Datasets:

---

DanielPFlorian

/

Transformers-Github-Issues

like 0

[]

[]

[ "@LysandreJik Is this PR done with your last changes in the tests?", "Ah, I had forgotten about this. I'll rebase and ping you for review\r\n", "should be good for review @jplu ", "They're not the same as they don't rely on the full checkpoints but on some random tiny ones, to make the CI faster.\r\nIt does test every same aspect, however: the weights loading, the full inference, the expected results." ]

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

[]

[ "Hi, could you mention the issue you're having?", "I want to use a natural question dataset and model trained on that. I have seen this code:\r\n```\r\nfrom transformers import DPRReader, DPRReaderTokenizer\r\ntokenizer = DPRReaderTokenizer.from_pretrained('facebook/dpr-reader-single-nq-base')\r\nmodel = DPRReader.from_pretrained('facebook/dpr-reader-single-nq-base', return_dict=True)\r\nencoded_inputs = tokenizer(\r\n questions=[\"What is love ?\"],\r\n titles=[\"Haddaway\"],\r\n texts=[\"'What Is Love' is a song recorded by the artist Haddaway\"],\r\n return_tensors='pt'\r\n )\r\noutputs = model(**encoded_inputs)\r\nstart_logits = outputs.stat_logits\r\nend_logits = outputs.end_logits\r\nrelevance_logits = outputs.relevance_logits\r\n```\r\nBut from here I can not know how to return the answer from the model. It gives us just starting and ending position of the answer as far as I understand.\r\nAnd also CAn I use my document as a context, and model search answer in this document. If yes please tell me how it is possible.", "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" ]

[]

[]

[ "> \r\n> \r\n> Awesome! Did you try it with a usual training example to see if everything is ok?\r\n\r\nYes, with example/text-classification.\r\nI didn't check yet with a usual `tf.keras.models.Model` (i.e. not TFPretrainedModel). But when I continue with `test_trainer_tf.py`, it will be tested.", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/master/CONTRIBUTING.md) are likely to be ignored.", "Hi, @sgugger & @LysandreJik ,\r\n\r\nI didn't realized that this PR is not merged into master. Since it has been for some time, I rebased the branch. All the suggestions from @sgugger are done in the latest version.\r\n\r\nIt would be great if you can merge this PR. Thanks!", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/master/CONTRIBUTING.md) are likely to be ignored." ]

[]

[]

[ "Hey @gmftbyGMFTBY - I think the better approach to tackle your problem would actually be this one here: https://github.com/huggingface/transformers/issues/3021#issuecomment-681792104 .\r\n\r\nThis way you should not run into any errors regarding the position_ids", "Hey, @patrickvonplaten, it works for me. Thank you so mcuh." ]

[]

[]

[ "This can happen if you're fetching a model from S3 but you have no internet access, or if you're using an incorrect URL to a local folder.", "Hello @LysandreJik \r\n\r\nI have uploaded the saved model in a folder on my EC2 instance. Therefore, the location for the model is from the instance file directory which I have verified multiple times.\r\n\r\nAlso, the model functions properly when deployed using Flask over localhost. \r\n\r\nDo I need to download the pre-trained models as a command in the dockerfile? \r\n\r\nKindly help.\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", "I'm seeing the same issue on my community model\r\n\r\n```\r\nOSError: Can't load config for 'model_path'. Make sure that:\r\n\r\n- 'model_path' is a correct model identifier listed on 'https://huggingface.co/models'\r\n\r\n- or 'model_path' is the correct path to a directory containing a config.json file\r\n```\r\n\r\nI verified the model files are there. How can i work around this?", "It would be helpful if you opened a new issue with everything related to your environment, as well as the code you use. Are you also on EC2? What is in the `model_path` folder? What is your `transformers` version? All that's asked in the template would be very helpful for us to help you.", "@LysandreJik Thanks for the reply! I created a new issue https://github.com/huggingface/transformers/issues/9106. Other old huggingtweets models still work but not the new ones, not sure what the problem is.", "> OSError: Can't load config for 'model_path'. Make sure that:\r\n> \r\n> - 'model_path' is a correct model identifier listed on 'https://huggingface.co/models'\r\n\r\nHye I'm facing the same. Did you solve that ? ", "I have the same error message. It also says that it can't find url\"/resolve/main/config.json\". I saved my model like they said but only have a folder \"results\" containing \"pytorch_model.bin\" and \"training_args.bin\".\r\n\r\nEdit: I tried to also save the tokenizer (despite only having fine-tuned). This gave me a tokenizer_config.json which still isn't enough.\r\n\r\nHow do I get a config.json in my directory? I'm using a custom BERT modeled after BertForTokenClassification (https://huggingface.co/transformers/_modules/transformers/models/bert/modeling_bert.html#BertForTokenClassification) which doesn't specify a config attribute.", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/master/CONTRIBUTING.md) are likely to be ignored." ]

