Datasets:

---

DanielPFlorian

/

Transformers-Github-Issues

like 0

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

[]

[ "Hello! Have you put these models in `eval()` mode so as to deactivate the dropout modules?", "For completeness sake: did you train both models with the same random seed? Or are you just trying to evaluate models that you trained?\r\n\r\nMy go-to method is:\r\n\r\n```python\r\ndef set_seed(seed):\r\n \"\"\" Set all seeds to make results reproducible (deterministic mode).\r\n When seed is a false-y value or not supplied, disables deterministic mode. \"\"\"\r\n\r\n if seed:\r\n logging.info(f\"Running in deterministic mode with seed {seed}\")\r\n torch.manual_seed(seed)\r\n torch.cuda.manual_seed_all(seed)\r\n torch.backends.cudnn.deterministic = True\r\n torch.backends.cudnn.benchmark = False\r\n np.random.seed(seed)\r\n random.seed(seed)\r\n os.environ['PYTHONHASHSEED'] = str(seed)\r\n else:\r\n logging.info(f\"Running in non-deterministic mode\")\r\n```", "It does work when I put generated tensors into 2 models but doesn't when I put tensors I save before. Maybe I will rewrite the inference code with new code base and retrain the model.", "Are you sure your model is in evaluation mode? ", "yes, I already put it in evaluation mode", "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" ]

[]

[]

[ "Great, that looks good to me!", "Ok, merging, thanks @brian41005 " ]

[]

[]

[ "Hello! \r\n\r\n1 - Indeed, if you want to have several sequences of variable length in a single batch, you should pad the shorter sequences.\r\n\r\n\r\n2 - In the [`BertForNextSentencePrediction ` documentation](https://huggingface.co/transformers/model_doc/bert.html#bertfornextsentenceprediction) is written the following: \r\n\r\n`attention_mask`: Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`: `1` for tokens that are NOT MASKED, `0` for MASKED tokens. \r\n\r\nIf using the `attention_mask`, there should be no difference between a model's predictions of a sequence and its padded counterpart. \r\n\r\n\r\n3 - The segment ids padding indices can change according to the model. I believe it is `0` for most models, but `4` in the case of XLNet. You seem to be padding with `0` in your example, which is the way to go!", "Thank you!!! This is exactly what I was missing. \r\n\r\n```python3\r\nimport torch\r\nfrom pytorch_transformers import BertTokenizer, BertModel, BertForMaskedLM,BertForNextSentencePrediction\r\n\r\n# Load pre-trained model tokenizer (vocabulary)\r\ntokenizer = BertTokenizer.from_pretrained('bert-base-uncased')\r\n\r\n# Tokenized inputs\r\ntext1 = \"[CLS] Who was Jim ? [SEP] Jim Henson was a puppeteer [SEP]\"\r\ntext2 = \"[CLS] Who was Jim ? [SEP] Jim Henson was a puppeteer [SEP] [PAD] [PAD]\"\r\ntokenized_text1 = tokenizer.tokenize(text1)\r\ntokenized_text2 = tokenizer.tokenize(text2)\r\n\r\n# Convert token to vocabulary indices\r\nindexed_tokens1 = tokenizer.convert_tokens_to_ids(tokenized_text1)\r\nindexed_tokens2 = tokenizer.convert_tokens_to_ids(tokenized_text2)\r\n\r\n# Define sentence A and B indices associated to 1st and 2nd sentences \r\nsegments_ids1 = [0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1]\r\nsegments_ids2 = [0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0]\r\n\r\n#Attention Mask [1] over tokens, [0] over padding\r\nattention_mask = torch.FloatTensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0]])\r\n\r\n# Convert inputs to PyTorch tensors\r\ntokens_tensor1 = torch.tensor([indexed_tokens1])\r\ntokens_tensor2 = torch.tensor([indexed_tokens2])\r\nsegments_tensors1 = torch.tensor([segments_ids1])\r\nsegments_tensors2 = torch.tensor([segments_ids2])\r\n\r\n# Load pre-trained model (weights)\r\nmodel = BertForNextSentencePrediction.from_pretrained('bert-base-uncased')\r\nmodel.eval()\r\n\r\n# Predict is Next Sentence ?\r\npredictions1 = model(tokens_tensor1, segments_tensors1 )\r\npredictions2 = model(tokens_tensor2, segments_tensors2, attention_mask=attention_mask )\r\n\r\nprint(predictions1) #(tensor([[ 5.6165, -5.2786]], grad_fn=<AddmmBackward>),)\r\nprint(predictions2) #(tensor([[ 5.6165, -5.2786]], grad_fn=<AddmmBackward>),)\r\n```\r\n\r\nWith the attention mask now over tokens I get the same output without degradation. \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 } ]

[]

[ "Hi! By \"GPT-2 training\" two different methods can be understood: training from scratch, and fine-tuning. \r\n\r\nIf you're looking at training GPT-2 on a different language such as Portuguese, then training from scratch seems necessary. You could use the [language modeling finetuning example](https://github.com/huggingface/transformers/blob/master/examples/run_lm_finetuning.py) as a start, but please be aware that training such a language model from scratch takes a humongous amount of power and data, which would cost a lot. I can point you to [this issue](https://github.com/huggingface/transformers/issues/1356) which discusses training such a model on French.\r\n\r\nIf you're looking at training your model on programming languages that have a lot of overlapping vocabulary with English (say Python with a lot of documentation), maybe you could fine-tune the original GPT-2 to your dataset (still using the [lm finetuning example](https://github.com/huggingface/transformers/blob/master/examples/run_lm_finetuning.py)), but I'm not sure of the results. ", "I'm training Russian GPT-2 at the moment. [I've tried to make Readme useful.](https://github.com/mgrankin/ru_transformers)", "> If you're looking at training GPT-2 on a different language such as Portuguese, then training from scratch seems necessary. \r\n\r\nIt is definitely not necessary to start from scratch. I'd argue the opposite, it'd be useful to start with pre-trained GPT-2 even if you replacing the whole vocabulary (English -> Portuguese).", "Alright @mgrankin, that's good to know, thanks!", "> > If you're looking at training GPT-2 on a different language such as Portuguese, then training from scratch seems necessary.\r\n> \r\n> It is definitely not necessary to start from scratch. I'd argue the opposite, it'd be useful to start with pre-trained GPT-2 even if you replacing the whole vocabulary (English -> Portuguese).\r\n\r\nBut in $$$ terms it would still be closer to training from scratch than fine-tuning, right?", "> I'm training Russian GPT-2 at the moment. [I've tried to make Readme useful.](https://github.com/mgrankin/ru_transformers)\r\n\r\nThank you @mgrankin for sharing your steps. I plan to do the same for Hindi language.\r\nHow much is it costing you to train?", "> How much is it costing you to train?\r\n\r\nIt’s hard to tell overall cost because the training is in the process. I’ve got a workstation with 4 Titan RTX and I don’t use cloud GPUs at the moment. I use one GPU per model. The training already lasted about two weeks now and gpt2-medium gives me perplexity = 21 on my validation set. \r\n\r\nSince PyTorch 1.3 was released recently with TPU support I’m thinking of trying to use TPU to speed up the training. I will update the repo in the next few days in case of success. \r\n", "> But in $$$ terms it would still be closer to training from scratch than fine-tuning, right?\r\n\r\nActually, in terms of quality it would be great if somebody try to train GPT2 on Portuguese from scratch vs fine-tune from pretrained English model. My guess that fine-tuning is better is based on intuition that non-random weights could be reused. Also, English is probably the most resourceful language and WebText is a great dataset. If you can build dataset with same or better quality you can give it a shot and train GPT-2 from scratch. \r\n\r\nIn terms of money it should be way cheaper to fine-tune. But I will say that with confidence then I'll finish the Russian GPT-2.\r\n\r\n", "Thanks for the answer @mgrankin i'm anxious to see your results!", "> I'm training Russian GPT-2 at the moment. [I've tried to make Readme useful.](https://github.com/mgrankin/ru_transformers)\r\n\r\n@mgrankin Could you explain to me how you trained your model from scratch with BERT?\r\n\r\nI would like to train BERT from scratch for a textual base in PT-BR (8GB data). Is it possible to use the run_lm_finetuning.py code to perform this process without using the multi-language bert model?\r\n\r\nI already have a vocab.txt for the PT-BR base and I don't want to load initial weights.\r\n\r\nIs there any script or tutorial to perform this process step by step?", "Hi, I also have a repo which allows to train gpt-2 language model on non-english text with a custom BPE tokenizer. But it uses a different gpt-2 implementation so currently it's unable to use pre-trained GPT-2 (although a conversion script should be possible, because it's a port of original TF implementation). Here is the link https://github.com/lopuhin/transformer-lm", "Hello, this thread is what I'm looking (with the one about GPT-2 and BERT into French) for but I'm not sure I found the answer to my questions:\r\n- how long does it take to go through GPT-2 on non-english text?\r\n- what configuration of GPUs?\r\n- what size of corpus?\r\n\r\nMany thanks in advance for your answers!", "Why don't you use [CamemBERT](https://camembert-model.fr/) model, which is dedicated to French language? **It's available in HuggingFace's Transformers** too (since few days ago, so try out :D)! If you want absolutely to use GPT2 model, I can answer to you too!\r\n\r\n> Hello, this thread is what I'm looking (with the one about GPT-2 and BERT into French) for but I'm not sure I found the answer to my questions:\r\n> \r\n> * how long does it take to go through GPT-2 on non-english text?\r\n> * what configuration of GPUs?\r\n> * what size of corpus?\r\n> \r\n> Many thanks in advance for your answers!", "Hi @piegu, please do not post the same message in two issues (that are linked with one another)", "> Hi @piegu, please do not post the same message in two issues (that are linked with one another)\r\n\r\nHello @julien-c. Ok but then I have to update in this thread my question to French and Portuguese (same 3 questions about fine-tuning GPT-2 and BERT). Thank you.\r\n\r\n", "> Why don't you use [CamemBERT](https://camembert-model.fr/) model, which is dedicated to French language? **It's available in HuggingFace's Transformers** too (since few days ago, so try out :D)! If you want absolutely to use GPT2 model, I can answer to you too!\r\n\r\nThanks @TheEdoardo93. For sure I will test CamemBERT but it does not answer my 3 questions :-) Great if you can answer about GPT-2 at least. Thank you.", "Hi @nikhilno1 , Did you manage to train it on Hindi?", "Hi @GladiatorX, No I didn't. Life got in the way. :)\r\nWould you like to work on it together?", "@mgrankin Out of curiosity, how did you collect your 230 GB Russian dataset?\r\n I would love to do something similar for another language, and I'm looking for tips", "@BoxxiDev you can use something like a scraper/crawler like [Scrapy](https://scrapy.org/) (or something like it) on a russian site, and then you can use something like AWS Comprehend to get the language (or make a language detector yourself) and filter only Russian results.\r\n\r\nto get tons of data use some distributed scraper on a cloud service like AWS.", "@BoxxiDev Library projects have been working in Russia for a very long time, and they publish a torrent file with all the contents in fb2. [example](https://booktracker.org/viewtopic.php?t=1198)", "Hi @nikhilno1 , +1, \r\nDid you manage to train it on Hindi?\r\n\r\n", "> > If you're looking at training GPT-2 on a different language such as Portuguese, then training from scratch seems necessary.\r\n> \r\n> It is definitely not necessary to start from scratch. I'd argue the opposite, it'd be useful to start with pre-trained GPT-2 even if you replacing the whole vocabulary (English -> Portuguese).\r\n\r\n@mgrankin you say that it is not necessary to train from scratch, but assumed the vocabulary will not overlap (let's say English and Russian), how you do it?\r\n\r\nAlso someone else is talking about BERT based models (like the French model CamemBERT), but those models are [MASK] token based models, so it would need a different approach for text generation à la GPT-2", "@loretoparisi \r\n\r\nBy using progressive unfreezing. This's a technique from Transfer Learning. First, you freeze all layers and unfreeze only those layers that you expect to change the most - the embeddings and adjacent to the embeddings, you train them, you unfreeze a bit more layers, repeat. I’d advise taking a [course.fast.ai](https://course.fast.ai) course to be very comfortable with the concept.\r\n\r\nYou can look at the code [here](https://github.com/mgrankin/ru_transformers/blob/64d7a68e067737c35c7bf3986cb1845aaf54a163/tpu_lm_finetuning.py#L677).\r\n", "@mgrankin thank you, in the meanwhile I'm following this approach [BERT has a Mouth, and It Must Speak: BERT as a Markov Random Field Language Model](https://arxiv.org/abs/1902.04094)", "> \r\n> By using progressive unfreezing. This's a technique from Transfer Learning. First, you freeze all layers and unfreeze only those layers that you expect to change the most - the embeddings and adjacent to the embeddings, you train them, you unfreeze a bit more layers, repeat. I’d advise taking a [course.fast.ai](https://course.fast.ai) course to be very comfortable with the concept.\r\n> \r\n> You can look at the code [here](https://github.com/mgrankin/ru_transformers/blob/64d7a68e067737c35c7bf3986cb1845aaf54a163/tpu_lm_finetuning.py#L677).\r\n\r\nHi Mikhail. In your (great) code, you unfreeze groups of 3 layers (see [code](https://github.com/mgrankin/ru_transformers/blob/64d7a68e067737c35c7bf3986cb1845aaf54a163/tpu_lm_finetuning.py#L684) and below). There is a specific reason or it is the result of your tests? Thanks.\r\n\r\n`need_grads = set(flat[:i_start+args.unfreeze_level*3]) | set(flat[-(i_end+args.unfreeze_level*3):])`", "@piegu that's a heuristic, feel free to experiment with the number.", "> Hi @nikhilno1 , +1,\r\n> Did you manage to train it on Hindi?\r\n\r\nStarting it now. Let me know if you want to work together.", "> > Hi @nikhilno1 , +1,\r\n> > Did you manage to train it on Hindi?\r\n> \r\n> Starting it now. Let me know if you want to work together.\r\n\r\n@nikhilno1 Im interested to do this for tamil, were you able to figure our hindi ?", "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/1406?src=pr&el=h1) Report\n> Merging [#1406](https://codecov.io/gh/huggingface/transformers/pull/1406?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/63ed224b7c550ead5f9599187e665ded57ce80d4?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1406?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1406 +/- ##\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\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1406?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/1406?src=pr&el=footer). Last update [63ed224...193bbda](https://codecov.io/gh/huggingface/transformers/pull/1406?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

[]

[]

