sequences with no other initial context.\r\n\r\nYou could do so like this:\r\n\r\n```\r\nfrom pytorch_transformers import GPT2Tokenizer\r\ntokenizer = GPT2Tokenizer.from_pretrained(\"gpt2\")\r\ncontext = [tokenizer.encoder[\"<|endoftext|>\"]]\r\n```", "@LysandreJik Yeah I did that but It is not performing good, it generated the start like this:\r\n```\r\n\"<|endoftext|>ingly anticipation passations, out he didn't realize that any or products in order to stand on. As Eric's disappointment, he threw into those o-bag's fanware vugeless chainsas. Finally, Chris went on a hob, and grabbed the crurne cartocos juice!\"\r\n```\r\nI wonder if I can add a starting special token (sth like [CLS]) to my inputs and finetune gpt2 with this added vocabulary?\r\nI have asked this [here](https://github.com/huggingface/pytorch-transformers/issues/1145) though", "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", "> @LysandreJik Yeah I did that but It is not performing good, it generated the start like this:\r\n> \r\n> ```\r\n> \"<|endoftext|>ingly anticipation passations, out he didn't realize that any or products in order to stand on. As Eric's disappointment, he threw into those o-bag's fanware vugeless chainsas. Finally, Chris went on a hob, and grabbed the crurne cartocos juice!\"\r\n> ```\r\n> \r\n> I wonder if I can add a starting special token (sth like [CLS]) to my inputs and finetune gpt2 with this added vocabulary?\r\n> I have asked this [here](https://github.com/huggingface/pytorch-transformers/issues/1145) though\r\n\r\nHi Have you figured out this issue?" ]