Datasets:

---

DanielPFlorian

/

Transformers-Github-Issues

like 0

[ { "id": 1108649053, "node_id": "MDU6TGFiZWwxMTA4NjQ5MDUz", "url": "https://api.github.com/repos/huggingface/transformers/labels/Help%20wanted", "name": "Help wanted", "color": "008672", "default": false, "description": "Extra attention is needed, help appreciated" } ]

[]

[ "any solution here?", "This looks like an incompatibility between apex and zope.\r\nHave you tried without installing apex?", "> This looks like an incompatibility between apex and zope.\r\n> Have you tried without installing apex?\r\n\r\nI uninstalled apex, it works now!\r\nThank you so much!!!! " ]

[]

[]

[]

[]

[]

[ "👍 " ]

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

[]

[ "I resolved this issue by adding the global_step to the skipping list. I think global_step is not required for using pretrained model. Please correct me if I am wrong.", "Is Pytorch requires a TF check point converted? am finding hard to load the checkpoint I generated.BTW is it safe to convert TF checkpoint ?", "> I resolved this issue by adding the global_step to the skipping list. I think global_step is not required for using pretrained model. Please correct me if I am wrong.\r\n\r\ncan you explain me what is skipping list?", "In the file `modeling.py` add it to the list at:\r\n`if any(n in [\"adam_v\", \"adam_m\"] for n in name):`", "Is it possible to load Tensorflow checkpoint using pytorch and do fine tunning? \r\nI can load pytorch_model.bin and finding hard to load my TF checkpoint.Documentation says it can load a archive with bert_config.json and model.chkpt but I have bert_model_ckpt.data-0000-of-00001 in my TF checkpoint folder so am confused. Is there specific example how to do this?\r\n\r\n\r\n\r\n\r\n", "There is a conversion script to convert a tf checkpoint to pytorch: https://github.com/huggingface/pytorch-pretrained-BERT/blob/master/pytorch_pretrained_bert/convert_tf_checkpoint_to_pytorch.py", "> In the file `modeling.py` add it to the list at:\r\n> `if any(n in [\"adam_v\", \"adam_m\"] for n in name):`\r\n\r\nadded global_step in skipping list but still getting same issue.\r\n\r\n", "@naga-dsalgo Is it fixed? I too added \"global_step\" to the list. But still get the error", "Yes it is fixed for me ... I edited installed version not the downloaded\ngit version ..\n\nOn Tue, Apr 2, 2019 at 4:37 AM Shivam Akhauri <[email protected]>\nwrote:\n\n> @naga-dsalgo <https://github.com/naga-dsalgo> Is it fixed? I too added\n> \"global_step\" to the list. But still get the error\n>\n> —\n> You are receiving this because you were mentioned.\n>\n>\n> Reply to this email directly, view it on GitHub\n> <https://github.com/huggingface/pytorch-pretrained-BERT/issues/306#issuecomment-478899861>,\n> or mute the thread\n> <https://github.com/notifications/unsubscribe-auth/AttINdN0Pj9IU0kcwNg_BtrnZdwF6Qjwks5vcxbZgaJpZM4bGhdh>\n> .\n>\n", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "👍 " ]

[]

[]

[ "Hi @graykode,\r\nWhat do you mean by \"code reference\"?", "@thomwolf \r\nHello thomwolf!\r\nIt mean that I apply your code about `GPT-2 model and transferring tensorflow checkpoint to pytorch` in my project!\r\n I show the origin of the information in my project code comment when I refer to your code.\r\nThanks", "Oh yes, no problem.\r\nJust reference the origin of the work and the licences (inherited from the relevant authors and code I started from)", "@thomwolf Sure, What license i follow? I already known original [openAi/gpt-2](https://github.com/openai/gpt-2) is MIT license, but pytorch-pretrained-BERT is Apache 2.0!\r\nI want to just use gpt-2 model and model transferring code!!" ]

[]

[]

[ "This could be related to #266 - are you using the latest version of `pytorch-pretrained-BERT`?", "No, I was on 0.4, I upgraded to 0.6.1 and it worked." ]

[]

[]

[ "👍 " ]

[]

[]

[ "Hi Ben,\r\nPlease read the [relevant example section in the readme](https://github.com/huggingface/pytorch-pretrained-BERT#fine-tuning-openai-gpt-on-the-rocstories-dataset).", "@thomwolf I see that the data is downloaded and cached in case of not providing the `train_dataset` and `eval_dataset` parameters: https://github.com/huggingface/pytorch-pretrained-BERT/blob/master/examples/run_openai_gpt.py#L152-L153, but it fails here: https://github.com/huggingface/pytorch-pretrained-BERT/blob/master/examples/run_openai_gpt.py#L162. So either we make the parameters required or do some additional steps which requires untar of the downloaded data and pointing towards that data. Let me know what you think and I can send a PR." ]

[]

[]

[ "Hi Jasdeep,\r\nNo, they are initialized from Google's pretrained model (they are trained for next sentence prediction task during pretraining)." ]

[]

[]

[ "Update:\r\n\r\nI manually uninstalled PyTorch (`torch | 1.0.1.post2 | 1.0.1.post2` in the list above) from my venv and installed PyTorch 0.4.0 (I couldn't find the whl file for 0.4.1 for Mac OS, which is my platform) by running\r\n\r\n`pip install https://download.pytorch.org/whl/torch-0.4.0-cp36-cp36m-macosx_10_7_x86_64.whl`\r\n\r\nI then ran the tests and I saw that 25 tests passed.\r\n\r\nSo I have just a couple questions at this point:\r\n1. Given that the requirements.txt file states that the expected torch version is >= 0.4.1, and the above observation that 25 tests passed with torch 0.4.0, would all the code in this repo work with 0.4.0?\r\n\r\n2. It is clear that the error described above occurs when using the latest version of torch, specifically torch 1.0.1.post2, but not with 0.4.0. Could you please make the necessary changes to your repo to support 1.0.1.post2?\r\n\r\nThanks!", "Update:\r\n\r\nWas able to find the whl for 0.4.1, so I uninstalled 0.4.0 from my venv and installed 0.4.1. Ran the tests again and 25 tests passed. So question 1 above is irrelevant now. Could I have an answer for question 2?\r\n\r\n", "Hi Karthik,\r\nOn my fresh install with pytorch 1.0.1.post2 (python 3.7) all 25 tests pass without error (also on the continuous integration by the way).\r\nMaybe try to create a clean environment? You don't have to install pytorch prior to pytorch-pretrained-bert, it's a dependency.", "Hi Thomas,\r\nI did start off with a clean virtual environment and I didn't install PyTorch prior to pytorch-pretrained-bert because I saw it's a dependency. The only difference I see between what you've described above and what I did is the version of Python: you used 3.7 while I used 3.6. Maybe that has something to do with this? Could you try with Python 3.6?", "Tested on a clean python 3.6 install and all the tests pass.\r\nHonestly there is not much more I can do at this stage.\r\nClosing for now. Feel free to re-open if you find something." ]