[ "Remaining CI failures are valid, they look to assume a `ijbn` ordering for all attention-based things, which no longer holds.\r\n\r\nI'm happy to add additional functionality to get these tests passing, but I'd like input on how you'd like that done (I'd lean to passing an optional `expected_attention_size` which is `[key_len, batch_size, num_heads]` by default, and checking that instead of assembling the expected sizes on-the-fly in the test(s))", "Please ignore above :) test errors were due to a missed transpose on my part (if attention outputs are returned, they need to be transposed from `bnij` to `ijbn` ordering to keep the interface from attention <-> the rest of the code the same as before.)", "This is a great work, thanks a lot @slayton58.\r\n\r\nCan you confirm this has no noticeable impact on downstream performances (in terms of evaluation metrics), for instance on your SQuAD tests?", "@thomwolf I have been testing against the config from https://github.com/huggingface/transformers/issues/947#issue-476001056 with seq-length=512, and obtained a consistent f1 score of 83 across both `ijbn` and `bnij` attention head orderings (also across fp32/fp16 O1/fp16 O2)\r\n\r\nWhen there's a PR issued for the changes in https://github.com/huggingface/transformers/issues/947#issuecomment-535989890 I'd be happy to go ahead and repro those numbers with this change if you'd like the additional security.", "Awesome, ok let's merge this then." ]

[]

[]

[ "Did you try using DistilBERT? Inference should be ~ 60% faster", "Turns out I wasn't using the using the gpu correctly. I moved the model and the inputs to gpu by doing `.to(device)` and it became 100x faster. Thanks for the suggestion." ]

[]

[]

[ "Hi! GPT-2, like all models in this library, directly inherit from pytorch's `nn.Module`, so you're free to finetune them or modify their parameters as you wish.", "> Hi! GPT-2, like all models in this library, directly inherit from pytorch's `nn.Module`, so you're free to finetune them or modify their parameters as you wish.\r\n\r\nThank you for your help!" ]

[ { "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" ]

[]

[]

[ "Hi! Those models actually are available, we just forgot to add them to the documentation :). Thanks for letting us know!" ]

[]

[]

[ "Great thanks!", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1400?src=pr&el=h1) Report\n> Merging [#1400](https://codecov.io/gh/huggingface/transformers/pull/1400?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/1400?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1400 +/- ##\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/1400?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_tf\\_xlm.py](https://codecov.io/gh/huggingface/transformers/pull/1400/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3hsbS5weQ==) | `88.23% <ø> (ø)` | :arrow_up: |\n| [transformers/modeling\\_tf\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/1400/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX2JlcnQucHk=) | `95.7% <ø> (ø)` | :arrow_up: |\n| [transformers/modeling\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/1400/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2JlcnQucHk=) | `88.17% <ø> (ø)` | :arrow_up: |\n| [transformers/modeling\\_tf\\_distilbert.py](https://codecov.io/gh/huggingface/transformers/pull/1400/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX2Rpc3RpbGJlcnQucHk=) | `96.61% <ø> (ø)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1400?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/1400?src=pr&el=footer). Last update [391db83...0c39053](https://codecov.io/gh/huggingface/transformers/pull/1400?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

[]

[]

[ "Hi! In Write With Transformer, we use the context to predict the following token. We then add that token to the initial context, to generate the following one. This way we can generate long sequences according to a given token.\r\n\r\nIn that app we stop generating tokens once we have reached a given time, or once we have seen an end of sentence token. We, therefore, don't generate variable length text suggestions, we just adjust the batch according to end tokens identified in our results.", "Thanks for the explanation!" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1398?src=pr&el=h1) Report\n> Merging [#1398](https://codecov.io/gh/huggingface/transformers/pull/1398?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/1398?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1398 +/- ##\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\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1398?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/1398?src=pr&el=footer). Last update [391db83...cd69bc9](https://codecov.io/gh/huggingface/transformers/pull/1398?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "👍 " ]

[]

[]

[]

[]

[]

[ "Thanks :)", "Your welcome, you are doing a great job!", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1396?src=pr&el=h1) Report\n> Merging [#1396](https://codecov.io/gh/huggingface/transformers/pull/1396?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/5c3b32d44d0164aaa9b91405f48e53cf53a82b35?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1396?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1396 +/- ##\n=======================================\n Coverage 84.69% 84.69% \n=======================================\n Files 84 84 \n Lines 12596 12596 \n=======================================\n Hits 10668 10668 \n Misses 1928 1928\n```\n\n\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1396?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/1396?src=pr&el=footer). Last update [5c3b32d...6b92911](https://codecov.io/gh/huggingface/transformers/pull/1396?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

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

[]

[ "The same issue here, using 4 RTX 2080Ti with Ubuntu 18.04.", "This issue exists as the `mask_tokens` function will sometimes replace `<s>` with a random word. Not sure whether `<s>` should be masked. A workaround would be adding a line \r\n```\r\nmasked_indices[:, 0] = 0 # tokenizer.bos_token_id\r\n```\r\nright below \r\n```\r\nmasked_indices = torch.bernoulli(torch.full(labels.shape, args.mlm_probability)).bool()\r\n```\r\nhttps://github.com/huggingface/transformers/blob/master/examples/run_lm_finetuning.py#L111.\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", "So the [CLS] and [SEP] are being masked during the training or not?" ]

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

[]

[ "Afaik the \"default\" loss function that gets computed if you pass your labels to `GPT2LMHeadModel` is `torch.nn.CrossEntropyLoss`. If you want to use a different loss function, can't you just grab the logits from the model and apply your own?\r\n\r\nSource:\r\nhttps://github.com/huggingface/transformers/blob/391db836ab7ed2ca61c51a7cf1b135b6ab92be58/transformers/modeling_gpt2.py#L539", "Unfortunately if I print a string inside the forward function, and then I run the training script, I don't get anything printed, so it seems like the training script is not using that function at all.", "Hello! The `GPT2LMHeadModel` does have a way to compute its own cross-entropy loss, but only when the `labels` are specified -> you're providing the values like so:\r\n```\r\nmodel(inputs, labels=inputs)\r\n```\r\nand the model takes care of shifting the inputs to calculate a causal language modeling loss on them with cross-entropy.\r\n\r\nIf you wish to use your own loss function, don't specify the labels and the model will return a tuple containing the language modeling logits as the first value.", "Hi,\r\n\r\nthanks for your answer.\r\nCan you tell me why a print statement inside the forward fuction of the GPT2LMHeadModel doesn't print anything when I run the run_lm_finetuning script ?\r\n\r\nWhich is the forward function I need to change?\r\n\r\nThanks.", "Where have you put your print statement? Do you have `transformers` installed in your environment or is it relying on the cloned repository? You could try to add a breakpoint and debug it to see which function calls are made and how the loss is calculated.\r\n\r\nOnce again, if you wish to use your own loss function, don't specify the labels and the model will return a tuple containing the language modeling logits as the first value.", "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! There is one big difference between BERT and GPT-2, in that BERT is trained using masked language modeling, whereas GPT-2 is trained using causal language modeling.\r\n\r\nDuring pre-training, BERT learns to predict masked words given a bi-directional context. GPT-2, on the other hand, learns to predict a word given only its left context. This is why GPT-2 is very good at text generation (it only needs the left-hand side context), while BERT isn't.\r\n\r\nGiven this, GPT-2 won't be able to do previous word prediction, as it does not handle the right-hand side context.", "If you want to train your own GPT-2 model to predict previous words, you could feed in your entire training set in reverse word order. Then GPT-2 would learn to predict text backwards, and that model would then be able to tell you what word should come before a piece of text." ]

[]

[]

[ "Hi! @thomwolf can correct me if I'm wrong, but I believe this keyword was changed to `output_hidden_states` in version 1.0.0.", "I can confirm what @LysandreJik suggests. The output of the embeddings is now also included as the first element. ", "Alright, thank you very much!" ]

[]

[]

[ "I've found it. It's a binary file. It's in ~/.cache/torch/transformers" ]

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

[]

[ "Hi! Yes, you understood the gist of it. The self-attention related to already computed tokens is not computed again.\r\n\r\nIn order to use the past, you would get the past from the model pass (I'm using GPT-2 in this example, XLNet would have `mems` instead of `past`):\r\n\r\n```py\r\nlogits, past = model(**inputs)\r\n```\r\n\r\nand you would then use the past on the following pass as follow:\r\n\r\n```py\r\nlogits, past = model(**inputs, past=past)\r\n```", "Thank you for your fast response @LysandreJik !\r\n\r\nNow it's very clear, but I have one more question :\r\n\r\nFor XLNet and TransfoXL, we need to use memory in order to not recompute previously generated token. This is ok when not using the memory for something else.\r\n\r\n**But what if the memory is already used for something else ?**\r\nLike we have a memory of 256 for XLNet, representing previous segments (or whatever), if we update the memory everytime a new token is generated, it means we are loosing part of the memory (and after generating 256 tokens, we will not be able to see anymore the memory of previous segment !).\r\n\r\n**Is there a way around this problem in the current API ?**", "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\nWe can accept this since it breaks lower versions of PyTorch.\r\nYou can just feed your mask as a FloatTensor (as indicated in the docstrings I think)." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1388?src=pr&el=h1) Report\n> Merging [#1388](https://codecov.io/gh/huggingface/transformers/pull/1388?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/5c3b32d44d0164aaa9b91405f48e53cf53a82b35?src=pr&el=desc) will **decrease** coverage by `0.16%`.\n> The diff coverage is `23.52%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1388?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1388 +/- ##\n==========================================\n- Coverage 84.69% 84.52% -0.17% \n==========================================\n Files 84 84 \n Lines 12596 12627 +31 \n==========================================\n+ Hits 10668 10673 +5 \n- Misses 1928 1954 +26\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1388?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/1388/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `61.17% <23.52%> (-10.05%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1388?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/1388?src=pr&el=footer). Last update [5c3b32d...1ba42ca](https://codecov.io/gh/huggingface/transformers/pull/1388?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "I am also working on reproducing the results reported in the Roberta paper and found two issues in this PR. One issue is explained in the comment above. The other issue is that it is required to insert two sep_tokens between question tokens and answer tokens for Roberta as implemented [here](https://github.com/huggingface/transformers/blob/master/transformers/tokenization_roberta.py#L101). Therefore, `max_tokens_for_doc` should be `max_seq_length - len(query_tokens) - 4`.", "In the Fairseq realisation of RoBERTa on Commonsense QA:\r\nhttps://github.com/pytorch/fairseq/tree/master/examples/roberta/commonsense_qa\r\nhttps://github.com/pytorch/fairseq/blob/master/examples/roberta/commonsense_qa/commonsense_qa_task.py\r\n\r\nThere is the only one sep_token between question and answer:\r\n`<s> Q: Where would I not want a fox? </s> A: hen house </s>`", "> In the Fairseq realisation of RoBERTa on Commonsense QA:\r\n> There is the only one sep_token between question and answer:\r\n> `<s> Q: Where would I not want a fox? </s> A: hen house </s>`\r\n\r\nThank you very much for your prompt reply. I did not know this. It seems to be appropriate to use single `sep_token` here because Commonsense QA is somewhat more similar to SQuAD than other tasks (e.g., GLUE).", "Thanks for this @vlarine! (and @ikuyamada)\r\n\r\nWould you agree to share the weights on our S3 as well?\r\n\r\nAlso, did you try with the same separators encoding scheme as the other RoBERTa models? \r\n`<s> Q: Where would I not want a fox? </s> </s> A: hen house </s>` – did the results differ significantly?", "No, I have not tried. But why there are two `</s>` tokens? I think more natural way is:\r\n`<s> Q: Where would I not want a fox? </s> <s> A: hen house </s>`", "@vlarine See this docstring in `fairseq`: https://github.com/pytorch/fairseq/pull/969/files\r\n\r\nDo you think you could try with this sep encoding scheme? Otherwise I'll do it in the next couple of days.\r\n\r\nI would like to merge your PR soon. Any way you can give me write access to your fork, cf. https://help.github.com/en/articles/committing-changes-to-a-pull-request-branch-created-from-a-fork – so that i can add commits on top of your PR?\r\n\r\n", "@vlarine @julien-c thanks for the amazing work! I can try it on SQuAD 2.0 and let you know if anything pops up there", "Nice work. I also tried to add roberta into run_squad.py several days ago. Hope that my implementation would be useful. [run_squad.py with roberta](https://github.com/erenup/pytorch-transformers/pull/4) ", "Folding this PR into #1386, which is close to being ready to being merged.\r\n\r\n@vlarine @ikuyamada @pminervini @erenup Can you guys please check it out?", "Closing in favor of #1386." ]

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

[]