[]

[]

[]

[ { "id": 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" ]

[]

[]

[ "Sorry duplicated my request." ]

[]

[]

[ "Indeed! Do you want to open a PR fixing the doc?", "Yes, I can do that. :)" ]

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

[]

[]

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

[]

[]

[]

[]

[ "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", "I was wondering if there was any follow up on this topic as I'd also be interested in continued pertaining on bart-base." ]

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

[]

[ "That's really cool, thanks for sharing @iliaschalkidis ", "Thanks @julien-c for your nice comments and for building and improving such a great library. Is there any chance, that we could place all 5 LEGAL-BERT variants in a sub-folder, i.e., `/legal-bert`, inside the account folder `/nlpaueb`? Kind of OCD though 🤓 \r\n", "I'm not sure what you mean :)\r\n\r\nDo you want to e.g. rename `bert-base-uncased-contracts` to `legal-bert-base-uncased-contracts`? Or do you want `nlpaueb/legal-bert/bert-base-uncased-contracts`? We don't really want to do the latter IMO (have increased levels of nesting) because:\r\n- I'm afraid it might get confusing for users of the models,\r\n- some of the tooling we are currently building is expecting a org_name/model_name layout.\r\n\r\nWhat do you think?", "I was referring to the second scenario, but I totally understand it will make things more complicated on your side. Thanks again!\r\n\r\n" ]

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

[]

[]

[]

[]