[ { "id": 1108649053, "node_id": "MDU6TGFiZWwxMTA4NjQ5MDUz", "url": "https://api.github.com/repos/huggingface/transformers/labels/Help%20wanted", "name": "Help wanted", "color": "008672", "default": false, "description": "Extra attention is needed, help appreciated" }, { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "Hm, I guess I was using the wrong `checkpoint` file? When I used `/mnt/transformer-xl/tf/sota/pretrained_xl/tf_lm1b/model/model.ckpt-1191000` weights are loaded, but another error occurs:\r\n\r\n```bash\r\nLoading TF weight transformer/r_r_bias/Adam_1 with shape [24, 16, 80]\r\nLoading TF weight transformer/r_w_bias with shape [24, 16, 80]\r\nLoading TF weight transformer/r_w_bias/Adam with shape [24, 16, 80]\r\nLoading TF weight transformer/r_w_bias/Adam_1 with shape [24, 16, 80]\r\nTraceback (most recent call last):\r\n File \"convert_transfo_xl_checkpoint_to_pytorch.py\", line 116, in <module>\r\n args.transfo_xl_dataset_file)\r\n File \"convert_transfo_xl_checkpoint_to_pytorch.py\", line 81, in convert_transfo_xl_checkpoint_to_pytorch\r\n model = load_tf_weights_in_transfo_xl(model, config, tf_path)\r\n File \"/usr/local/lib/python3.6/dist-packages/pytorch_pretrained_bert/modeling_transfo_xl.py\", line 169, in load_tf_weights_in_transfo_xl\r\n assert pointer.shape == array.shape\r\nAssertionError: (torch.Size([3, 1024]), (3, 1280))\r\n```", "Ok, the `TransfoXLConfig` for the `lm1b` model is a bit different. I tried:\r\n\r\n```python\r\nconfig = TransfoXLConfig(vocab_size_or_config_json_file=793472,\r\n cutoffs=[0, 60000, 100000, 640000, 793472],\r\n d_model=1280,\r\n d_embed=1280,\r\n n_head=16,\r\n d_head=80,\r\n d_inner=8192,\r\n div_val=4,\r\n pre_lnorm=False,\r\n n_layer=24,\r\n tgt_len=32,\r\n ext_len=0,\r\n mem_len=128,\r\n clamp_len=-1,\r\n same_length=True,\r\n proj_share_all_but_first=False,\r\n attn_type=0,\r\n sample_softmax=-1,\r\n adaptive=True,\r\n tie_weight=True,\r\n dropout=0.0,\r\n dropatt=0.0,\r\n untie_r=True,\r\n init=\"normal\",\r\n init_range=0.01,\r\n proj_init_std=0.01,\r\n init_std=0.02)\r\n```\r\n\r\nwhich seems not to be 100% correct. Where do I get the model json configuration from (so I can easily pass it to the `convert_transfo_xl_checkpoint_to_pytorch.py` script 🤔", "Hi Stefan,\r\nYou have to create the configuration yourself indeed 🙂\r\nI usually do it by looking at the training parameters of the Tensorflow code related to the model you are trying to load.", "The voab `cutoffs` were wrong. I changed the configuration to:\r\n\r\n```python\r\nconfig = TransfoXLConfig(vocab_size_or_config_json_file=793472,\r\n cutoffs=[60000, 100000, 640000],\r\n d_model=1280,\r\n d_embed=1280,\r\n n_head=16,\r\n d_head=80,\r\n d_inner=8192,\r\n div_val=4,\r\n pre_lnorm=False,\r\n n_layer=24,\r\n tgt_len=32,\r\n ext_len=0,\r\n mem_len=128,\r\n clamp_len=-1,\r\n same_length=True,\r\n proj_share_all_but_first=False,\r\n attn_type=0,\r\n sample_softmax=-1,\r\n adaptive=True,\r\n tie_weight=True,\r\n dropout=0.0,\r\n dropatt=0.0,\r\n untie_r=True,\r\n init=\"normal\",\r\n init_range=0.01,\r\n proj_init_std=0.01,\r\n init_std=0.02,\r\n )\r\n```\r\n\r\nAnd then the `transformer/adaptive_softmax/cutoff_0/proj` key wasn't found in the `tf_weights` dict:\r\n\r\n```python\r\ntransformer/adaptive_softmax/cutoff_0/proj\r\nTraceback (most recent call last):\r\n File \"convert_transfo_xl_checkpoint_to_pytorch.py\", line 142, in <module>\r\n args.transfo_xl_dataset_file)\r\n File \"convert_transfo_xl_checkpoint_to_pytorch.py\", line 107, in convert_transfo_xl_checkpoint_to_pytorch\r\n model = load_tf_weights_in_transfo_xl(model, config, tf_path)\r\n File \"/mnt/pytorch-pretrained-BERT/pytorch_pretrained_bert/modeling_transfo_xl.py\", line 150, in load_tf_weights_in_transfo_xl\r\n assert name in tf_weights\r\nAssertionError\r\n```\r\n", "That's probably a question of weights or projection tying, try to set `tie_weight` or `proj_share_all_but_first` to `False` (the correct value should be indicated in Google/CMU hyper-parameters for lm1b).\r\n\r\n(I can convert this model later if you don't manage to but not before next week unfortunately)", "Thanks for your help @thomwolf ! I'll try to find the correct configuration settings.\r\n\r\nWe are currently trying to integrate the Transformer-XL model into [flair](https://github.com/zalandoresearch/flair), and we would really like to use a larger (in terms of training size) model for downstream tasks like NER :)", "Here's the last configuration I tried:\r\n\r\n```json\r\n{\r\n \"adaptive\": true,\r\n \"attn_type\": 0,\r\n \"clamp_len\": -1,\r\n \"cutoffs\": [\r\n 60000,\r\n 100000,\r\n 640000\r\n ],\r\n \"d_embed\": 1280,\r\n \"d_head\": 80,\r\n \"d_inner\": 8192,\r\n \"d_model\": 1280,\r\n \"div_val\": 4,\r\n \"dropatt\": 0.0,\r\n \"dropout\": 0.1,\r\n \"ext_len\": 0,\r\n \"init\": \"normal\",\r\n \"init_range\": 0.01,\r\n \"init_std\": 0.02,\r\n \"mem_len\": 32,\r\n \"n_head\": 16,\r\n \"n_layer\": 24,\r\n \"n_token\": 793472,\r\n \"pre_lnorm\": false,\r\n \"proj_init_std\": 0.01,\r\n \"same_length\": true,\r\n \"sample_softmax\": -1,\r\n \"tgt_len\": 32,\r\n \"tie_weight\": true,\r\n \"untie_r\": true,\r\n \"proj_share_all_but_first\": false,\r\n \"proj_same_dim\": false,\r\n \"tie_projs\": [\r\n true,\r\n false,\r\n true\r\n ]\r\n}\r\n````\r\n\r\nUnfortunately, an error is thrown. @thomwolf it would be awesome if you can take a look on this :)", "Did you manage to convert this model @stefan-it?", "Sadly, I couldn't managed to convert it (I tried several options)", "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", " @stefan-it did you ever manage to convert this model?", "Hi @irugina, unfortunately, I wasn't able to convert the model 😞" ]