[ "I see now that I missed something. The documentation uses the parameter 'lm_labels' but the correct parameter is just 'labels'. The documentation says that when this parameter is present, prediction logits will not be output, but this is incorrect. They are output regardless of the presence of 'labels'.", "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 / @LysandreJik / @VictorSanh / @julien-c Could you help review this PR? Thanks!", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1386?src=pr&el=h1) Report\n> Merging [#1386](https://codecov.io/gh/huggingface/transformers/pull/1386?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/be916cb3fb4579e278ceeaec11a6524662797d7f?src=pr&el=desc) will **decrease** coverage by `0.15%`.\n> The diff coverage is `21.21%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1386?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1386 +/- ##\n==========================================\n- Coverage 86.16% 86.01% -0.16% \n==========================================\n Files 91 91 \n Lines 13593 13626 +33 \n==========================================\n+ Hits 11713 11720 +7 \n- Misses 1880 1906 +26\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1386?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/1386/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `69.18% <21.21%> (-11.39%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1386?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/1386?src=pr&el=footer). Last update [be916cb...ee83f98](https://codecov.io/gh/huggingface/transformers/pull/1386?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Hi @thomwolf / @LysandreJik / @VictorSanh / @julien-c \r\n\r\nI have also run experiments using RoBERT large setting in original paper and reproduced their results,\r\n- **SQuAD v1.1**\r\n{\r\n \"exact\": 88.25922421948913,\r\n \"f1\": 94.43790487416292,\r\n \"total\": 10570,\r\n \"HasAns_exact\": 88.25922421948913,\r\n \"HasAns_f1\": 94.43790487416292,\r\n \"HasAns_total\": 10570\r\n}\r\n- **SQuAD v2.0**\r\n{\r\n \"exact\": 86.05238777057188,\r\n \"f1\": 88.99602665148535,\r\n \"total\": 11873,\r\n \"HasAns_exact\": 83.38394062078272,\r\n \"HasAns_f1\": 89.27965999208608,\r\n \"HasAns_total\": 5928,\r\n \"NoAns_exact\": 88.71320437342304,\r\n \"NoAns_f1\": 88.71320437342304,\r\n \"NoAns_total\": 5945,\r\n \"best_exact\": 86.5914259243662,\r\n \"best_exact_thresh\": -2.146007537841797,\r\n \"best_f1\": 89.43104312625539,\r\n \"best_f1_thresh\": -2.146007537841797\r\n}", "Awesome @stevezheng23. Can I push on top of your PR to change a few things before we merge?\r\n\r\n(We refactored the tokenizer to handle the encoding of sequence pairs, including special tokens. So we don't need to do it inside each example script anymore)", "@julien-c sure, please add changes in this PR if needed 👍 ", "@julien-c I've also upload the roberta large model finetuned on squad v2.0 data together with its prediction & evaluation results to public cloud storage https://storage.googleapis.com/mrc_data/squad/roberta.large.squad.v2.zip", "Can you check my latest commit @stevezheng23? Main change is that I removed the `add_prefix_space` for RoBERTa (which the RoBERTa authors don't use, as far as I know) which doesn't seem to make a significant difference.\r\n\r\n@thomwolf @LysandreJik this is ready for review.", "Everything looks good.\r\n\r\nAs for the `add_prefix_space` flag,\r\n- For `add_prefix_space=True`, I have run the experiment, the F1 score is around 89.4\r\n- For `add_prefix_space=False`, I have also run the experiment, the F1 score is around 88.2", "Great! Good job on reimplementing the cross-entropy loss when start/end positions are given.", "Look good to me.\r\nWe'll probably be able to simplify `utils_squad` a lot soon but that will be fine for now.\r\nDo you want to add your experimental results with RoBERTa in `examples/readme`, with a recommendation to use `add_prefix_space=True` (fyi it's the opposite for NER)?", "@julien-c do you want to add the roberta model finetuned on squad by @stevezheng23 in our library?", "Yep @thomwolf ", "@thomwolf I have updated README file as you suggested, you can merge this PR when you think it's good to go. BTW, it seems CI build is broken", "Ok thanks, I'll let @julien-c finish to handle this PR when he's back.", "> @julien-c I've also upload the roberta large model finetuned on squad v2.0 data together with its prediction & evaluation results to public cloud storage https://storage.googleapis.com/mrc_data/squad/roberta.large.squad.v2.zip\r\n\r\nHey @stevezheng23 !\r\n\r\nI just tried to reproduce your model with slightly different hyperparameters (`batch_size=2` and `gradient_accumulation=6` instead of `batch_size=12`), and I am currently getting worse results.\r\n\r\nResults with your model:\r\n\r\n```\r\n{\r\n \"exact\": 86.05238777057188,\r\n \"f1\": 88.99602665148535,\r\n \"total\": 11873,\r\n \"HasAns_exact\": 83.38394062078272,\r\n \"HasAns_f1\": 89.27965999208608,\r\n \"HasAns_total\": 5928,\r\n \"NoAns_exact\": 88.71320437342304,\r\n \"NoAns_f1\": 88.71320437342304,\r\n \"NoAns_total\": 5945\r\n}\r\n```\r\n\r\nResults with the model I trained, on the best checkpoint I was able to obtain after training for 8 epochs:\r\n\r\n```\r\n{\r\n \"exact\": 82.85184873241809,\r\n \"f1\": 85.85477834702593,\r\n \"total\": 11873,\r\n \"HasAns_exact\": 77.80026990553306,\r\n \"HasAns_f1\": 83.8147407750069,\r\n \"HasAns_total\": 5928,\r\n \"NoAns_exact\": 87.88898233809924,\r\n \"NoAns_f1\": 87.88898233809924,\r\n \"NoAns_total\": 5945\r\n}\r\n```\r\n\r\nYour hyperparameters:\r\n\r\n```\r\nNamespace(adam_epsilon=1e-08, cache_dir='', config_name='', device=device(type='cuda', index=0), do_eval=True, do_lower_case=False, do_train=True, doc_stride=128, eval_all_checkpoints=False, evaluate_during_training=False, fp16=False, fp16_opt_level='O1', gradient_accumulation_steps=1, learning_rate=1.5e-05, local_rank=0, logging_steps=50, max_answer_length=30, max_grad_norm=1.0, max_query_length=64, max_seq_length=512, max_steps=-1, model_name_or_path='roberta-large', model_type='roberta', n_best_size=20, n_gpu=1, no_cuda=False, null_score_diff_threshold=0.0, num_train_epochs=2.0, output_dir='output/squad/v2.0/roberta.large', overwrite_cache=False, overwrite_output_dir=False, per_gpu_eval_batch_size=12, per_gpu_train_batch_size=12, predict_file='data/squad/v2.0/dev-v2.0.json', save_steps=500, seed=42, server_ip='', server_port='', tokenizer_name='', train_batch_size=12, train_file='data/squad/v2.0/train-v2.0.json', verbose_logging=False, version_2_with_negative=True, warmup_steps=500, weight_decay=0.01)\r\n```\r\n\r\nMy hyperparameters:\r\n\r\n```\r\nNamespace(adam_epsilon=1e-08, cache_dir='', config_name='', device=device(type='cuda'), do_eval=True, do_lower_case=False, do_train=True, doc_stride=128, eval_all_checkpoints=False, evaluate_during_training=False, fp16=False, fp16_opt_level='O1', gradient_accumulation_steps=6, learning_rate=1.5e-05, local_rank=-1, logging_steps=50, max_answer_length=30, max_grad_norm=1.0, max_query_length=64, max_seq_length=512, max_steps=-1, model_name_or_path='roberta-large', model_type='roberta', n_best_size=20, n_gpu=1, no_cuda=False, null_score_diff_threshold=0.0, num_train_epochs=8.0, output_dir='../roberta.large.squad2.v1p', overwrite_cache=False, overwrite_output_dir=False, per_gpu_eval_batch_size=2, per_gpu_train_batch_size=2, predict_file='/home/testing/drive/invariance//workspace/data/squad/dev-v2.0.json', save_steps=500, seed=42, server_ip='', server_port='', tokenizer_name='', train_batch_size=2, train_file='/home/testing/drive/invariance//workspace/data/squad/train-v2.0.json', verbose_logging=False, version_2_with_negative=True, warmup_steps=500, weight_decay=0.01)\r\n```\r\n\r\nDo you have any ideas why this is happening ?\r\n\r\nOne thing that may be happening is that, when using `max_grad_norm` and `gradient_accumulation=n`, the clipping of the gradient norm seems to be done `n` times rather than just 1, but I need to look deeper into this.\r\n\r\nI'd like to see what happens without the need of gradient accumulation - anyone with a spare TPU to share? 😬", "> Ok thanks, I'll let @julien-c finish to handle this PR when he's back.\r\n\r\nthanks, @thomwolf ", "@pminervini I haven't tried out using `max_grad_norm` and `gradient_accumulation=n` combination before. One thing you could pay attention to is that the checkpoint is trained with `add_prefix_space=True` for RoBERTa tokenizer.", "@stevezheng23 if you look at it, the `max_grad_norm` is performed on all the gradients in the accumulation - I think it should be done just before the `optimizer.step()` call.\r\n\r\nhttps://github.com/huggingface/transformers/blob/master/examples/run_squad.py#L163\r\n\r\n@thomwolf what do you think ? should I go and do a PR ?", "@LysandreJik just significantly rewrote our SQuAD integration in https://github.com/huggingface/transformers/pull/1984 so we were holding out on merging this.\r\n\r\nDoes anyone here want to revisit this PR with the changes from #1984? Otherwise, we'll do it, time permitting.", "cool, I'm willing to revisit it. I will take a look at your changes and tansformers' recent updates today （have been away from the Master branch for some time😊).", "> > @julien-c I've also upload the roberta large model finetuned on squad v2.0 data together with its prediction & evaluation results to public cloud storage https://storage.googleapis.com/mrc_data/squad/roberta.large.squad.v2.zip\r\n> \r\n> Hey @stevezheng23 !\r\n> \r\n> I just tried to reproduce your model with slightly different hyperparameters (`batch_size=2` and `gradient_accumulation=6` instead of `batch_size=12`), and I am currently getting worse results.\r\n> Your hyperparameters:\r\n> \r\n> ```\r\n> Namespace(adam_epsilon=1e-08, cache_dir='', config_name='', device=device(type='cuda', index=0), do_eval=True, do_lower_case=False, do_train=True, doc_stride=128, eval_all_checkpoints=False, evaluate_during_training=False, fp16=False, fp16_opt_level='O1', gradient_accumulation_steps=1, learning_rate=1.5e-05, local_rank=0, logging_steps=50, max_answer_length=30, max_grad_norm=1.0, max_query_length=64, max_seq_length=512, max_steps=-1, model_name_or_path='roberta-large', model_type='roberta', n_best_size=20, n_gpu=1, no_cuda=False, null_score_diff_threshold=0.0, num_train_epochs=2.0, output_dir='output/squad/v2.0/roberta.large', overwrite_cache=False, overwrite_output_dir=False, per_gpu_eval_batch_size=12, per_gpu_train_batch_size=12, predict_file='data/squad/v2.0/dev-v2.0.json', save_steps=500, seed=42, server_ip='', server_port='', tokenizer_name='', train_batch_size=12, train_file='data/squad/v2.0/train-v2.0.json', verbose_logging=False, version_2_with_negative=True, warmup_steps=500, weight_decay=0.01)\r\n> ```\r\n> \r\n> My hyperparameters:\r\n> \r\n> ```\r\n> Namespace(adam_epsilon=1e-08, cache_dir='', config_name='', device=device(type='cuda'), do_eval=True, do_lower_case=False, do_train=True, doc_stride=128, eval_all_checkpoints=False, evaluate_during_training=False, fp16=False, fp16_opt_level='O1', gradient_accumulation_steps=6, learning_rate=1.5e-05, local_rank=-1, logging_steps=50, max_answer_length=30, max_grad_norm=1.0, max_query_length=64, max_seq_length=512, max_steps=-1, model_name_or_path='roberta-large', model_type='roberta', n_best_size=20, n_gpu=1, no_cuda=False, null_score_diff_threshold=0.0, num_train_epochs=8.0, output_dir='../roberta.large.squad2.v1p', overwrite_cache=False, overwrite_output_dir=False, per_gpu_eval_batch_size=2, per_gpu_train_batch_size=2, predict_file='/home/testing/drive/invariance//workspace/data/squad/dev-v2.0.json', save_steps=500, seed=42, server_ip='', server_port='', tokenizer_name='', train_batch_size=2, train_file='/home/testing/drive/invariance//workspace/data/squad/train-v2.0.json', verbose_logging=False, version_2_with_negative=True, warmup_steps=500, weight_decay=0.01)\r\n> ```\r\n> \r\n> Do you have any ideas why this is happening ?\r\n\r\n\r\nYou're using num_train_epochs=8 instead of 2, which makes the learning rate decay more slowly. Maybe that is causing the difference?", "Regarding `max_grad_norm` - RoBERTa doesn't use gradient clipping, so the `max_grad_norm` changes aren't strictly necessary here\r\n\r\nRoBERTa also uses `adam_epsilon=1e-06` as I understand, but I'm not sure if it would change the results here", "Hi @stevezheng23 @julien-c @thomwolf @ethanjperez , I updated the run squad with roberta in #2173 \r\n based on #1984 and #1386. Could you please help to review it? Thank you very much.", "Closed in favor of #2173 which should be merged soon. \r\n\r\n" ]

[]

[]

[ "Great addition. I feel like using enums would be especially helpful for the truncating strategy, indeed.", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1385?src=pr&el=h1) Report\n> Merging [#1385](https://codecov.io/gh/huggingface/transformers/pull/1385?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/5c3b32d44d0164aaa9b91405f48e53cf53a82b35?src=pr&el=desc) will **decrease** coverage by `0.07%`.\n> The diff coverage is `41.66%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1385?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1385 +/- ##\n==========================================\n- Coverage 84.69% 84.61% -0.08% \n==========================================\n Files 84 84 \n Lines 12596 12610 +14 \n==========================================\n+ Hits 10668 10670 +2 \n- Misses 1928 1940 +12\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1385?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1385/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl91dGlscy5weQ==) | `87.73% <41.66%> (-2.46%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1385?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/1385?src=pr&el=footer). Last update [5c3b32d...9e136ff](https://codecov.io/gh/huggingface/transformers/pull/1385?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "I have reviewed this PR and It looks good to me. Thank you! @julien-c . I added two lines of comments above. Hope they are useful. Thank you. ", "Merged in, and superseded by, #1384" ]

[]

[]

[ "I think we should drop the `always_truncate` param, and just set it to `True` iff `max_length is not None`", "Other than that I like it.", "As seen with @julien-c , `always_truncate` really should be enabled by default when a `max_length` is specified." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1383?src=pr&el=h1) Report\n> Merging [#1383](https://codecov.io/gh/huggingface/transformers/pull/1383?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/1c5079952f5f10eeac4cb6801b4fd1f36b0eff73?src=pr&el=desc) will **increase** coverage by `1.63%`.\n> The diff coverage is `92.38%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1383?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1383 +/- ##\n==========================================\n+ Coverage 83.79% 85.42% +1.63% \n==========================================\n Files 84 91 +7 \n Lines 12587 13464 +877 \n==========================================\n+ Hits 10547 11502 +955 \n+ Misses 2040 1962 -78\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1383?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/1383/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `80.57% <ø> (+15.1%)` | :arrow_up: |\n| [transformers/tests/modeling\\_tf\\_gpt2\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1383/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2dwdDJfdGVzdC5weQ==) | `94.73% <0%> (ø)` | :arrow_up: |\n| [transformers/tests/modeling\\_tf\\_common\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1383/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2NvbW1vbl90ZXN0LnB5) | `95.38% <100%> (+7.88%)` | :arrow_up: |\n| [transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/1383/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX29wZW5haS5weQ==) | `81.75% <100%> (+1.35%)` | :arrow_up: |\n| [transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1383/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2ZpbGVfdXRpbHMucHk=) | `74.17% <100%> (ø)` | :arrow_up: |\n| [transformers/modeling\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/1383/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2F1dG8ucHk=) | `51.85% <20%> (-2.1%)` | :arrow_down: |\n| [transformers/configuration\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/1383/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2NvbmZpZ3VyYXRpb25fYXV0by5weQ==) | `58.82% <33.33%> (-2.47%)` | :arrow_down: |\n| [transformers/tokenization\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/1383/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl9hdXRvLnB5) | `67.64% <33.33%> (-3.33%)` | :arrow_down: |\n| [transformers/tokenization\\_ctrl.py](https://codecov.io/gh/huggingface/transformers/pull/1383/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl9jdHJsLnB5) | `83.6% <83.6%> (ø)` | |\n| [transformers/configuration\\_ctrl.py](https://codecov.io/gh/huggingface/transformers/pull/1383/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2NvbmZpZ3VyYXRpb25fY3RybC5weQ==) | `88.88% <88.88%> (ø)` | |\n| ... and [18 more](https://codecov.io/gh/huggingface/transformers/pull/1383/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1383?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/1383?src=pr&el=footer). Last update [1c50799...d9e60f4](https://codecov.io/gh/huggingface/transformers/pull/1383?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Ok for merge", "Thanks for adding this! I'm currently doing some experiments with the CTRL model, and I've a question about the tokenization:\r\n\r\n```bash\r\ntokenizer.tokenize(\"Munich and Berlin are nice cities.\") \r\nOut[6]: ['m@@', 'unic@@', 'h', 'and', 'ber@@', 'lin', 'are', 'nice', 'cities', '.']\r\n```\r\n\r\nDo you have any idea, why the output returns lowercased tokens only - `Berlin` and `Munich` do both appear in the vocab file (cased, and the splitting of `Munich` looks really weird 😅).", "Yes, we are aware of the issue.\r\n\r\nWe are fixing this problem in #1480." ]