[ "> If that's all it takes, that's fantastic! Did you manage to obtain the performance improvements that were initially mentioned thanks to this?\r\n\r\nOn my machine with my GPU yes.\r\n\r\n>Also I'm realizing now that we don't have integration testing for our TensorFlow models, and this seems like a situation where having some would be needed. Could we work on adding these tests for the models modified here at first, and then add it to the rest of the models?\r\n\r\nSure! It is a good idea!\r\n\r\n> I can help you work on it if you're lacking time!\r\n\r\nI would appreciate if you have time yes 😃 ", "Okay, will take a look at doing the integrations tests sometimes tonight. Will let you know!", "@jplu \r\n\r\nFor learning purpose, I am wondering which operations was done on CPU instead of GPU. I saw you changed `Dense` to `EinsumDense` in several places, and remove several operations about shape changing. Is shape changing done on CPU and `EinsumDense` could avoid this? Could you give me some information about this, so I can read and learn it? Thanks.\r\n", "@chiapas \r\n\r\nIf you take a look at #6771 is it quite well detailed. The issue was coming from transpose+matmul that was done on CPU. einsumDense allows you to do all these computation directly in the layer but at the cost of changing the shapes of the original layers, that why we have modified the way we load the TF models.\r\n\r\nTo do this PR I basically took example on the original BERT implementation right [here](https://github.com/tensorflow/models/blob/master/official/nlp/transformer/attention_layer.py).\r\n\r\n", "Thanks a lot @LysandreJik !!\r\n\r\nAs I'm currently working on from scratch LM training for TF models, I don't have much time to really focus on this.", "> transpose+matmul\r\n\r\n@jplu Thanks. I am superised by this `transpose+matmul that was done on CPU`.", "> \r\n> \r\n> Thanks a lot @LysandreJik !!\r\n> \r\n> As I'm currently working on from scratch LM training for TF models, I don't have much time to really focus on this.\r\n\r\n@jplu You also works on LM training for TF models? I plan to go back to a pending PR #6955 I created once the `test_trainer_tf.py` is done. Do PR #6955 and your work on TF models LM training overlap? Currently that PR is still empty though.", "@chiapas This is exactly what I'm doing, and the models needs some rework that's why I'm mostly focus on BERT to have at least one model working.\r\n\r\nI just done yesterday the data pipeline with random masking generation.", "> @chiapas This is exactly what I'm doing, and the models needs some rework that's why I'm mostly focus on BERT to have at least one model working.\r\n> \r\n> I just done yesterday the data pipeline with random masking generation.\r\n\r\nAh, ok. I guess my PR was pending too long and it is my bad not to communicate with you first. I planed to do this while I finished a notebook on Kaggle [Masked, My Dear Watson - MLM with TPU](https://www.kaggle.com/yihdarshieh/masked-my-dear-watson-mlm-with-tpu), which also works on MLM.\r\n\r\nSince you already have more progresses (and also you are HF member), it is better for you to continue. However, if there is something I can contribute for this TF LM task, I would love to do it.\r\n\r\n", "> Since you already have more progresses (and also you are HF member), it is better for you to continue. However, if there is something I can contribute for this TF LM task, I would love to do it.\r\n\r\nThanks! I will let you know.", "That's awesome! I will see what results the TF benchmark scripts give before/after this PR.\r\n\r\nStrongly agree with @LysandreJik that we should add integration tests before merging this PR.", "I ran the benchmarks: `python examples/benchmarking/run_benchmark_tf.py --models bert-base-cased --env_print` in the following environment:\r\n\r\n```\r\n- transformers_version: 3.3.1 \r\n- framework: TensorFlow \r\n- eager_mode: False \r\n- use_xla: False \r\n- framework_version: 2.3.0 \r\n- python_version: 3.6.10 \r\n- system: Linux \r\n- cpu: x86_64 \r\n- architecture: 64bit \r\n- date: 2020-10-06 \r\n- time: 19:06:48.378935 \r\n- fp16: False \r\n- use_multiprocessing: True \r\n- only_pretrain_model: False \r\n- cpu_ram_mb: 32088 \r\n- use_gpu: True \r\n- num_gpus: 1 \r\n- gpu: TITAN RTX \r\n- gpu_ram_mb: 24217 \r\n- gpu_power_watts: 280.0 \r\n- gpu_performance_state: 8 \r\n- use_tpu: False \r\n```\r\n\r\n\r\nCurrently, on master:\r\n\r\n``` \r\n==================== INFERENCE - SPEED - RESULT ====================\r\n--------------------------------------------------------------------------------\r\n Model Name Batch Size Seq Length Time in s \r\n--------------------------------------------------------------------------------\r\n bert-base-cased 8 8 0.085 \r\n bert-base-cased 8 32 0.166 \r\n bert-base-cased 8 128 0.513 \r\n bert-base-cased 8 512 2.629 \r\n-------------------------------------------------------------------------------- \r\n```\r\n\r\nIn this `tf-optim` branch, the results are:\r\n\r\n```\r\n==================== INFERENCE - SPEED - RESULT ====================\r\n--------------------------------------------------------------------------------\r\n Model Name Batch Size Seq Length Time in s \r\n--------------------------------------------------------------------------------\r\n bert-base-cased 8 8 0.088 \r\n bert-base-cased 8 32 0.176 \r\n bert-base-cased 8 128 0.531 \r\n bert-base-cased 8 512 3.028 \r\n--------------------------------------------------------------------------------\r\n```\r\n\r\n=> So the speed results are more or less identical with the way the benchmarks are used.\r\n\r\nI don't compile the model with Keras, but just add the \"@tf.function\" decorator to the function to transform the function into graph mode. So not sure what to think of that.... => @jplu - colud you maybe check the benchmark script and see if you can get a speed-up there? Or if the benchmark script is wrong? \r\n\r\n```\r\npython examples/benchmarking/run_benchmark_tf.py --models bert-base-cased --env_print\r\n```", "The benchmark script is ok, but to see the difference you have to create a saved_model and run the model in TF Serving. Your benchmark don't take into account all the optimization TF serving does for inference.\r\n\r\nWe should update the benchmark script to include:\r\n\r\n- Saved model creation\r\n- run a the saved model with the TF Serving tool\r\n- adapt the the benchmark to include gRPC calls to use the model from TF Serving.", "Will be integrated into the PR #7753" ]