[]

[]

[ "I manage to solve this problem. There is an issue in the calculation of the total optimization steps in `run_squad.py` example that results in a negative learning rate because of the `warmup_linear` schedule. This happens because `t_total` is calculated based on `len(train_examples)` instead of `len(train_features)`. That may not be a problem for datasets with short sentences, but, for long sentences, one example may generate many entries in `train_features` due to the strategy of dividing an example in `DocSpan's`.", "@fabiocapsouza I am trying to handle text classification but my dataset is also highly unbalanced. I am trying to find where I can adjust the class weights when training transformers. Which parameter you changed in your case?", "@MendesSP , since the provided BERT model classes have the loss function hardcoded in the `forward`method, I had to write a subclass to override the `CrossEntropyLoss` definition passing a `weight` tensor." ]

[]

[]

[ "I have the same question. I need to change the `hidden_dropout_prob`. How is that possible?", "Same question. Any best practice? ", "Oh, I find this code works:\r\n```python\r\nhidden_droput_prob = 0.3\r\nconfig = BertConfig.from_pretrained(\"bert-base-uncased\", num_labels=num_labels, hidden_dropout_prob=hidden_dropout_prob)\r\nmodel = BertForMultiLabelClassification.from_pretrained(\"bert-base-uncased\", config=config)\r\n```\r\nAnd `print(model)` could see that drop_out changes:\r\n```sh\r\n (output): BertOutput(\r\n (dense): Linear(in_features=3072, out_features=768, bias=True)\r\n (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\r\n (dropout): Dropout(p=0.3, inplace=False)\r\n```\r\n", "Hi @Opdoop - this code overrides the config parameters of the pertained BERT model. Did I understand correctly? \r\n\r\nAlso, how can we ensure that the `num_labels` parameter is also updated? I don't see it in the output of `print(model)`.", "@kaankork \r\n* For override:\r\nYes. Your understand is correct.\r\n* For `num_labels`:\r\nIt's been a long time. I didn't sure. But you should see `num_labels` by the last layer shape in `print(model)` .", "Any update on this?", "I need to pass some values to config as well, will save me a lot of time....", "Just use the `update` method. \r\nFor example, if you want to change the number of hidden layers, simply use `config.update({'num_hidden_layers': 1})`." ]

[]

[]

[ "Thanks!" ]

[]

[]

[ "Hi Dan,\r\nYou can extract all the hidden-states of Transformer-XL using the snippet indicated in the readme [here](https://github.com/huggingface/pytorch-pretrained-BERT#12-transfoxlmodel).\r\nFor the GPT-2 it's not possible right now.\r\nI can add it in the next release (or you can submit a PR).", "@thomwolf Can we extract final hidden layer representations from GPT-2 1.5 billion models now?" ]

[]

[]

[ "Thanks!" ]

[]

[]

[ "Thanks!" ]

[]

[]

[ "Oh that's right, this one should be a logging.info event like the other ones.", "Fixed" ]

[]

[]

[ "Changes are... even more minor then. I'll close this and open another (hope that's OK)." ]

[]

[]

[ "Thanks, I'll fix that in a future release." ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "Thanks!" ]

[ { "id": 1260952223, "node_id": "MDU6TGFiZWwxMjYwOTUyMjIz", "url": "https://api.github.com/repos/huggingface/transformers/labels/Discussion", "name": "Discussion", "color": "22870e", "default": false, "description": "Discussion on a topic (keep it focused or open a new issue though)" }, { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "Closing for now.", "Why ? @thomwolf ", "I'm trying to clean up the issues to get a better view of what needs to be fixed.\r\nBut you are right opening/closing issue is too binary. Let's add labels instead.", "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" ]

[]

[]

[ "This was an error in `apex`, due to mismatched compiled libraries. Fix can be found [here](https://github.com/NVIDIA/apex/issues/156#issuecomment-465301976).\r\n\r\n> Try a full `pip uninstall apex`, then `cd apex_repo_dir; rm-rf build; python setup.py install --cuda_ext --cpp_ext` and see if the segfault persists." ]

[]

[]