[]

[]

[ "Can't reproduce this on master now. Seems to be fixed.", "Thanks a lot. It seems to be fixed. Now I get `'[SEP]'` and `' [SEP]'` consecutively with the first and the second command above. So we can close this issue." ]

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

[]

[ "https://github.com/pytorch/fairseq/blob/master/examples/roberta/README.pretraining.md", "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", "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", "When I put new --config_name and --tokenizer_name. It shows me that \r\njson.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0)\r\nAnyone can help me?" ]

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

[]

[ "Hey @malmaud I think this #1196 can help you. The Roberta/GPT2 tokenizer expect a space to start. Without that, it sounds like you'll get strange behaviors.\r\n\r\nTo get the same output, in your first example, change it to \r\n```\r\nt.tokenize(\"mystery\", add_prefix_space=True)\r\n['Ġmystery']\r\n```", "That does work, thanks. I'm still confused why this doesn't work, though:\r\n\r\n```\r\nt.tokenize(\"<s> mystery </s>\")\r\n```\r\n\r\ngives `['<s>', 'my', 'stery', '</s>']`", "Hey @malmaud, spent some time going through the source code. So like above this gives the correct result:\r\n```\r\nt.tokenize(\"mystery\", add_prefix_space=True)\r\n['Ġmystery']\r\n```\r\nHowever\r\n```\r\nt.tokenizer(\" mystery\")\r\n['my', 'stery']\r\n```\r\nI thought these should be doing the same thing. In the tokenization_gpt2.py file, it says:\r\n```\r\nif add_prefix_space:\r\n text = ' ' + text\r\n```\r\nThis should give the same results in both files then however when I add a print(text) statement before and after that I noticed I got these results. (using your example now)\r\n```\r\nt.tokenize(\"<s> mystery\")\r\nmystery\r\nmystery\r\n['<s>', 'my', 'stery']\r\n\r\nt.tokenize(\"<s> mystery\", add_prefix_space=True)\r\nmystery\r\n mystery\r\n['<s>', 'Ġmystery']\r\n```\r\n\r\nThis means that even though we are putting a single word in with a leading space, something in the preprocessing is getting rid of the initial space(s). So we need to use the add_prefix_space=True in order to get the space back or else the function won't be using the string we are expecting it will be using.", "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" ]

[]

[]

[ "so, how to solve this problem?", "Should be solved on master and the latest release." ]

[ { "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" ]

[]

[]

[]

[]

[]

[ "Hi! The TensorFlow components are only available when you have TF2 installed on your system. Could you please check that you have it in the environment in which you're running your code?", "It worked. Thanks" ]

[ { "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" ]

[]

[]

[ "Amazing!" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1372?src=pr&el=h1) Report\n> Merging [#1372](https://codecov.io/gh/huggingface/transformers/pull/1372?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/fd97761c5a977fd22df789d2851cf57c7c9c0930?src=pr&el=desc) will **increase** coverage by `1.42%`.\n> The diff coverage is `83.33%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1372?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1372 +/- ##\n==========================================\n+ Coverage 84.74% 86.16% +1.42% \n==========================================\n Files 91 91 \n Lines 13593 13593 \n==========================================\n+ Hits 11519 11713 +194 \n+ Misses 2074 1880 -194\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1372?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1372/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl91dGlscy5weQ==) | `91.43% <83.33%> (ø)` | :arrow_up: |\n| [transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/1372/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX29wZW5haS5weQ==) | `81.75% <0%> (+1.35%)` | :arrow_up: |\n| [transformers/modeling\\_ctrl.py](https://codecov.io/gh/huggingface/transformers/pull/1372/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2N0cmwucHk=) | `95.45% <0%> (+2.27%)` | :arrow_up: |\n| [transformers/modeling\\_xlnet.py](https://codecov.io/gh/huggingface/transformers/pull/1372/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3hsbmV0LnB5) | `73.28% <0%> (+2.46%)` | :arrow_up: |\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/1372/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `80.57% <0%> (+15.1%)` | :arrow_up: |\n| [transformers/tests/modeling\\_tf\\_common\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1372/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2NvbW1vbl90ZXN0LnB5) | `96.8% <0%> (+17.02%)` | :arrow_up: |\n| [transformers/modeling\\_tf\\_pytorch\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1372/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3B5dG9yY2hfdXRpbHMucHk=) | `92.95% <0%> (+83.09%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1372?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/1372?src=pr&el=footer). Last update [fd97761...ba6f2d6](https://codecov.io/gh/huggingface/transformers/pull/1372?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "We'll handle this by dropping python2 support in the next release (and using flake8) cc @aaugustin " ]

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

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1371?src=pr&el=h1) Report\n> Merging [#1371](https://codecov.io/gh/huggingface/transformers/pull/1371?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/ae50ad91ea2fedb64ecd2e7c8e2d0d4778dc03aa?src=pr&el=desc) will **increase** coverage by `0.97%`.\n> The diff coverage is `85.71%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1371?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1371 +/- ##\n==========================================\n+ Coverage 83.76% 84.74% +0.97% \n==========================================\n Files 84 84 \n Lines 12596 12559 -37 \n==========================================\n+ Hits 10551 10643 +92 \n+ Misses 2045 1916 -129\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1371?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/1371/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX29wZW5haS5weQ==) | `80.41% <100%> (ø)` | :arrow_up: |\n| [transformers/modeling\\_xlnet.py](https://codecov.io/gh/huggingface/transformers/pull/1371/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3hsbmV0LnB5) | `72.02% <100%> (+0.77%)` | :arrow_up: |\n| [transformers/modeling\\_gpt2.py](https://codecov.io/gh/huggingface/transformers/pull/1371/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2dwdDIucHk=) | `83.88% <100%> (-0.11%)` | :arrow_down: |\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/1371/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `71.01% <100%> (+5.54%)` | :arrow_up: |\n| [transformers/modeling\\_distilbert.py](https://codecov.io/gh/huggingface/transformers/pull/1371/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2Rpc3RpbGJlcnQucHk=) | `95.8% <100%> (-0.03%)` | :arrow_down: |\n| [transformers/modeling\\_transfo\\_xl.py](https://codecov.io/gh/huggingface/transformers/pull/1371/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RyYW5zZm9feGwucHk=) | `75.16% <100%> (ø)` | :arrow_up: |\n| [transformers/modeling\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/1371/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2JlcnQucHk=) | `88.41% <100%> (+0.23%)` | :arrow_up: |\n| [transformers/modeling\\_xlm.py](https://codecov.io/gh/huggingface/transformers/pull/1371/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3hsbS5weQ==) | `88.36% <100%> (-0.07%)` | :arrow_down: |\n| [transformers/modeling\\_transfo\\_xl\\_utilities.py](https://codecov.io/gh/huggingface/transformers/pull/1371/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RyYW5zZm9feGxfdXRpbGl0aWVzLnB5) | `54.16% <37.5%> (+0.27%)` | :arrow_up: |\n| [transformers/modeling\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1371/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3V0aWxzLnB5) | `92.57% <90%> (-0.12%)` | :arrow_down: |\n| ... and [7 more](https://codecov.io/gh/huggingface/transformers/pull/1371/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1371?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/1371?src=pr&el=footer). Last update [ae50ad9...716d783](https://codecov.io/gh/huggingface/transformers/pull/1371?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "Unstale. ", "Would feel syntactically cleaner if we could do `ACT2FN.gelu()` instead of a dict (also gives some IDE goodness like autocomplete) (I guess through a class or namespace or something), what do you guys think?", "> Would feel syntactically cleaner if we could do `ACT2FN.gelu()` instead of a dict (also gives some IDE goodness like autocomplete) (I guess through a class or namespace or something), what do you guys think?\r\n\r\nSounds good but note that this is not something I introduced. The ACT2FN dict already existed, but wasn't used consistently it seemed.", "Ah yeah, I see. Would you want to do this change, if you have the time/bandwidth? (+ rebasing on current master so we can merge easily?)", "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", "AFAICT, this has been done by @sshleifer on master. Re-open if necessary!" ]

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

[]

[ "Would definitely love to see an implementation of ALBERT added to this repository. Just for completeness:\r\n\r\n* paper: https://arxiv.org/abs/1909.11942\r\n* reddit: https://www.reddit.com/r/MachineLearning/comments/d9tdfo/albert_a_lite_bert_for_selfsupervised_learning_of/\r\n* medium: https://medium.com/syncedreview/googles-albert-is-a-leaner-bert-achieves-sota-on-3-nlp-benchmarks-f64466dd583\r\n\r\nThat said, it could be even more interesting to implement the core improvements (factorized embedding parameterization, cross-layer parameter sharing) from ALBERT in (some?/all?) other transformers as optional features?\r\n", "Knowing how fast the team works, I would expect ALBERT to be implemented quite soon. That being said, I haven't had time to read the ALBERT paper yet, so it might be more difficult than previous BERT iterations such as distilbert and RoBERTa.", "I think ALBERT is very cool! Expect...", "And in pytorch (using code from this repo and weights from brightmart) https://github.com/lonePatient/albert_pytorch", "Any Update on the progress?", "The ALBERT paper will be presented at ICLR in April 2020. From what I last heard, the huggingface team has been talking with the people over at Google AI to share the details of the model, but I can imagine that the researchers rather wait until the paper has been presented. One of those reasons being that they want to get citations from their ICLR talk rather than an arXiv citation which, in the field, is \"worth less\" than a big conference proceeding. \r\n\r\nFor now, just be patient. I am sure that the huggingface team will have a big announcement (follow their Twitter/LinkedIn channels) with a new version bump. No need to keep bumping this topic.", "https://github.com/interviewBubble/Google-ALBERT", "The official code and models got released :slightly_smiling_face: \r\nhttps://github.com/google-research/google-research/tree/master/albert ", "[WIP]\r\nALBERT in tensorflow 2.0\r\nhttps://github.com/kamalkraj/ALBERT-TF2.0\r\n", "https://github.com/lonePatient/albert_pytorch\r\n\r\nDataset: MNLI\r\nModel: ALBERT_BASE_V2\r\nDev accuracy : 0.8418\r\n\r\nDataset: SST-2\r\nModel: ALBERT_BASE_V2\r\nDev accuracy :0.926", "PR was created, see here:\r\n\r\nhttps://github.com/huggingface/transformers/pull/1683", "> [WIP]\r\n> ALBERT in tensorflow 2.0\r\n> https://github.com/kamalkraj/ALBERT-TF2.0\r\n\r\nVerison 2 weights added.\r\nSupport for SQuAD 1.1 and 2.0 added. \r\nReproduces the same results from paper. From my experiments, ALBERT model is very sensitive to hyperparameter like Batch Size. FineTuning using AdamW as Default in Original Repo. AdamW performs better than LAMB on Model finetuning. ", "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/1369?src=pr&el=h1) Report\n> Merging [#1369](https://codecov.io/gh/huggingface/transformers/pull/1369?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/ae50ad91ea2fedb64ecd2e7c8e2d0d4778dc03aa?src=pr&el=desc) will **increase** coverage by `0.92%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1369?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1369 +/- ##\n==========================================\n+ Coverage 83.76% 84.69% +0.92% \n==========================================\n Files 84 84 \n Lines 12596 12596 \n==========================================\n+ Hits 10551 10668 +117 \n+ Misses 2045 1928 -117\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1369?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_xlnet.py](https://codecov.io/gh/huggingface/transformers/pull/1369/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3hsbmV0LnB5) | `72.14% <0%> (+0.89%)` | :arrow_up: |\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/1369/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `71.22% <0%> (+5.75%)` | :arrow_up: |\n| [transformers/tests/modeling\\_tf\\_common\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1369/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2NvbW1vbl90ZXN0LnB5) | `95% <0%> (+7.5%)` | :arrow_up: |\n| [transformers/modeling\\_tf\\_pytorch\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1369/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3B5dG9yY2hfdXRpbHMucHk=) | `76.92% <0%> (+66.43%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1369?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/1369?src=pr&el=footer). Last update [ae50ad9...d1176d5](https://codecov.io/gh/huggingface/transformers/pull/1369?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Great, thanks for updating the README!" ]

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

[]

[ "Hey @mandavachetana its not just a google colab thing. Take a look here #1375 You need to make sure you are using tensorflow 2.0 and it should work.", "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 check the two migration guides, they explain all the differences:\r\n- https://github.com/huggingface/transformers#Migrating-from-pytorch-transformers-to-transformers\r\n- https://github.com/huggingface/transformers#migrating-from-pytorch-pretrained-bert-to-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" ]

[]

[]

[ "Sorry, I will fix this " ]

[ { "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 } ]

[]