[ "Yes, the examples are not adapted for Python 2, only the library.\r\nI don't plan to adapt or maintain them but feel free to submit a PR!", "env: python2.7\r\nline 662: writer.write(json.dumps(all_predictions, indent=4) + \"\\n\")\r\nchange as :writer.write(json.dumps(all_predictions, indent=4).decode('utf-8') + \"\\n\")" ]

[]

[]

[ "Ok, thanks @wlhgtc!" ]

[]

[]

[ "I'd like to help out on this. I will have a look and try to understand the earlier bridges in this repo. Let me know if you see anywhere a newcomer can be helpful with.", "Sure, would be happy to welcome a PR.\r\nYou can start from `modeling_openai.py` and `tokenization_openai.py`'s codes.\r\nIt's pretty much the same architecture (read OpenAI's paper first!)\r\nYou should mostly reimplement the BPE tokenization to work byte-level and move the layer norms modules to the input rather than the output of the layers.", "Ok, GPT-2 should be in the coming 0.6.0 release (see #287)", "Ok it's on pip: https://github.com/huggingface/pytorch-pretrained-BERT/releases/tag/v0.6.0\r\nPlease read the updated README for details. All should be there (model and examples).\r\nHave a nice week y'all.", "@thomwolf do we have pytorch implementation of GPT-2 small? ", "Yes, just read the README" ]

[]

[]

[]

[]

[]

[ "Managed to get volatile GPU to work properly but memory allocation is sitll imbalanced\r\n\r\n+-----------------------------------------------------------------------------+\r\n| NVIDIA-SMI 410.78 Driver Version: 410.78 CUDA Version: 10.0 |\r\n|-------------------------------+----------------------+----------------------+\r\n| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |\r\n| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |\r\n|===============================+======================+======================|\r\n| 0 Tesla P40 Off | 00000275:00:00.0 Off | 0 |\r\n| N/A 43C P0 61W / 250W | 11151MiB / 22919MiB | 99% Default |\r\n+-------------------------------+----------------------+----------------------+\r\n| 1 Tesla P40 Off | 00003984:00:00.0 Off | 0 |\r\n| N/A 41C P0 60W / 250W | 5979MiB / 22919MiB | 99% Default |\r\n+-------------------------------+----------------------+----------------------+\r\n| 2 Tesla P40 Off | 00005B97:00:00.0 Off | 0 |\r\n| N/A 38C P0 63W / 250W | 5979MiB / 22919MiB | 100% Default |\r\n+-------------------------------+----------------------+----------------------+\r\n| 3 Tesla P40 Off | 0000EA90:00:00.0 Off | 0 |\r\n| N/A 42C P0 61W / 250W | 5979MiB / 22919MiB | 99% Default |\r\n+-------------------------------+----------------------+----------------------+\r\n\r\n+-----------------------------------------------------------------------------+\r\n| Processes: GPU Memory |\r\n| GPU PID Type Process name Usage |\r\n|=============================================================================|\r\n| 0 5927 C python 11141MiB |\r\n| 1 5927 C python 5969MiB |\r\n| 2 5927 C python 5969MiB |\r\n| 3 5927 C python 5969MiB |\r\n+-----------------------------------------------------------------------------+", "Yes, there is no mechanism to balance memory in the examples. In NVIDIA's tests, it didn't help." ]

[]

[]

[ "Hi Judit:\r\n- Regarding the padding: you should send an `attention_mask` with the input if the input is smaller than the tensor you are sending in (see the description on `BertModel` in the README).\r\n- Regarding the seed: don't forget to put your model in eval mode (`model.eval()`) to disable the dropout layers.", "@thomwolf \r\n\r\nDespite the `attention_mask` the values are a slightly different.\r\n\r\nIt is normal that `[PAD]` vectors have different values?\r\n\r\n```\r\nfrom pytorch_transformers import BertModel\r\nfrom rest.run_glue import *\r\n\r\ntokenizer = BertTokenizer.from_pretrained('bert-base-uncased', do_lower_case=False)\r\nmodel = BertModel.from_pretrained('bert-base-uncased')\r\nmodel.eval()\r\n\r\ntorch.manual_seed(0)\r\nsent = \"this is a complicated sentence [SEP]\"\r\ntokens = ['[CLS]'] + tokenizer.tokenize(sent)\r\nids = tokenizer.convert_tokens_to_ids(tokens)\r\nt = torch.LongTensor([ids])\r\n\r\nwith torch.no_grad():\r\n out = model(t)[0]\r\n\r\ntorch.manual_seed(0)\r\nsent = \"this is a complicated sentence [SEP]\"\r\ntokens = ['[CLS]'] + tokenizer.tokenize(sent)\r\ntokens.extend(['[PAD]'] * 3)\r\nids = torch.tensor(tokenizer.convert_tokens_to_ids(tokens)).unsqueeze(0)\r\nmask = torch.zeros((1, ids.shape[1], ids.shape[1]), dtype=torch.float)\r\nmask[:, :, 0:-3] = 1.0\r\n\r\nwith torch.no_grad():\r\n out2 = model(ids, attention_mask = mask[:, 0])[0]\r\n\r\nprint('------------')\r\nfor i in range(out.shape[1]):\r\n print(i, out[0][0, i].item())\r\n\r\nprint('------------')\r\nfor i in range(out2.shape[1]):\r\n torch.manual_seed(0)\r\n print(i, out2[0][0, i].item())\r\n```\r\nhere is the output\r\n\r\n```\r\n------------\r\n0 -0.10266201943159103\r\n1 0.11214534193277359\r\n2 -0.1575649380683899\r\n3 -0.3163739740848541\r\n4 -0.4168904423713684\r\n5 -0.4069269001483917\r\n6 0.28849801421165466\r\n------------\r\n0 -0.10266169905662537\r\n1 0.1121453121304512\r\n2 -0.15756472945213318\r\n3 -0.3163738548755646\r\n4 -0.41689014434814453\r\n5 -0.40692687034606934\r\n6 0.288497656583786\r\n7 0.28312715888023376\r\n8 0.08457585424184799\r\n9 -0.3077544569969177\r\n```\r\n\r\n`[PAD]`'s are different, is that normal? \r\n\r\n**7 0.28312715888023376\r\n8 0.08457585424184799\r\n9 -0.3077544569969177**", "I am having same problem and couldn't find a reason or fix yet.", "Due to Position Embeddings every token results in different vectors.\r\nYou might want to google \"How the Embedding Layers in BERT Were Implemented\"", "> Due to Position Embeddings every token results in different vectors.\r\n\r\nCould you be more specific what is the source of this numerical instability? Perhaps refer to exact code? I am still not exactly sure why output changes slightly when using attention mask, when I use differently padded inputs. There should be no self-attention over padded inputs. Self-attention scores are set to large negative number before softmax:\r\n`attention_scores = attention_scores + attention_mask`\r\nCould it be that sometimes -10_000 might not be enough to get 0 from softmax? I have recorded differences at most in the order of 2e-6.\r\n\r\nOr is it because of arithmetic errors? According to https://en.wikipedia.org/wiki/Machine_epsilon, upped bound for the relative error in 32bit format is somewhere at 1.19e-07, which is still an order away. Could that be because of the error propagation through many FP32 operations?" ]

[]

[]

[ "You should use 16bit training (`--fp16` argument). You can use the dynamic loss scaling or tune the loss scale yourself if the results are not the best.", "@thomwolf Thanks! I enabled 16bit training and it took about 20min/epoch. Is that what you experienced?", "Sounds good.", "@thomwolf \r\nMay I know what is the expected EM & F1 score if users train for 2-3 epochs? I got 43 and 48 respectively.", "You can have a look at the readme examples but it should be a lot higher, around 88-90.\r\nMaybe your batch size is too small, look at the readme for more information." ]

[]

[]

[ "Hi @dileep1996, this has just been fixed in master (#275)!" ]

[]

[]

[ "Thanks @davidefiocco!" ]

[]

[]

[ "Hi,\r\nThere is no pretrained character-level model for text8 right now.\r\nOnly a word-level model trained on wikitext 103." ]

[]

[]

[]

[]

[]

[ "Yes, you need documents with multiple lines because only sentences from the same doc are used as positive examples for the nextSentence prediction. ", "Seems like the expected behavior. Feel free to open a PR to extend the example if you want @tuhinjubcse." ]

[ { "id": 1108649053, "node_id": "MDU6TGFiZWwxMTA4NjQ5MDUz", "url": "https://api.github.com/repos/huggingface/transformers/labels/Help%20wanted", "name": "Help wanted", "color": "008672", "default": false, "description": "Extra attention is needed, help appreciated" }, { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "Yeah, the re-encoding seems to fix the bug:\r\n```\r\nIn [6]: \"'Enquêtes\".encode('latin1').decode('utf8')\r\nOut[6]: \"'Enquêtes\"\r\n```", "which version of python are you using?", "It's python3.6. Does the snippet above gives different result on other version?\r\n\r\nJFYI: I'm using the script below to create a fixed vocab and save in the current folder.\r\n```(python)\r\nimport collections\r\nimport urllib.request\r\n\r\nfrom pytorch_pretrained_bert import tokenization_transfo_xl\r\nimport torch\r\n\r\n\r\ndef fix(x):\r\n return x.encode('latin1').decode('utf8')\r\n\r\n\r\ndef main():\r\n basedir = '.'\r\n good_vocab_path = basedir + '/' + tokenization_transfo_xl.VOCAB_NAME\r\n vocab_url = tokenization_transfo_xl.PRETRAINED_VOCAB_ARCHIVE_MAP['transfo-xl-wt103']\r\n urllib.request.urlretrieve(vocab_url, basedir + '/vocab.buggy.bin')\r\n vocab = torch.load(basedir + '/vocab.buggy.bin')\r\n\r\n vocab['counter'] = collections.Counter({fix(k): v for k, v in vocab['counter'].items()})\r\n vocab['sym2idx'] = {fix(k): v for k, v in vocab['sym2idx'].items()}\r\n vocab['idx2sym'] = [fix(k) for k in vocab['idx2sym']]\r\n torch.save(vocab, good_vocab_path)\r\n\r\n\r\nif __name__ == \"__main__\":\r\n main()\r\n```", "Maybe indeed. Do you want to submit a PR to fix this?", "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" ]

[]

[]

[ "This is only Transformer-XL related. For GPT the output is:\r\n\r\n```python\r\ntensor([[[ 0.1963, 0.0367, -0.2051, ..., 0.7062, -0.2786, 0.1352],\r\n [-0.4705, 0.1581, 0.0452, ..., 0.7809, -0.2519, 0.4257],\r\n [-0.2602, -0.7126, -0.7966, ..., 0.6364, -0.1560, -0.6084],\r\n ...,\r\n [-0.3665, 1.2743, -2.4027, ..., -1.7271, -1.7892, 0.7689],\r\n [-1.3017, 2.7999, -2.8868, ..., -1.3412, 0.2787, -0.0605],\r\n [ 0.2648, 0.3508, 0.2894, ..., -0.7471, 0.1855, -0.0492]]])\r\n```", "Same situation, on GPU.\r\n\r\n", "Indeed, there was a bug in the loading of the `TransfoXLModel` from the S3 dump (which is a converted `TransfoXLLMHeadModel`) so the weights were not loaded.\r\n\r\nYou can see that the weights are not loaded if you activate the logger before loading the model:\r\n```python\r\nimport logging\r\nlogging.basicConfig(level=logging.INFO)\r\n```\r\n\r\nI've fixed it in release 0.5.1.\r\nI've also fixed another issue you (@stefan-it) mentions in https://github.com/zalandoresearch/flair/issues/68 which is the dependency of `OpenAIGPTTokenizer` on SpaCy and ftfy by adding a fallback on BERT's `BasicTokenizer` (should be fine for normal usage, SpaCy+ftfy were included to exactly reproduce the paper's pre-processing steps).", "Publishing 0.5.1 as soon as all the tests are checked.", "Ok 0.5.1 is published: https://github.com/huggingface/pytorch-pretrained-BERT/releases/tag/v0.5.1" ]

[]

[]