[ "What's your precise question?", "> What's your precise question?\r\n\r\nthe self.padding_idx's meaning in modeling_roberta.py", "It's the position of the padding vector. It's not unique to RoBERTa but far more general, especially for embeddings. Take a look at [the PyTorch documentation](https://pytorch.org/docs/stable/nn.html#embedding).", "> It's the position of the padding vector. It's not unique to RoBERTa but far more general, especially for embeddings. Take a look at [the PyTorch documentation](https://pytorch.org/docs/stable/nn.html#embedding).\r\n\r\nI know that, but I confuse about why there is 1 and the \\<s\\> is 0, is it ignore and why the max_position_embeddings size is 512+2=514?", "Because that's their index [in the vocab](https://s3.amazonaws.com/models.huggingface.co/bert/roberta-base-vocab.json). The max_position_embeddings size is indeed 514, I'm not sure why. The tokenizer seems to handle text correctly with a max of 512. Perhaps someone of the developers can help with that. I would advise you to change the title of your topic.\r\n\r\nhttps://github.com/huggingface/transformers/blob/ae50ad91ea2fedb64ecd2e7c8e2d0d4778dc03aa/transformers/tokenization_roberta.py#L84-L85", "@LysandreJik can chime in if I’m wrong, but afaik `max_position_embeddings` is just the name of the variable that we use to encode the size of the embedding matrix. Max_len is correctly set to 512.", "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", "Answer here in case anyone from the future is curious: https://github.com/pytorch/fairseq/issues/1187", "> Answer here in case anyone from the future is curious: [pytorch/fairseq#1187](https://github.com/pytorch/fairseq/issues/1187)\r\n\r\n@morganmcg1 Tks for this, was getting all kinds of CUDA errors because i setted `max_position_embeddings=512`, now that i setted 514 it's running ok..." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1362?src=pr&el=h1) Report\n> Merging [#1362](https://codecov.io/gh/huggingface/transformers/pull/1362?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/a6a6d9e6382961dc92a1a08d1bab05a52dc815f9?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1362?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1362 +/- ##\n=======================================\n Coverage 84.69% 84.69% \n=======================================\n Files 84 84 \n Lines 12596 12596 \n=======================================\n Hits 10668 10668 \n Misses 1928 1928\n```\n\n\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1362?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/1362?src=pr&el=footer). Last update [a6a6d9e...60f7916](https://codecov.io/gh/huggingface/transformers/pull/1362?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "👍 " ]

[]

[]

[ "cf https://github.com/huggingface/transformers/issues/1193#issuecomment-534740929", "Ok, as discussed let's copy this script to the `examples/distillation` folder and keep `run_squad` barebone for now as it's going to evolve in the short term.", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1361?src=pr&el=h1) Report\n> Merging [#1361](https://codecov.io/gh/huggingface/transformers/pull/1361?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/2dc8cb87341223e86220516951bb4ad84f880b4a?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1361?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1361 +/- ##\n=======================================\n Coverage 84.69% 84.69% \n=======================================\n Files 84 84 \n Lines 12596 12596 \n=======================================\n Hits 10668 10668 \n Misses 1928 1928\n```\n\n\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1361?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/1361?src=pr&el=footer). Last update [2dc8cb8...b4df865](https://codecov.io/gh/huggingface/transformers/pull/1361?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "no squash @VictorSanh? 😬" ]

[]

[]

[ "I'm not sure that I understand. As you say, you can see it implemented in the run_squad example. What else would you like? ", "Hello Bram,\r\n\r\nI mean I want to apply it with a sequence classification task like BertForSequenceClassification, for example, versus what is being done in squad.\r\n\r\nI don't think it should be too hard but I'm not exactly sure how a long document that is being chunked gets trained. Do we ignore the fact that these are chunks of the same document and just treat them as independent docs? Or do we do some sort of trick to join the tokens/embeddings with the first chunk? \r\n\r\nHow would this be implemented for sequence classification?", "I quickly glared over the `convert_examples_to_features` function, and it seems that given some stride different parts are used as input. So, yes, as far as I can see they are treated as independent docs.\r\n\r\nhttps://github.com/huggingface/transformers/blob/ae50ad91ea2fedb64ecd2e7c8e2d0d4778dc03aa/examples/utils_squad.py#L189-L397\r\n\r\n", "isn't there a way to deal with long documents without ignoring the fact that the chunks represent the same doc? \r\n\r\nMaybe something along the lines of https://finetune.indico.io/chunk.html?highlight=long or https://explosion.ai/blog/spacy-pytorch-transformers#batching", "After a first look, I don't see how `spacy-pytorch-transformers` does anything special rather than processing a document sentence-per-sentence. `finetune`'s approach might be what you after (taking the mean over all the slided windows), but as always: \"a mean is just a mean\", so the question remains how representative it is of the whole document. I am not saying that slicing is _better_ by any means, but averaging can distort \"real\" values greatly.", "Yeah I see your point. I'm starting to think that maybe trying out chunking with a couple of different strides and maybe at inference time taking a voting approach would be a better option.\r\n\r\nIn any case, thank you for your feedback!", "I agree that that might be the more efficient approach. No worries, thanks for the interesting question. If you think it's okay the question, please close it so it's easy to keep track of all open issues.", "Hi, just to let you know that there is an option to manage strides in the `encode_plus` method. It handles special tokens and returns the overflowing elements in the `overflowing_tokens` field of the returned dictionary." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1359?src=pr&el=h1) Report\n> Merging [#1359](https://codecov.io/gh/huggingface/transformers/pull/1359?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/ca559826c4188be8713e46f191ddf5f379c196e7?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1359?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1359 +/- ##\n=======================================\n Coverage 84.73% 84.73% \n=======================================\n Files 84 84 \n Lines 12573 12573 \n=======================================\n Hits 10654 10654 \n Misses 1919 1919\n```\n\n\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1359?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/1359?src=pr&el=footer). Last update [ca55982...9478590](https://codecov.io/gh/huggingface/transformers/pull/1359?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Thanks @dennymarcels!" ]

[ { "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": "" } ]

[]

[ "What is it that you can contribute? The only (yet impressive) thing that is going on is language modeling. Can you contribute a pre-trained French model for one of the frameworks? That's (as far as I know) the only way to contribute. ", "Thanks Bram, I’m going to investigate what the cost could be for XLNet on clevergrid https://www.clevergrid.io/?pk_campaign=ga-gpu-1&pk_source=adwords&pk_medium=sem&pk_content=gpuasaservicefr&gclid=CjwKCAjwibzsBRAMEiwA1pHZrvm8ozRMrbcDR7YoYiKqsq6gEnPo9AecJwjKzBxa8L-4_hB6ny4uARoCwfMQAvD_BwE\n\nEnvoyé de mon iPad\n\n> Le 28 sept. 2019 à 09:44, Bram Vanroy <[email protected]> a écrit :\n> \n> What is it that you can contribute? The only (yet impressive) thing that is going on is language modeling. Can you contribute a pre-trained French model for one of the frameworks? That's (as far as I know) the only way to contribute.\n> \n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub, or mute the thread.\n", "Hi all,\r\nWe (ovh) are open to calculate it for free.", "Not sure if a new French language model is still necessary after Camembert has been introduced.", "That's awesome news, @jqueguiner! Let us know if we can help.\r\n\r\n@BramVanroy To work well with Write With Transformer, we would want more like a FR-pretrained GPT-2-like model. CamemBERT wouldn't do on generation out of the box.\r\n\r\nSee also the more specific issue: https://github.com/huggingface/transformers/issues/1356", "For generation CamemBERT is of no use I think...\n\nEnvoyé de mon iPad\n\n> Le 22 nov. 2019 à 14:02, Bram Vanroy <[email protected]> a écrit :\n> \n> \n> Not sure if a new French language model is still necessary after Camembert has been introduced.\n> \n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub, or unsubscribe.\n", "Yes, CamemBERT is awesome, but for WWT we need a FR-Pretrained GPT-2 model!\n\nEnvoyé de mon iPad\n\n> Le 22 nov. 2019 à 14:55, Julien Chaumond <[email protected]> a écrit :\n> \n> \n> That's awesome news, @jqueguiner! Let us know if we can help.\n> \n> @BramVanroy To work well with Write With Transformer, we would want more like a FR-pretrained GPT-2-like model. CamemBERT wouldn't do on generation out of the box.\n> \n> See also the more specific issue: #1356\n> \n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub, or unsubscribe.\n", "Camembert doesnt offer generation, only syntax analysis and masking due to the nature of the network. Multiple mask generation (<mask><mask><mask>) gives really uggly results as you cna test here : https://market-place.ai.ovh.net/#!/apis/43323c37-59e7-4092-b23c-3759e7c09288/pages/94d31892-4e64-446f-9318-924e64346f9e\r\n\r\nIMO we should start training using OSCAR dataset \r\nhttps://traces1.inria.fr/oscar/\r\n\r\n@julien-c yes we can start with a collab GPT2 french training ipynb together then I'll prepare the env for a DGX1 or something similar. I didn't train a GPT2 before. IS it scaling over multiple GPU's ? do we need horovod adaptation ?", "Oops, sorry everyone. I thought this was a general French model question. My bad. ", "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", "So is HuggingFace going to write the finetuning implementation for SuperGlue?", "Hi @jiachangliu, did you have any news about support for superglue?", "> Hi @jiachangliu, did you have any news about support for superglue?\r\n\r\nNo I have not heard any HugginFace support on SuperGlue. It was not urgent for me to run those experiments. However, if you want to run SuperGlue, I guess you need to install JIANT, which uses the model structures built by HuggingFace.", "> > Hi @jiachangliu, did you have any news about support for superglue?\r\n> \r\n> No I have not heard any HugginFace support on SuperGlue. It was not urgent for me to run those experiments. However, if you want to run SuperGlue, I guess you need to install JIANT, which uses the model structures built by HuggingFace.\r\n\r\nThank you !! " ]

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

[]