[ "Hi @hugochan, actually that what's in the `mems` of the Transformer-XL are (maybe you can read again the paper).\r\n\r\nOne thing to be careful about is that the `mems` have transposed first dimensions and are longer (see the readme). Here is how to extract the hidden states from the model output:\r\n```python\r\nhidden_states, mems = model(tokens_tensor)\r\nseq_length = hidden_states.size(1)\r\nlower_hidden_states = list(t[-seq_length:, ...].transpose(0, 1) for t in mems)\r\nall_hidden_states = lower_hidden_states + [hidden_states]\r\n```", "Hi @thomwolf , thank you for your answer! Just one quick question. It seems that `mems` already contains a list of num_layer hidden states, what is the difference between `lower_hidden_states[-1]` and `hidden_states` in your code? Thank you!", "Actually `mems` contains all the hidden states PLUS the output of the embeddings (`lower_hidden_states[0]`) so `lower_hidden_states[-1]` is the output of the hidden state of the layer below the last layer and `hidden_states` is the output of the last layer (before the softmax).\r\n\r\nI will add a note on that in the readme." ]

[]

[]

[ "Thanks @wangxiaodiu " ]

[]

[]

[ "Oh yes, that was a typo, there is only one example for Transformer-XL and it's the `run_transfo_xl.py` file detailed [here](https://github.com/huggingface/pytorch-pretrained-BERT/blob/master/README.md#openai-gpt-and-transformer-xl-running-the-examples). Read the API details in the readme for more information on the input/outputs of the two Transformer-XL models.\r\nI've updated the readme, thanks." ]

[]

[]

[ "Though we are not facing the same issue……\r\n\r\nI followed the instruction from the readme, the `tokenized_text` is expected by assertion to be:\r\n `['[CLS]', 'who', 'was', 'jim', 'henson', '?', '[SEP]', 'jim', '[MASK]', 'was', 'a', 'puppet', '##eer', '[SEP]']`. \r\n\r\nHowever, the actual `tokenized_text` is:\r\n`['[CLS]', 'who', 'was', 'jim', 'henson', '?', '[MASK]', 'jim', 'henson', 'was', 'a', 'puppet', '##eer', '[SEP]']`.\r\n\r\nI believe this is because of `masked_index = 6` in the example code from the readme. If you let it to `masked_index = 8`, everything is perfect.\r\n\r\nI opened a PR to fix this minor bug.", "I think the tokenizer issue has been resolved in the latest version (0.5.0). ", "Yes!", "I just encountered the same issue as @dhirajmadan1 with `transformers==2.2.1`. Is this expected somehow?\r\n\r\nI am following the quickstart guide: https://huggingface.co/transformers/quickstart.html\r\n\r\n```\r\ntokenizer = BertTokenizer.from_pretrained('bert-base-uncased')\r\n\r\n# Run an example text through this:\r\ntext = \"[CLS] Who was Jim Henson ? [SEP] Jim Henson was a puppeteer [SEP]\"\r\ntokenized_text = tokenizer.tokenize(text)\r\n\r\nmasked_index = 8\r\ntokenized_text[masked_index] = '[MASK]'\r\npredicted_tokenized_sentence = ['[CLS]', 'who', 'was', 'jim', 'henson', '?', '[SEP]', 'jim', '[MASK]', 'was', 'a', 'puppet', '##eer', '[SEP]']\r\n \r\n```", "What is your issue @yenicelik? Do you mind opening a new issue with your problem?", "ah, my apologies: https://github.com/huggingface/transformers/issues/2047\r\n\r\napparently a PR is on the way!" ]

[]

[]

[ "Hi Carolin,\r\n\r\nDepending on the model you are using, not all the weights are initialized from the pre-trained models. Check the details in the [overview](https://github.com/huggingface/pytorch-pretrained-BERT#overview) section of the readme to see if it's the case for you.\r\n\r\nApart from weights initialization and dataset shuffling other typical source of variances are the dropout layers.\r\n\r\nBert fine-tuning has been reported to be a high-variance process indeed, in particular on small datasets.", "Hi Thomas,\r\n\r\nthanks for the quick reply!\r\n\r\nI'm using `BertForMaskedLM`, so the weights should be set. But yes, I didn't think of dropout, thanks for pointing that out!\r\n" ]

[]

[]

[ "reduce batch size?", "It is 32 as of now . What do you think I should reduce it to ?", "Start very low and increase while looking at `nvidia-smi` or a similar GPU memory visualization tool.", "Closing this for now, feel free to re-open if you have other issues." ]

[]

[]

[ "Hi Jie,\r\n`extract_feature.py` is an example script. If you want to adapt it for sentences-pair, we would be happy to welcome a PR :)" ]

[ { "id": 1108649053, "node_id": "MDU6TGFiZWwxMTA4NjQ5MDUz", "url": "https://api.github.com/repos/huggingface/transformers/labels/Help%20wanted", "name": "Help wanted", "color": "008672", "default": false, "description": "Extra attention is needed, help appreciated" }, { "id": 1108649070, "node_id": "MDU6TGFiZWwxMTA4NjQ5MDcw", "url": "https://api.github.com/repos/huggingface/transformers/labels/Need%20more%20information", "name": "Need more information", "color": "d876e3", "default": false, "description": "Further information is requested" }, { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "I know this is closed, but I'm running into a similar issue. For the first ~100 batches, the script runs with an OK speed (1s/it, batch size 64, 512 tokens, 8x GTX 1080Ti) (specific for `DistilBertForSequenceClassification` in my case).\r\nAfter that, the speed drops significantly, to about 10s/it, with the GPUs mostly sitting idle. (0% on `gpustat`.\r\n\r\nAny idea on what could be causing this?", "Update: Looks like I was accumulating the gradient for too long.", "> Update: Looks like I was accumulating the gradient for too long.\r\n\r\n@ArthurCamara So, how did you fix the problem in your case? Did you change the `gradient_accumulation_steps` parameter of `Trainer`? What were the initial value and the value which helped to resolve the problem?\r\nAnd how did you understand that the problem was really in this?", "I have the same problem. I noticed that the training speed slows down as GPU temperature goes up... When the temperature goes down (if I wait after terminating the process), the speed becomes okay again.\r\n\r\nThis issue happens only when I use `Trainer`. When I don't use it (i.e. use PyTorch utilities directly), the training speed is stable and the temperature doesn't go up.\r\n\r\n@taesikna Did you fix your issue?", "> > Update: Looks like I was accumulating the gradient for too long.\r\n> \r\n> @ArthurCamara So, how did you fix the problem in your case? Did you change the `gradient_accumulation_steps` parameter of `Trainer`? What were the initial value and the value which helped to resolve the problem?\r\n> And how did you understand that the problem was really in this?\r\n\r\nI think the initial setting was 5 or something. I dropped to 1 and it was fine then. ", "> > > Update: Looks like I was accumulating the gradient for too long.\n> > \n> > @ArthurCamara So, how did you fix the problem in your case? Did you change the `gradient_accumulation_steps` parameter of `Trainer`? What were the initial value and the value which helped to resolve the problem?\n> > And how did you understand that the problem was really in this?\n> \n> I think the initial setting was 5 or something. I dropped to 1 and it was fine then. \n\nAs I see now the default value is 1, but I still observe the slow down. @ArthurCamara Do you have any idea on that?", "> > > > Update: Looks like I was accumulating the gradient for too long.\r\n> > > \r\n> > > \r\n> > > @ArthurCamara So, how did you fix the problem in your case? Did you change the `gradient_accumulation_steps` parameter of `Trainer`? What were the initial value and the value which helped to resolve the problem?\r\n> > > And how did you understand that the problem was really in this?\r\n> > \r\n> > \r\n> > I think the initial setting was 5 or something. I dropped to 1 and it was fine then.\r\n> \r\n> As I see now the default value is 1, but I still observe the slow down. @ArthurCamara Do you have any idea on that?\r\n\r\nI was not using the Trainer, but my own training loop. dropping the accumulation steps to 1 helped because it was overwhelming the GPUs memory and that makes the GPUs sit idly. If the GPUs on `nvidia-smi` are idle, but their memory is full, it's probably something related to memory usage. Otherwise, no idea. ", "\r\n\r\n> facing this same issue too, on 2080. will not use Trainer.\r\n\r\n```\r\nen ignored: tokens, ner_tags, id.\r\n[INFO|trainer.py:1156] 2021-06-05 16:46:31,375 >> ***** Running training *****\r\n[INFO|trainer.py:1157] 2021-06-05 16:46:31,386 >> Num examples = 2021\r\n[INFO|trainer.py:1158] 2021-06-05 16:46:31,397 >> Num Epochs = 1\r\n[INFO|trainer.py:1159] 2021-06-05 16:46:31,407 >> Instantaneous batch size per device = 10\r\n[INFO|trainer.py:1160] 2021-06-05 16:46:31,418 >> Total train batch size (w. parallel, distributed & accumulation) = 10\r\n[INFO|trainer.py:1161] 2021-06-05 16:46:31,428 >> Gradient Accumulation steps = 1\r\n[INFO|trainer.py:1162] 2021-06-05 16:46:31,439 >> Total optimization steps = 203\r\n 1%|█▍ | 2/203 [00:12<16:46, 5.01s/it\r\n 1%|██ | 3/203 [00:20<19:53, 5.97s/it\r\n 2%|██▊ | 4/203 [00:37<30:55, 9.33s/i\r\n 2%|███▍ | 5/203 [00:50<34:44, 10.53s/\r\n 3%|████▏ | 6/203 [01:01<34:07, 10.39s\r\n 3%|████▊ | 7/203 [01:02<25:16, 7.74s\r\n 4%|█████▌ | 8/203 [01:16<31:19, 9.64\r\n 4%|██████▎ | 9/203 [01:30<34:47, 10.7\r\n 5%|██████▉ | 10/203 [01:43<37:30, 11.6\r\n 5%|███████▌ | 11/203 [01:56<38:22, 11.\r\n 6%|████████▎ | 12/203 [02:09<38:37, 12\r\n 6%|████████▉ | 13/203 [02:23<40:23, 12\r\n 7%|█████████▋ | 14/203 [02:33<37:51, 1\r\n 7%|██████████▎ | 15/203 [02:44<36:31,\r\n 8%|███████████ | 16/203 [02:55<35:39,\r\n 8%|███████████▋ | 17/203 [03:07<36:25,\r\n 9%|████████████▍ | 18/203 [03:20<37:19\r\n 9%|█████████████ | 19/203 [03:34<38:25\r\n 10%|█████████████▊ | 20/203 [03:47<38:4\r\n 10%|██████████████▍ | 21/203 [04:00<38:\r\n 11%|███████████████▏ | 22/203 [04:13<39\r\n 11%|███████████████▊ | 23/203 [04:27<39\r\n:42, 13.24s/it]\r\n```" ]

[]

[]

[ "Nice!" ]

[ { "id": 1108649053, "node_id": "MDU6TGFiZWwxMTA4NjQ5MDUz", "url": "https://api.github.com/repos/huggingface/transformers/labels/Help%20wanted", "name": "Help wanted", "color": "008672", "default": false, "description": "Extra attention is needed, help appreciated" }, { "id": 1314768611, "node_id": "MDU6TGFiZWwxMzE0NzY4NjEx", "url": "https://api.github.com/repos/huggingface/transformers/labels/wontfix", "name": "wontfix", "color": "ffffff", "default": true, "description": null } ]

[]