[ "Pre-training is indeed a tough pill to swallow. First of all you need a good dataset (does such dataset exist for French?), second you need a lot of processing power. A lot. If a dataset is available (preprocessed, ready to train) then I'd be willing to look into training the model on hardware that I have available. ", "Have you an example of a good dataset prepared for the english language (my experience on such things is limited to training Glove on a cleaned dump of the french wikipedia) ?", "English BERT was trained on Wikipedia and BookCorpus for 1M steps.\r\n\r\nAfter reading throug hthe BERT readme, I have to retract my previous statement, though. I do not have the resources to pretrain such a model. I thought it would be max one week on a V100, but they speak of four days on *4 to 16 cloud TPUs*. I do not possess such power!", "Hi Bram,\n\nI planned to use the French Wikipedia and some Gutenberg famous French works like La comédie humaine for a start, I let you know when I finish to preprocess them. Concerning the hardware I would like to use gpu ec2 spot instances but I do not know how long I’ll have to run them and if it exceeds my meagre financial resources.\n\n\n\nEnvoyé de mon iPad\n\n> Le 28 sept. 2019 à 10:53, Nestor Demeure <[email protected]> a écrit :\n> \n> Have you an example of a good dataset prepared for the english language (my experience on such things is limited to training Glove on a cleaned dump of the french wikipedia) ?\n> \n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub, or mute the thread.\n", "Reading [this](https://cloud.google.com/blog/products/ai-machine-learning/now-you-can-train-ml-models-faster-and-lower-cost-cloud-tpu-pods) comparison post, 16 TPUv2's are about twice as fast as 8x V100's that are in the ec2 instances. I would then guess that you'd have to run training for a week.", "Order of magnitude for the compute cost (on cloud platforms) of pre-training a large model is anywhere between $10k and $100k. That's for one pre-training, and you usually at least start multiple ones to search the hyperparameter space.\r\n\r\nRoBERTa was pre-trained for 24 hours on 1,024 (full size, 32GB) V100s.", "> Order of magnitude for the compute cost (on cloud platforms) of pre-training a large model is anywhere between $10k and $100k. That's for one pre-training, and you usually at least start multiple ones to search the hyperparameter space.\r\n> \r\n> RoBERTa was pre-trained for 24 hours on 1,024 (full size, 32GB) V100s.\r\n\r\nPretty sure that [this](https://media1.tenor.com/images/dbf3ee8c8e92b4c1bd3492636a774dc7/tenor.gif) is applicable for everyone here.", "i made a dataset by converting books from [bibebook](http://www.bibebook.com/) package to text files.\r\nit's a package of 1 700 Créative Commons BY-SA and public domain book in french \r\n\r\n[livre en francais kaggle dataset](https://www.kaggle.com/cedriclacrambe/livres-en-francais)", "Wonderful! Thank you very much!\n\n> Le 30 sept. 2019 à 12:33, cedspam <[email protected]> a écrit :\n> \n> i made a dataset by converting books from bibebook <http://www.bibebook.com/> to text files.\n> it's a package of 1 700 Créative Commons BY-SA and public domain book in french\n> \n> livre francais kaggle dataset <https://www.kaggle.com/cedriclacrambe/livres-en-francais>\n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub <https://github.com/huggingface/transformers/issues/1356?email_source=notifications&email_token=AAHXAP2JVSBU2KSTRLJI6HDQMHIZJA5CNFSM4I3IELGKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD75GLQY#issuecomment-536503747>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAHXAP7ER7H4ERVY7J7JS7LQMHIZJANCNFSM4I3IELGA>.\n> \n\n", "Hi all,\r\n\r\nI'm currently preparing the `.tfrecords` (both cased and uncased) for a French BERT model (corpus is mainly taken from Wikipedia + OPUS corpora, resulting in ~20GB of text). \r\n\r\nI'll share the results (TF checkpoints + Transformers weights) whenever the training on TPU has finished.\r\n\r\nEvaluation tasks for that model are a bit limited, so I would evaluate the model for PoS tagging and NER (Universal Dependencies and WikiANN) and compare the model with mBERT.", "Great news!\n\nEnvoyé de mon iPad\n\n> Le 5 oct. 2019 à 20:20, Stefan Schweter <[email protected]> a écrit :\n> \n> \n> Hi all,\n> \n> I'm currently preparing the .tfrecords (both cased and uncased) for a French BERT model (corpus is mainly taken from Wikipedia + OPUS corpora, resulting in ~20GB of text).\n> \n> I'll share the results (TF checkpoints + Transformers weights) whenever the training on TPU has finished.\n> \n> Evaluation tasks for that model are a bit limited, so I would evaluate the model for PoS tagging and NER (Universal Dependencies and WikiANN) and compare the model with mBERT.\n> \n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub, or mute the thread.\n", "That's awesome @stefan-it. Let us know if we can help.", "I'm training the GPT-2 on corpus of Russian classical literature. I've modified training script to make it more robust and useful. You can find it [here](https://github.com/mgrankin/ru_transformers). ", "Thanks for sharing Mikhail :)\n\nEnvoyé de mon iPad\n\n> Le 7 oct. 2019 à 17:53, Mikhail Grankin <[email protected]> a écrit :\n> \n> \n> I'm training the GPT-2 on corpus of Russian classical literature. I've modified training script to make it more robust and useful. You can find it here.\n> \n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub, or mute the thread.\n", "> Hi all,\r\n> \r\n> I'm currently preparing the `.tfrecords` (both cased and uncased) for a French BERT model (corpus is mainly taken from Wikipedia + OPUS corpora, resulting in ~20GB of text).\r\n> \r\n> I'll share the results (TF checkpoints + Transformers weights) whenever the training on TPU has finished.\r\n> \r\n> Evaluation tasks for that model are a bit limited, so I would evaluate the model for PoS tagging and NER (Universal Dependencies and WikiANN) and compare the model with mBERT.\r\n\r\n@stefan-it Could you explain to me how you trained your model from scratch without using Bert multilingual?\r\n\r\nI would like to train BERT from scratch for a textual base in PT-BR (8GB data). Is it possible to use the run_lm_finetuning.py code to perform this process without using the multi-language bert model?\r\n\r\nI already have a vocab.txt for the PT-BR base and I don't want to load initial weights.\r\n\r\nIs there any script or tutorial to perform this process step by step?", "I don’t know if this link https://github.com/facebookresearch/XLM can answer your question. \n\nEnvoyé de mon iPad\n\n> Le 17 oct. 2019 à 20:03, calusbr <[email protected]> a écrit :\n> \n> \n> Hi all,\n> \n> I'm currently preparing the .tfrecords (both cased and uncased) for a French BERT model (corpus is mainly taken from Wikipedia + OPUS corpora, resulting in ~20GB of text).\n> \n> I'll share the results (TF checkpoints + Transformers weights) whenever the training on TPU has finished.\n> \n> Evaluation tasks for that model are a bit limited, so I would evaluate the model for PoS tagging and NER (Universal Dependencies and WikiANN) and compare the model with mBERT.\n> \n> @stefan-it Could you explain to me how you trained your model from scratch without using Bert multilingual?\n> \n> I would like to train BERT from scratch for a textual base in PT-BR (8GB data). Is it possible to use the run_lm_finetuning.py code to perform this process without using the multi-language bert model?\n> \n> I already have a vocab.txt for the PT-BR base and I don't want to load initial weights.\n> \n> Is there any script or tutorial to perform this process step by step?\n> \n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub, or unsubscribe.\n", "Hi @calusbr,\r\n\r\nI'm using the official Google BERT implementation from [this repository](https://github.com/google-research/bert) on a TPU. Then the trained model TensorFlow model can easily be converted into a Transformers-compatible one (so I can be used with this library).\r\n\r\nRegarding to your question: if you don't want to use and fine-tune the multi-lingual BERT model, you could try to train a model with the official BERT implementation for a few steps (Google Colab has TPU support). Then you can fine-tune this model with `transformers` (or you can try to use the Colab instance) :)", "> Hi all,\r\n> \r\n> I'm currently preparing the `.tfrecords` (both cased and uncased) for a French BERT model (corpus is mainly taken from Wikipedia + OPUS corpora, resulting in ~20GB of text).\r\n> \r\n> I'll share the results (TF checkpoints + Transformers weights) whenever the training on TPU has finished.\r\n> \r\n> Evaluation tasks for that model are a bit limited, so I would evaluate the model for PoS tagging and NER (Universal Dependencies and WikiANN) and compare the model with mBERT.\r\n\r\nHi @stefan-it !\r\nVery happy to know that you will possibly able to share this model with us!\r\nDo you have any update on it?\r\nMany thanks!! :)", "Sure, no problem :)\r\n\r\nI did some experiments with a training corpus size from 16 to 40 GB. I used the same fine-tuning parameters as used in the SciBERT paper/repository. That means training with a sequence length of 128, then fine-tuning with a sequence length of 512.\r\n\r\nUnfortunately, the model trained from scratch is ~ 0.5% worse than the multilingual model on a WikiNER split (80/10/10). In another experiment I used the TensorFlow checkpoint from the multilingual cased model and did training with a sequence length of 128. This results in a +0.2% \"boost\" on WikiNER.\r\n\r\nHowever, for PoS tagging the model (trained from scratch) is always better (~0.3%) than the BERT multilingual cased model (I used 4 PoS tagging datasets).\r\n\r\nI'm currently doing more experiments (mainly focussing on training corpus cleaning...) and will report back here :)", "Thanks Stefan !\n\n> Le 4 nov. 2019 à 11:33, Stefan Schweter <[email protected]> a écrit :\n> \n> Sure, no problem :)\n> \n> I did some experiments with a training corpus size from 16 to 40 GB. I used the same fine-tuning parameters as used in the SciBERT paper/repository. That means training with a sequence length of 128, then fine-tuning with a sequence length of 512.\n> \n> Unfortunately, the model trained from scratch is ~ 0.5% worse than the multilingual model on a WikiNER split (80/10/10). In another experiment I used the TensorFlow checkpoint from the multilingual cased model and did training with a sequence length of 128. This results in a +0.2% \"boost\" on WikiNER.\n> \n> However, for PoS tagging the model (trained from scratch) is always better (~0.3%) than the BERT multilingual cased model (I used 4 PoS tagging datasets).\n> \n> I'm currently doing more experiments (mainly focussing on training corpus cleaning...) and will report back here :)\n> \n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub <https://github.com/huggingface/transformers/issues/1356?email_source=notifications&email_token=AAHXAP7ZVWXK4GP236MLDIDQR726NA5CNFSM4I3IELGKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEC6ZPXY#issuecomment-549296095>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAHXAP3PKUDBDELDVEGMUT3QR726NANCNFSM4I3IELGA>.\n> \n\n", "Thanks for your work @stefan-it. It's nice, but perhaps disappointing, to see that the multilingual models aren't that bad after all. From what I read, the multilingual models were said to perform poorly but from your tests it seems that is not (laways?) the case.", "I think we should wait for CamemBERT then 😅\r\n\r\nhttps://camembert-model.fr/", "Coming soon! cc @louismartin @LysandreJik ", "Two days ago they released on arXiv the [https://128.84.21.199/pdf/1911.03894.pdf](url)\r\n\r\n> I think we should wait for CamemBERT then \r\n> \r\n> https://camembert-model.fr/", "CamemBERT was merged into master: https://github.com/huggingface/transformers/pull/1822\r\n\r\nI'll keep this issue open for GPT.", "Hello, this thread is what I was looking for but I'm not sure I found the answer to my questions:\r\n- how long does it take to go through GPT-2 and BERT in French?\r\n- what configuration of GPUs?\r\n- what size of corpus?\r\n\r\nThanks a lot in advance.", "We trained CamemBERT on 138GB of raw text on 256 GPUs (32 GB Tesla V100) for 1 day.", "Thank you very much for this valuable information !\n\nChristian Mauceri, PhD\nLe 4 déc. 2019 à 16:17 +0100, Louis Martin <[email protected]>, a écrit :\n> We trained CamemBERT on 138GB of raw text on 258 GPUs (32 GB Tesla V100) for 1 day.\n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub, or unsubscribe.\n", "> We trained CamemBERT on 138GB of raw text on 258 GPUs (32 GB Tesla V100) for 1 day.\r\n\r\nThanks @louismartin. I find great what your did and published with CamemBERT (I'm French :-) ) and the fact you share as well this kind of information. \r\n\r\nAbout your answer: 258 GPUs Tesla V100... waoooooo!!!!!\r\nWhere did you find this power of computation? In [Facebook AI](https://ai.facebook.com)?\r\n\r\nI read in the [Download section of CamemBERT site](https://camembert-model.fr/#download\r\n) that the model has only 110 millions of parameters. Was it worth to train it on 132 GB of data? ", "> Hi all,\r\n> \r\n> I'm currently preparing the `.tfrecords` (both cased and uncased) for a French BERT model (corpus is mainly taken from Wikipedia + OPUS corpora, resulting in ~20GB of text).\r\n> \r\n> I'll share the results (TF checkpoints + Transformers weights) whenever the training on TPU has finished.\r\n> \r\n> Evaluation tasks for that model are a bit limited, so I would evaluate the model for PoS tagging and NER (Universal Dependencies and WikiANN) and compare the model with mBERT.\r\n\r\nHi @stefan-it , do you mind to upload your French Bert check point ? I am interested in your model for generation task. Thanks" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1355?src=pr&el=h1) Report\n> Merging [#1355](https://codecov.io/gh/huggingface/transformers/pull/1355?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/ca559826c4188be8713e46f191ddf5f379c196e7?src=pr&el=desc) will **decrease** coverage by `0.04%`.\n> The diff coverage is `50%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1355?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1355 +/- ##\n==========================================\n- Coverage 84.73% 84.68% -0.05% \n==========================================\n Files 84 84 \n Lines 12573 12592 +19 \n==========================================\n+ Hits 10654 10664 +10 \n- Misses 1919 1928 +9\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1355?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/data/processors/utils.py](https://codecov.io/gh/huggingface/transformers/pull/1355/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2RhdGEvcHJvY2Vzc29ycy91dGlscy5weQ==) | `46.66% <100%> (+1.21%)` | :arrow_up: |\n| [transformers/data/processors/glue.py](https://codecov.io/gh/huggingface/transformers/pull/1355/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2RhdGEvcHJvY2Vzc29ycy9nbHVlLnB5) | `27.98% <47.36%> (+1.76%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1355?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/1355?src=pr&el=footer). Last update [ca55982...795b3e7](https://codecov.io/gh/huggingface/transformers/pull/1355?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Nice and clean, thanks a lot @agrinh and @philipp-eisen!", "@thomwolf Happy to help, we're finding this package super useful!" ]

[]

[]

[ "Fixed with #1355" ]

[]

[]

[ "👍" ]

[ { "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" ]

[]

[]

[ "Need to use the flag \r\n--version_2_with_negative" ]

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

[]