[ "Hi @WilliamTambellini, have you tried to follow the standard ONNX procedure for converting a PyTorch model?\r\nThe model in this repo are just regular PyTorch models.", "Hello Thomas, I ve not yet tried, just seen : \r\nhttps://github.com/onnx/models/issues/130\r\nhttps://stackoverflow.com/questions/54220042/how-do-you-generate-an-onnx-representation-of-a-pytorch-bert-pretrained-neural-n\r\nWill try, tks. ", "Hi, when I try to export a TokenClassification model to a ONNX model, I encounter `RuntimeError: ONNX export failed: Couldn't export operator aten::erf`, does that mean some part of BERT model layers not supported by ONNX?\r\nI think that problem comes from the definition of GELU function, which is `x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))`. Should I try to use other way to calculate this function or wait for ONNX to support this opertator? ", "@geekboood update your pytorch version to latest and the problem will most likely go away.", "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", "for anyone who is looking for the answer:\r\ntorch=1.1.0 \r\npython=3.6\r\n\r\n`torch.onnx.export(model, (input_ids, segment_ids, input_mask), \"bert.onnx\", verbose=False)`\r\n\r\nworks well for me", "> for anyone who is looking for the answer:\r\n> torch=1.1.0\r\n> python=3.6\r\n> \r\n> `torch.onnx.export(model, (input_ids, segment_ids, input_mask), \"bert.onnx\", verbose=False)`\r\n> \r\n> works well for me\r\n\r\nHi, thanks for the answer. Do you get good results when using the exported model for inference in another framework? I exported a BertForQuestionAnswering model to ONNX without errors, but I'm getting wrong predictions when using onnxruntime or a second export to TF Serving and I can't figure out why!", "Not sure if this is still an issue for you but in the BertForSequenceClassification model the parameters are in a different order\r\n\r\n`torch.onnx.export(model, (input_ids, input_mask, segment_ids), \"bert.onnx\", verbose=False)`\r\n\r\nworks as intended", "@chessgecko wow you're right, thanks! working now", "cc @mfuntowicz :)" ]

[]

[]

[ "Indeed, this will be in the next release, thanks!", "is there is a change according the CPU issue ?", "Should be fixed now. Do you still have an error?", "i am loading the model and i dont know how to load on CPU , gives me \"model.to\" to is not defined. Can you tell me how to send the model to run on CPU if trained on GPU.", "Would love some explanation on how to do this as well!", "I have the same question. how to load a model trained on GPU to CPU?", "I am watching this as well.", "I have the same question. Is there any option for using CPU-only?", "watching this as well", "same" ]

[]

[]

[ "Thanks @BoeingX !" ]

[]

[]

[ "Yes, feel free to submit a PR. Otherwise, I'll fix it in the next release." ]

[]

[]

[ "No, there's no evaluation step in the example script yet. What I can recommend is using your downstream task for evaluation of the pretrained BERT. Alternatively, you could of course also add some evaluation of the LM / nextSentence loss on a validation set.", "Perhaps for clarity then, that parameter should be taken out of the script? ", "Sure, makes sense. I created a PR. Thanks for pointing this out @bsugerman .", "Fixed in master now." ]

[]

[]

[ "Hi @chenyangh,\r\nYou need to install `apex` with the C++ and CUDA extensions:\r\n```bash\r\ngit clone https://github.com/NVIDIA/apex.git\r\ncd apex\r\npython setup.py install --cuda_ext --cpp_ext\r\n```", "@thomwolf \r\nThanks!", "@thomwolf After doing what you wrote, I got this error.\r\n\r\ntorch.__version__ = 1.0.1.post2\r\nTraceback (most recent call last):\r\n File \"setup.py\", line 60, in <module>\r\n raise RuntimeError(\"--cuda_ext was requested, but nvcc was not found. Are you sure your environment has nvcc available? If you're installing within a container from https://hub.docker.com/r/pytorch/pytorch, only images whose names contain 'devel' will provide nvcc.\")\r\nRuntimeError: --cuda_ext was requested, but nvcc was not found. Are you sure your environment has nvcc available? If you're installing within a container from https://hub.docker.com/r/pytorch/pytorch, only images whose names contain 'devel' will provide nvcc.\r\n\r\nWhat else should I do for this?", "You should refer to apex installation instructions.\r\nApex has slightly changed since my comment so best is to go read NVIDIA's README and installation instructions here: https://github.com/NVIDIA/apex", "@kbulutozler You can change pytorch docker image version to pytorch/pytorch:1.3-cuda10.1-cudnn7-devel" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "In the examples scripts, it tunes the whole model.\r\nBut BERT models classes are just regular PyTorch `nn.Modules` so you can also freeze layer like you would do in any PyTorch module." ]

[]

[]

[ "Thanks Thibault!" ]

[]

[]

[ "Thanks @cooelf, there was a PR merging `run_squad` and `run_squad2` that also fixed this issue." ]

[]

[]

[ "The weights of the BERT base model are getting updated while finetuning the network.", "Indeed" ]

[]

[]

[ "Thanks @JoeDumoulin " ]

[]

[]

[ "Hey, I am working on something similar. I feel like the original code might be incorrect. They seem to directly take the output of the model as 'loss' without applying any criteria. But I might be totally wrong. ", "Hey! :)\r\n\r\nReally? What do you mean by criteria?\r\n\r\nI tried to artificially change my dataset so that the target outputs are\r\n[1, 0, 0,..., 0] for every sample. I wanted to see whether the model was\r\nable to learn this dummy case. However, it fails and it predicts the exact\r\nopposite, i.e. [0, 1, 1,.... 1]. That's why I think I must be doing\r\nsomething wrong somewhere.\r\n", "Hi @nicolas-mng @zhipeng-fan Did you guys manage to get a multilabel problem to work ? Could you please share a gist ?", "Hi, you should have a look here: https://medium.com/huggingface/multi-label-text-classification-using-bert-the-mighty-transformer-69714fa3fb3d ;) " ]

[]

[]

[]

[]

[]

[ "Hi Joel, yes the new release (0.5.0) is pretty much ready (remaining work on branches `fifth-release` and `transfo-xl` to finish testing the newly added pre-trained OpenAI GPT and Transformer-XL).\r\n\r\nLikely next week.", "Awesome, thanks!\n\nOn Thu, Jan 31, 2019, 11:06 AM Thomas Wolf <[email protected]> wrote:\n\n> Hi Joel, yes the new release (0.5.0) is pretty much ready (remaining work\n> on branches fifth-release and transfo-xl to finish testing the newly\n> added pre-trained OpenAI GPT and Transformer-XL).\n>\n> Likely next week.\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/pytorch-pretrained-BERT/issues/245#issuecomment-459506373>,\n> or mute the thread\n> <https://github.com/notifications/unsubscribe-auth/ABP2mYQxXrEKfGvJhESKqKIQzI2dvcolks5vI1rUgaJpZM4ac94Y>\n> .\n>\n", "Ok @joelgrus, the new release is out: https://github.com/huggingface/pytorch-pretrained-BERT/releases/tag/v0.5.0" ]

[]

[]

[ "Great, thanks @tholor!" ]