[ "The main issue is at line 85 on the forward pass of `TFRobertaMainLayer`:\r\n\r\nhttps://github.com/huggingface/transformers/blob/ca559826c4188be8713e46f191ddf5f379c196e7/transformers/modeling_tf_roberta.py#L85\r\n\r\nIt seems that passing Input placeholders mess up this comparison:\r\n\r\n> OperatorNotAllowedInGraphError: using a `tf.Tensor` as a Python `bool` is not allowed in Graph execution. Use Eager execution or decorate this function with @tf.function.\r\n\r\nWhen I comment-out this block of code, the training process works... I can't find any way to by-pass this error without commenting-out though....", "@iliaschalkidis I've also run into this issue when trying to make a plug-and-play wrapper around the numerous TF-compatible models. Like you, I was able to get the RoBERTa model working by hacking around it a bit. Not ideal, but it works.\r\n\r\nFor anyone else that's interested, the line above that raises the error occurs in `TFRobertaMainLayer.call`. You can get around it by wrapping the `call` as a TensorFlow 2.0 `function` whenever you want to use a model that depends on `TFRobertaMainLayer` (which is all of them?). Here I'm using `TFRobertaForSequenceClassification`:\r\n\r\n```python\r\nfrom transformers import TFRobertaForSequenceClassification\r\nimport tensorflow as tf\r\n\r\n# Establish a RoBERTa-based classifier.\r\nclf = TFRobertaForSequenceClassification.from_pretrained(\"roberta-base\", num_labels=5)\r\n\r\n# \"Decorate\" the `call` method as a TensorFlow2.0 function.\r\nclf.roberta.call = tf.function(clf.transformer.roberta.call)\r\n```\r\n\r\nUsing that, I was successfully able to fine-tune the classifier on a multi-GPU setup without much trouble. I still get a ton of warnings from ZMQ and TensorFlow, but I'm not yet sure they're official `transformer` issues. \r\n\r\n**Note:** I suspect you'll have to wrap the `call` instance method any time you initialize this model (e.g., if you save your pre-trained model and re-load for prediction/inference, you may not be able to just use `TFRobertaForSequenceClassification`). In that case, it may be simpler to define a minimal subclass that does it for you. This is untested code, but I suspect it'd work alright:\r\n\r\n```python\r\nimport tensorflow as tf\r\nimport transformers\r\n\r\n\r\nclass _TFRobertaForSequenceClassification(transformers.TFRobertaForSequenceClassification):\r\n \r\n def __init__(self, config, *inputs, **kwargs):\r\n super(_TFRobertaForSequenceClassification, self).__init__(config, *inputs, **kwargs)\r\n self.roberta.call = tf.function(self.roberta.call)\r\n```\r\n\r\nHope this helps!\r\n\r\n---\r\n\r\nUnrelated tip: I also had a bit of trouble using TFv2 metrics (e.g., `tf.keras.metrics.[Precision/Recall/AUC]` because the `TFRobertaClassificationHead` outputs logits (no softmax activation). If anybody else is wondering, you can set the classifier head's output layer to use softmax quite easily:\r\n\r\n```python\r\n# Continuing from the previous setup.\r\nclf.classifier.out_proj.activation = tf.keras.activations.softmax\r\n```\r\n\r\nThis way, you can monitor Precision/Recall/AUC in the call to `clf.compile`:\r\n\r\n```python\r\n# Compile our model.\r\nclf.compile(\r\n optimizer=...,\r\n loss=...,\r\n metrics=[\r\n tf.keras.metrics.CategoricalCrossentropy(from_logits=False),\r\n tf.keras.metrics.Precision(thresholds=.50, name=\"precision\"),\r\n tf.keras.metrics.Recall(thresholds=.50, name=\"recall\"),\r\n tf.keras.metrics.AUC(curve=\"PR\", name=\"auc-pr\")\r\n ]\r\n)\r\n```\r\n\r\nFurthermore, if you want to just fine-tune the classifier layer, you can easily freeze the core RoBERTa layers:\r\n\r\n```python\r\n# Note you have ~125M trainable parameters. This'll take a while!\r\nclf.summary()\r\n\r\n# Freeze core RoBERTa model (embeddings, encoder, pooler).\r\nclf.roberta.trainable = False\r\n\r\n# Note you have ~600K trainable parameters. Much better!\r\nclf.summary()\r\n```", "@dataframing thanx a lot, this was really helpful! I opted to go with a very similar solution...\r\n\r\nDefine a meta-model on top of `TFRobertaModel`:\r\n\r\n```python\r\nimport tensorflow as tf\r\nimport transformers\r\n\r\n\r\nclass ROBERTA(transformers.TFRobertaModel):\r\n\r\n def __init__(self, config, *inputs, **kwargs):\r\n super(ROBERTA, self).__init__(config, *inputs, **kwargs)\r\n self.roberta.call = tf.function(self.roberta.call)\r\n\r\n```\r\n\r\nBuild a wrapper `tf.keras.Model`:\r\n\r\n```python\r\n# Define inputs (token_ids, mask_ids, seg_ids)\r\ntoken_inputs = Input(shape=(None,), name='word_inputs', dtype='int32')\r\nmask_inputs = Input(shape=(None,), name='mask_inputs', dtype='int32')\r\nseg_inputs = Input(shape=(None,), name='seg_inputs', dtype='int32')\r\n\r\n# Load model and collect encodings\r\nroberta = ROBERTA.from_pretrained('roberta-base')\r\nroberta_encodings = roberta([token_inputs, mask_inputs, seg_inputs])[0]\r\n\r\n# Keep [CLS] token encoding\r\ndoc_encoding = tf.squeeze(roberta_encodings[:, 0:1, :], axis=1)\r\n\r\n# Apply dropout\r\ndoc_encoding = Dropout(0.1)(doc_encoding)\r\n\r\n# Final output (projection) layer\r\noutputs = Dense(self.n_classes, activation='sigmoid', name='outputs')(doc_encoding)\r\n\r\n# Wrap-up model\r\nmodel = Model(inputs=[word_inputs, mask_inputs, seg_inputs], outputs=[outputs])\r\nmodel.compile(optimizer=Adam(lr=3e-4), loss='binary_crossentropy')\r\n```\r\n\r\nEverything works like a charm, except the annoying warnings. Although working on a single RTX 2080Ti or any other 12GB GPU has a limitation of batch size up to 4-5 samples of 512 subword units (the same applies for BERT), while I was able to go up to 8 when I was calling `bert-base` via Tensorflow Hub and wrap it as Keras layer, which is really weird... Any idea why, moving to transformers library and TF2 will make such a different?", "Thanks for the report.\r\nWe can probably get rid of this test in the TF version of RoBERTa if it's a blocking element for integrating with other Keras modules.\r\nI've never been a huge fan of this hacky solution anyway. In the future, we should probably move forward with a breaking change in the tokenizers and have control tokens included by default in the tokenizer encoding output instead of having them as an option.\r\ncc @LysandreJik @julien-c ", "@thomwolf Having the tokenizers include special tokens in the call to `tokenizer.encode[_plus]` seems like a pretty safe default, but I think it also makes sense to have this inline inspection to make sure that the end user has properly encoded their tokens. Wrapping the method in a `tf.function` like above call seems to make it work fine as-is, so maybe there's a way to have the best of both worlds?", "@dataframing BERT and ROBERTa work like a charm with the tweaks you proposed, although with XLNet I still have issues:\r\n\r\n```python\r\n# Define token ids as inputs\r\nword_inputs = Input(batch_shape=(2, 2000), name='word_inputs', dtype='int32')\r\n\r\n# Call XLNet model\r\nxlnet = TFXLNetModel.from_pretrained('xlnet-base-cased')\r\nxlnet_encodings = xlnet(word_inputs)\r\n\r\n# Collect last hidden step (CLS)\r\ndoc_encoding = tf.squeeze(xlnet_encodings[:, -1:, :], axis=1)\r\n\r\n# Apply dropout\r\ndoc_encoding = Dropout(dropout_rate)(doc_encoding)\r\n\r\n# Final output (projection) layer\r\noutputs = Dense(n_classes, activation='softmax', name='outputs')(doc_encoding)\r\n\r\n# Compile model\r\nmodel = Model(inputs=[word_inputs], outputs=[outputs])\r\nmodel.compile(optimizer=Adam(lr=lr, loss='categorical_crossentropy'))\r\n```\r\n\r\n> xlnet_encodings = xlnet(word_inputs)\r\n> .../tensorflow_core/python/keras/engine/base_layer.py\", line 842, in __call__\r\n> outputs = call_fn(cast_inputs, *args, **kwargs)\r\n> .../tensorflow_core/python/autograph/impl/api.py\", line 237, in wrapper\r\n> raise e.ag_error_metadata.to_exception(e) AttributeError: in converted code:\r\n> relative to .../transformers/modeling_tf_xlnet.py:810 call *\r\n> outputs = self.transformer(inputs, **kwargs)\r\n> tensorflow_core/python/keras/engine/base_layer.py:874 __call__\r\n> inputs, outputs, args, kwargs)\r\n> tensorflow_core/python/keras/engine/base_layer.py:2038 _set_connectivity_metadata_\r\n> input_tensors=inputs, output_tensors=outputs, arguments=arguments)\r\n> tensorflow_core/python/keras/engine/base_layer.py:2068 _add_inbound_node\r\n> arguments=arguments)\r\n> tensorflow_core/python/keras/engine/node.py:110 __init__\r\n> self.output_shapes = nest.map_structure(backend.int_shape, output_tensors)\r\n> tensorflow_core/python/util/nest.py:535 map_structure\r\n> structure[0], [func(*x) for x in entries],\r\n> tensorflow_core/python/util/nest.py:535 <listcomp>\r\n> structure[0], [func(*x) for x in entries],\r\n> tensorflow_core/python/keras/backend.py:1185 int_shape\r\n> shape = x.shape\r\n> AttributeError: 'NoneType' object has no attribute 'shape'\r\n\r\nPretty much the same story happens using the `TFXLNetForSequenceClassification` class:\r\n\r\n```python\r\n# Call TFXLNetForSequenceClassification model\r\nmodel = TFXLNetForSequenceClassification.from_pretrained('xlnet-base-cased', num_labels=n_classes)\r\n\r\n# Amend activation functions\r\nmodel.logits_proj.activation = tf.keras.activations.softmax\r\n\r\n# Compile model\r\nmodel.compile(optimizer=Adam(lr=lr, loss='categorical_crossentropy'))\r\n```\r\n\r\n> File .../tensorflow_core/python/keras/engine/training.py\", line 2709, in _set_inputs\r\n> outputs = self(inputs, **kwargs)\r\n> File .../tensorflow_core/python/keras/engine/base_layer.py\", line 842, in __call__\r\n> outputs = call_fn(cast_inputs, *args, **kwargs)\r\n> File .../tensorflow_core/python/autograph/impl/api.py\", line 237, in wrapper\r\n> raise e.ag_error_metadata.to_exception(e)\r\n> TypeError: in converted code:\r\n> transformers/modeling_tf_xlnet.py:916 call *\r\n> output = self.sequence_summary(output)\r\n> tensorflow_core/python/keras/engine/base_layer.py:842 __call__\r\n> outputs = call_fn(cast_inputs, *args, **kwargs)\r\n> transformers/modeling_tf_utils.py:459 call *\r\n> output = self.first_dropout(output)\r\n> tensorflow_core/python/autograph/impl/api.py:396 converted_call\r\n> return py_builtins.overload_of(f)(*args)\r\n> TypeError: 'NoneType' object is not callable", "In your case, it might be because you are not extracting the hidden states from the model tuple output.\r\nThis line: `xlnet_encodings = xlnet(word_inputs)`\r\nShould be like this:\r\n```\r\noutputs = xlnet(word_inputs)\r\nxlnet_encodings = outputs[0]\r\n```\r\n\r\nI'm working on adding some tests on this integration with other Keras modules here: #1482", "Hi @thomwolf,\r\n\r\nEven with this update it keeps producing the exact same error. The actual error happens internally on TF2, when the abstract `keras.Layer` calls the Autograph API to do some adjustments. This actually parse the whole network layer by layer and convert the `call()` functions for some reason. It fails in the very end, when it tries to convert the final (outer) call of the `TFXLNetMainLayer`:\r\n\r\n```python\r\noutputs = self.transformer(inputs, **kwargs)\r\n```\r\nThe main reason, as I see it through debugging, is the fact that you return by default as part of the outputs a list called `new_mems`. This returns a list of `None`, if the user do not provide such an input, that later the internal Keras engine cannot handle, because the elements of this list lack of shape and lead to the aforementioned error `AttributeError: 'NoneType' object has no attribute 'shape'`. \r\n\r\nThe only way to surpass this at this stage, is again with some hacking in line 653 of `modeling_tf_xlnet.py`:\r\n\r\n```python\r\noutputs = (tf.transpose(output, perm=(1, 0, 2)), new_mems)\r\n```\r\n\r\nto\r\n\r\n```python\r\noutputs = tf.transpose(output, perm=(1, 0, 2))\r\n```\r\n\r\nProbably, if I pass memories as an input in `TFXLNetModel`, this won't happen any more and I'll avoid hacking. Could you please remind me the notion of memories and how should I pass this information when I'm calling the model? Is this a single integer denoting how many steps back can the Transformer-XL use?", "In two words memories are cached hidden-states to be reused to speed up or allow for longer sentence inputs. The best to understand the notion of memory is to read the Transformer-XL paper which is here: http://arxiv.org/abs/1901.02860\r\n\r\nWe have a couple of models outputting memories and it seems to be a problem for Keras indeed (GPT-2) has the same.\r\n\r\nSo the best (non-breaking) solution is probably to add a flag in the configuration that you can set to False to avoid outputting memories or cache. ", "Great, I read Transformer-XL a few months ago. Maybe if I pass memories as input, I'll avoid this error, and probably I have to do so, if i want the model to act as a real Transformer-XL and not forget all previous timesteps at each segment... What's the specification for `mems`: a tensor of shape (batch_size, ) including integers (e.g., 200 steps back) for the memory length?", "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\nWhy not just take the language model layer from the transformer using `roberta_model = roberta_model.layers[0]` and then build on top of it? \r\n\r\n\r\n" ]

[]

[]

[ "👍 ", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1349?src=pr&el=h1) Report\n> Merging [#1349](https://codecov.io/gh/huggingface/transformers/pull/1349?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/d83d295763b738aa0c071f8b63ad6e155b6cf515?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1349?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1349 +/- ##\n=======================================\n Coverage 84.73% 84.73% \n=======================================\n Files 84 84 \n Lines 12573 12573 \n=======================================\n Hits 10654 10654 \n Misses 1919 1919\n```\n\n\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1349?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/1349?src=pr&el=footer). Last update [d83d295...d2de5b9](https://codecov.io/gh/huggingface/transformers/pull/1349?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

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

[]

[ "Please choose a better title for your post and specify (or remove) the first part of your post. \r\n\r\nAs far as I can tell this is an issue specific to the mnli model. As you say it's pre-trained with three final out features. When loading the state dict into the model, all weights from the pretrained model are \"transferred\" to the initialized model, this is a one-to-one mapping. Since RobertaForSequenceClassification has a classification head, which you can configure a.o. with `num_labels` it _can_ clash with the classification head of the pretrained model.\r\n\r\nThe intuitive solution would be to just load all weight excluding the classifier - so that torch doesn't try to load those mis-matching states, if and only if the num_labels specified in `from_pretrained` are not the same as the ones inside the models `self.config`. **However**, I'm not sure if that is the right approach, since one method (from_pretrained) then does different things with the same given pretrained model. In one case you use it completely, in the other you only use part.", "Thanks for the hint. But I do not think this is the mnli model problem, because it worked before even the label size is not 3. This is my old log:\r\n\r\n09/11/2019 22:37:46 - INFO - pytorch_transformers.modeling_utils - loading configuration file https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-mnli-config.json from cache at /root/.cache/torch/pytorch_transformers/54eef9bf74f919edd81b765fee413c8229620f3e271a51bdcdc67797422ef3f3.233bd69ec613d2ebcb1d55823dfc5b1e109157918e13bdbde6db7f694e1a0039\r\n09/11/2019 22:37:46 - INFO - pytorch_transformers.modeling_utils - Model config {\r\n \"attention_probs_dropout_prob\": 0.1,\r\n \"finetuning_task\": null,\r\n \"hidden_act\": \"gelu\",\r\n \"hidden_dropout_prob\": 0.1,\r\n \"hidden_size\": 1024,\r\n \"initializer_range\": 0.02,\r\n \"intermediate_size\": 4096,\r\n \"layer_norm_eps\": 1e-05,\r\n \"max_position_embeddings\": 514,\r\n \"num_attention_heads\": 16,\r\n \"num_hidden_layers\": 24,\r\n \"num_labels\": 2,\r\n \"output_attentions\": false,\r\n \"output_hidden_states\": false,\r\n \"pruned_heads\": {},\r\n \"torchscript\": false,\r\n \"type_vocab_size\": 1,\r\n \"vocab_size\": 50265\r\n}\r\n\r\n09/11/2019 22:37:46 - INFO - pytorch_transformers.modeling_utils - loading weights file https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-mnli-pytorch_model.bin from cache at /root/.cache/torch/pytorch_transformers/1c2e185bc053ae7261ce2289653438a4c05b871ff7f30eaee1cdb787154410e0.c1823b934e18e923174ff260ba955eef25b2205f48fe2655c432a5fb805f8c8a\r\n09/11/2019 22:38:02 - INFO - pytorch_transformers.modeling_utils - Weights of RobertaForSequenceClassification not initialized from pretrained model: ['classifier.dense.weight', 'classifier.dense.bias', 'classifier.out_proj.weight', 'classifier.out_proj.bias']\r\n09/11/2019 22:38:02 - INFO - pytorch_transformers.modeling_utils - Weights from pretrained model not used in RobertaForSequenceClassification: ['lm_head.weight', 'lm_head.bias', 'lm_head.dense.weight', 'lm_head.dense.bias', 'lm_head.layer_norm.weight', 'lm_head.layer_norm.bias']\r\n\r\nYou can see that the system can automatically detect which part of the mnli parameters not used to initialize, then it will neglect it; but now the transformers only output error message, I found it from yesterday.\r\n\r\nMy code is the same, but behave differently now", "It's odd since `modeling_utils` hasn't seen any updates apart from the naming update. I'm not sure where else to look for this issue.", "My \"pytorch_transformers\" was installed maybe 3 weeks ago, but the latest \"transformers\" was installed yesterday. But both will make the same error, in different lines:\r\n\r\n\"transformers\" in line 411:\r\nFile \"/opt/conda/lib/python3.6/site-packages/transformers/modeling_utils.py\", line 411, in from_pretrained\r\n model.__class__.__name__, \"\\n\\t\".join(error_msgs)))\r\nRuntimeError: Error(s) in loading state_dict for RobertaForSequenceClassification:\r\n\tsize mismatch for classifier.out_proj.weight: copying a param with shape torch.Size([3, 1024]) from checkpoint, the shape in current model is torch.Size([2, 1024]).\r\n\tsize mismatch for classifier.out_proj.bias: copying a param with shape torch.Size([3]) from checkpoint, the shape in current model is torch.Size([2]).\r\n\r\n\"pytorch_transformers\" in line 594:\r\nFile \"/opt/conda/lib/python3.6/site-packages/pytorch_transformers/modeling_utils.py\", line 594, in from_pretrained\r\n model.__class__.__name__, \"\\n\\t\".join(error_msgs)))\r\nRuntimeError: Error(s) in loading state_dict for RobertaForSequenceClassification:\r\n\tsize mismatch for classifier.out_proj.weight: copying a param with shape torch.Size([3, 1024]) from checkpoint, the shape in current model is torch.Size([2, 1024]).\r\n\tsize mismatch for classifier.out_proj.bias: copying a param with shape torch.Size([3]) from checkpoint, the shape in current model is torch.Size([2]).\r\n\r\n\r\nBut both come from the same reason. Since the \"pytorch_transformers\" worked for me a couple days ago (even though i found \"transformer\" did not work for me yesterday, but very likely something changed before yesterday, since I haven't run the piece of code for some days).\r\n\r\nNow, I kind of agree with you that the \"roberta-large-mnli\" model itself had something changed recently, which makes it unable to neglect the mismatch of hyperparameters.\r\n", "Also having this issue using the `roberta-large-mnli` model on a single-document (not paired) multiclass classification task.", "I guess the simplest solution would be to load the model with previous num_labels and than directly change its num_labels and initialize a new classifier layer in the `run_glue.py` script. This way you won't need to modify any of the `transformers` code.\r\n\r\nThis is what I do:\r\n```\r\n # for num_labels(mnli)\r\n num_labels_old = config_class.from_pretrained(args.model_name_or_path).num_labels\r\n config = config_class.from_pretrained(args.config_name if args.config_name else args.model_name_or_path,\r\n num_labels=num_labels_old,\r\n finetuning_task=args.task_name,\r\n cache_dir=args.cache_dir if args.cache_dir else None)\r\n tokenizer = tokenizer_class.from_pretrained(args.tokenizer_name if args.tokenizer_name else args.model_name_or_path,\r\n do_lower_case=args.do_lower_case,\r\n cache_dir=args.cache_dir if args.cache_dir else None)\r\n if num_labels != num_labels_old:\r\n config.num_labels = num_labels_old\r\n model = model_class.from_pretrained(args.model_name_or_path,\r\n from_tf=bool('.ckpt' in args.model_name_or_path),\r\n config=config,\r\n cache_dir=args.cache_dir if args.cache_dir else None)\r\n config.num_labels = num_labels\r\n logger.info('Reintializing model classifier layer...')\r\n model.num_labels = num_labels\r\n model.classifier = RobertaClassificationHead(config)\r\n\r\n else:\r\n model = model_class.from_pretrained(args.model_name_or_path,\r\n from_tf=bool('.ckpt' in args.model_name_or_path),\r\n config=config,\r\n cache_dir=args.cache_dir if args.cache_dir else None)\r\n\r\n```\r\n\r\nOf course, it would be better to modify the `transformers` code directly. \r\n", "Hi felicity,\n\nSorry for the late reply. I actually have forgotten how i solve that, or\ngave up. I will check my code when I finish a deadline in the next couple\nof days. Thanks for sharing your experience.\n\nBest.\n\nOn Thu, Nov 28, 2019 at 8:01 AM felicitywang <[email protected]>\nwrote:\n\n> I'm getting the same error. Did you solve this problem? @wyin-Salesforce\n> <https://github.com/wyin-Salesforce> @pmbaumgartner\n> <https://github.com/pmbaumgartner> Would really appreciate it if you\n> could share your solutions. Thank you.\n>\n> —\n> You are receiving this because you authored the thread.\n> Reply to this email directly, view it on GitHub\n> <https://github.com/huggingface/transformers/issues/1348?email_source=notifications&email_token=AM2XN4JEMO7KYWJ4OFXFSYDQV7TPFA5CNFSM4I3D6BP2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEFM7XIA#issuecomment-559545248>,\n> or unsubscribe\n> <https://github.com/notifications/unsubscribe-auth/AM2XN4IUZKSK4D22E2WI3ZDQV7TPFANCNFSM4I3D6BPQ>\n> .\n>\n\n\n-- \n\n\nWenpeng Yin\nResearch Scientist @ Salesforce Research, Palo Alto\nhttps://sites.google.com/site/yinwenpeng1987/\n", "Thanks @wyin-Salesforce . If people are still having trouble with this, the solution above worked. ", "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 want to evaluate the pre-trained **roberta-large-mnli** model on a **2-way classification task**. I tried to imitate what @felicitywang posted by adding these four lines after calling config/tokenizer/model in run_glue.py (after line 134):\r\n```\r\n num_labels = 2 # ADDED\r\n config.num_labels = num_labels # ADDED\r\n model.num_labels = num_labels # ADDED\r\n model.classifier = RobertaClassificationHead(config) # ADDED\r\n```\r\n\r\nHowever, I'm still getting the following error (from line 131) when I run my modified run_glue.py:\r\n```\r\nRuntimeError: Error(s) in loading state_dict for RobertaForSequenceClassification:\r\n\tsize mismatch for classifier.out_proj.weight: copying a param with shape torch.Size([3, 1024]) from checkpoint, the shape in current model is torch.Size([2, 1024]).\r\n\tsize mismatch for classifier.out_proj.bias: copying a param with shape torch.Size([3]) from checkpoint, the shape in current model is torch.Size([2]).\r\n```\r\n\r\nwhere the line 131 is the last line ('cache_dir=...') of this code block:\r\n```\r\n model = AutoModelForSequenceClassification.from_pretrained(\r\n model_args.model_name_or_path,\r\n from_tf=bool(\".ckpt\" in model_args.model_name_or_path),\r\n config=config,\r\n cache_dir=model_args.cache_dir,\r\n )\r\n```\r\n\r\nDoes anyone know how to make it work?", "> I want to evaluate the pre-trained **roberta-large-mnli** model on a **2-way classification task**. I tried to imitate what @felicitywang posted by adding these four lines after calling config/tokenizer/model in run_glue.py (after line 134):\r\n> \r\n> ```\r\n> num_labels = 2 # ADDED\r\n> config.num_labels = num_labels # ADDED\r\n> model.num_labels = num_labels # ADDED\r\n> model.classifier = RobertaClassificationHead(config) # ADDED\r\n> ```\r\n> \r\n> However, I'm still getting the following error (from line 131) when I run my modified run_glue.py:\r\n> \r\n> ```\r\n> RuntimeError: Error(s) in loading state_dict for RobertaForSequenceClassification:\r\n> \tsize mismatch for classifier.out_proj.weight: copying a param with shape torch.Size([3, 1024]) from checkpoint, the shape in current model is torch.Size([2, 1024]).\r\n> \tsize mismatch for classifier.out_proj.bias: copying a param with shape torch.Size([3]) from checkpoint, the shape in current model is torch.Size([2]).\r\n> ```\r\n> \r\n> where the line 131 is the last line ('cache_dir=...') of this code block:\r\n> \r\n> ```\r\n> model = AutoModelForSequenceClassification.from_pretrained(\r\n> model_args.model_name_or_path,\r\n> from_tf=bool(\".ckpt\" in model_args.model_name_or_path),\r\n> config=config,\r\n> cache_dir=model_args.cache_dir,\r\n> )\r\n> ```\r\n> \r\n> Does anyone know how to make it work?\r\n\r\n@scarletcho \r\nIf you just want to evaluate the pretrained roberta-large-mnli on a new dataset without any fine-tuning; let's say your new dataset has two classes \"entail\" and \"non_entail\", then you just manually combine the outputs \"neutral\" and \"contradict\" as a single output \"non_entail\".\r\n\r\nIf you want to load this pretrained model and fine-tune on your 2-way dataset, today I just found the following approach works for using N-way fine-tuning:\r\n\r\n` model_config = BartConfig.from_pretrained(pretrain_model_dir)\r\n model_config.num_labels=new_num_labels\r\n model = BartForSequenceClassification.from_pretrained(pretrain_model_dir, config=model_config)`\r\n \r\nI tried Bart, but it should work for roberta too (here \"pretrain_model_dir\" is string \"facebook/bart-large\", you can use \"roberta-large-mnli\" instead)", "> I guess the simplest solution would be to load the model with previous num_labels and than directly change its num_labels and initialize a new classifier layer in the `run_glue.py` script. This way you won't need to modify any of the `transformers` code.\r\n> \r\n> This is what I do:\r\n> \r\n> ```\r\n> # for num_labels(mnli)\r\n> num_labels_old = config_class.from_pretrained(args.model_name_or_path).num_labels\r\n> config = config_class.from_pretrained(args.config_name if args.config_name else args.model_name_or_path,\r\n> num_labels=num_labels_old,\r\n> finetuning_task=args.task_name,\r\n> cache_dir=args.cache_dir if args.cache_dir else None)\r\n> tokenizer = tokenizer_class.from_pretrained(args.tokenizer_name if args.tokenizer_name else args.model_name_or_path,\r\n> do_lower_case=args.do_lower_case,\r\n> cache_dir=args.cache_dir if args.cache_dir else None)\r\n> if num_labels != num_labels_old:\r\n> config.num_labels = num_labels_old\r\n> model = model_class.from_pretrained(args.model_name_or_path,\r\n> from_tf=bool('.ckpt' in args.model_name_or_path),\r\n> config=config,\r\n> cache_dir=args.cache_dir if args.cache_dir else None)\r\n> config.num_labels = num_labels\r\n> logger.info('Reintializing model classifier layer...')\r\n> model.num_labels = num_labels\r\n> model.classifier = RobertaClassificationHead(config)\r\n> \r\n> else:\r\n> model = model_class.from_pretrained(args.model_name_or_path,\r\n> from_tf=bool('.ckpt' in args.model_name_or_path),\r\n> config=config,\r\n> cache_dir=args.cache_dir if args.cache_dir else None)\r\n> ```\r\n> \r\n> Of course, it would be better to modify the `transformers` code directly.\r\n\r\nHi,\r\n\r\nI am using this code to solve this issue. What is `RobertaClassificationHead(config)` ? I cannot find this from huggingface.", "> > I want to evaluate the pre-trained **roberta-large-mnli** model on a **2-way classification task**. I tried to imitate what @felicitywang posted by adding these four lines after calling config/tokenizer/model in run_glue.py (after line 134):\r\n> > ```\r\n> > num_labels = 2 # ADDED\r\n> > config.num_labels = num_labels # ADDED\r\n> > model.num_labels = num_labels # ADDED\r\n> > model.classifier = RobertaClassificationHead(config) # ADDED\r\n> > ```\r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > However, I'm still getting the following error (from line 131) when I run my modified run_glue.py:\r\n> > ```\r\n> > RuntimeError: Error(s) in loading state_dict for RobertaForSequenceClassification:\r\n> > \tsize mismatch for classifier.out_proj.weight: copying a param with shape torch.Size([3, 1024]) from checkpoint, the shape in current model is torch.Size([2, 1024]).\r\n> > \tsize mismatch for classifier.out_proj.bias: copying a param with shape torch.Size([3]) from checkpoint, the shape in current model is torch.Size([2]).\r\n> > ```\r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > where the line 131 is the last line ('cache_dir=...') of this code block:\r\n> > ```\r\n> > model = AutoModelForSequenceClassification.from_pretrained(\r\n> > model_args.model_name_or_path,\r\n> > from_tf=bool(\".ckpt\" in model_args.model_name_or_path),\r\n> > config=config,\r\n> > cache_dir=model_args.cache_dir,\r\n> > )\r\n> > ```\r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > Does anyone know how to make it work?\r\n> \r\n> @scarletcho\r\n> If you just want to evaluate the pretrained roberta-large-mnli on a new dataset without any fine-tuning; let's say your new dataset has two classes \"entail\" and \"non_entail\", then you just manually combine the outputs \"neutral\" and \"contradict\" as a single output \"non_entail\".\r\n> \r\n> If you want to load this pretrained model and fine-tune on your 2-way dataset, today I just found the following approach works for using N-way fine-tuning:\r\n> \r\n> ` model_config = BartConfig.from_pretrained(pretrain_model_dir) model_config.num_labels=new_num_labels model = BartForSequenceClassification.from_pretrained(pretrain_model_dir, config=model_config)`\r\n> \r\n> I tried Bart, but it should work for roberta too (here \"pretrain_model_dir\" is string \"facebook/bart-large\", you can use \"roberta-large-mnli\" instead)\r\n\r\nIt is ok to use roberta-large, but it stills has the error in roberta-large-mnli.", "I fix the issue when I use `transformers=2.3.0` and I put num_labels in the config and then put the config into the model.", "I'm using `transformers=4.20.1` and [this example code](https://github.com/huggingface/transformers/blob/24a85cca61fda92b9376fe45da1dcb10c8853066/examples/pytorch/text-classification/run_glue.py) (the most recent commit that passed all the automated testing) and I'm still running into this error.\r\n\r\nIn the code, it looks like they do [add the num_labels](https://github.com/huggingface/transformers/issues/1348#issuecomment-888779209) to the config, and then put the config into the model, but I'm still getting the error.\r\n\r\nThe exact command I'm running is `python run_glue.py --train_file sg_train_dataset.csv --validation_file sg_test_dataset.csv --do_train --do_eval --model_name roberta-large-mnli --output_dir output --overwrite_output_dir` where `sg_test_dataset.csv` is a CSV file with three columns, \"sentence1\", \"sentence2\" and \"label\" and \"label\" is either 0 or 1.\r\n\r\nAny suggestions on how to fix it?" ]

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

[]

[ "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/1346?src=pr&el=h1) Report\n> Merging [#1346](https://codecov.io/gh/huggingface/transformers/pull/1346?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/da2e47ad15e552b84815da20daf3282b517103f7?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1346?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1346 +/- ##\n=======================================\n Coverage 84.73% 84.73% \n=======================================\n Files 84 84 \n Lines 12573 12573 \n=======================================\n Hits 10654 10654 \n Misses 1919 1919\n```\n\n\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1346?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/1346?src=pr&el=footer). Last update [da2e47a...15749bf](https://codecov.io/gh/huggingface/transformers/pull/1346?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Awesome, thanks @BramVanroy!" ]

[ { "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" ]