[]

[]

[ "Closing this and moving the conversation (w/ my answer) to [the forums](https://discuss.huggingface.co/t/way-to-make-inference-zero-shot-pipeline-faster/1384/2?u=joeddav)." ]

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

[]

[ "Thanks!" ]

[]

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

[ "Hey @shamanez - what do you mean by \"update frequency\"? You don't need to use 8 GPUS => you can just reduce the number of gpus as you wish and keep the \"same\" batch size by increasing the `gradient_accumulation_steps` - does this make sense? ", "Let's say if the effective batch size is 32 with 8 GPUs and I want to keep the same batch size with 4 GPUs, I just need to change the _gradient_accumulation_steps_ to 2 right?\r\n\r\n[Update_Freq](https://fairseq.readthedocs.io/en/latest/command_line_tools.html#fairseq-train) is what fairseq used to keep the effective batch size same with less number of GPUs.", "Yeah exactly, in `examples/rag/finetune.sh` the default is `gpus=8` and `gradient_accumalation_steps=1`. So if you want to use less gpus while keeping the same \"effective\" batch size you should increase `gradient_accumalation_steps` accordingly", "Thanks a lot. :) " ]

[]

[]

[ "Hi, indeed GPUs are not used when doing tokenization. There are no matrix operations and there's no need for heavy parallelization, so no need to rely on GPUs for this operation." ]

[]

[]

[ "I deleted my virtual environment and workspace, re-installed it, re-runed the above codes, and found that the above codes worked without error\r\n" ]

[]

[]

[]

[]

[]

[ "Hi @AdrienDS, \r\n\r\nThanks for suggesting the changes.\r\n\r\nDid you try building the image locally? I totally understand the motivation behind the use of the `-devel` layer parent, but I have concern regarding the final image size. Would it be possible for you to include the resulting size for th `devel` based image?\r\n\r\nOtherwise look good for me!\r\n", "Hi @mfuntowicz It does increase the size, from 4.46GB (v3.3.1) to 6.53GB for `transformers-pytorch-gpu`. \r\n\r\nIf it's too large, could we create a separate image ? (like: `transformers-pytorch-gpu-apex`)", "Ok, that shouldn't hurt too much, let's go!\r\n\r\nThanks for the contribution 👍 " ]

[]

[]

[ "Closing in favor of #7619 " ]

[]

[]

[ "Finally! Can't wait for this PR to be merged.\r\nI've briefly looked at the code and from my understanding it should support this case, but correct me if I'm wrong:\r\n\r\n_When saving checkpoints (including model weights as well as scheduler and optimizer states), I will be able to attach to this process and store the checkpoint in some external repository (i.e GCS / W&B artifact)_,\r\n\r\nright?", "Yes, you will be able to inject custom behavior to the saved checkpoint with the `on_save` event." ]

[]

[]

[]

[]

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

[ "This may be related to an incompatible pytorch or cuda version", "@mthielk - this can very well be due to the PyTorch version. Did you try with a more current version of PyTorch? ", "@patrickvonplaten I face the same issue with PyTorch version 1.4.0. ", "I can confirm that this error occurs with PyTorch version 1.4.0!", "Okey, after some internal discussion the error is the following. PyTorch changed its `torch.save()` method officially in PyTorch 1.6.0 (check https://github.com/pytorch/pytorch/releases for 1.6.0 under \"Deprecations\") which means that models saved with torch >= 1.6.0 are not loadable with torch <= 1.4.0 -> hence this error. So for RAG the minimum required torch version is torch 1.5.0 it seems. (thanks @sgugger @LysandreJik )" ]

[]

[]

[ "This breaks backwards compatibility on saved `classif_dropout`, but from my checks this is always set to 0 (so incorrect) anyways and will stay 0, so I'm not too concerned.", "```python\r\nfrom transformers import BartConfig\r\nconfig_to_save = BartConfig.from_pretrained('facebook/bart-base', classif_dropout=0.42)\r\nconfig_to_save.classif_dropout # AttributeError\r\n```" ]

[]

[]

[ "@Paul-Trax ,\r\n\r\nThe warning you saw is not because you have some `-1` in the labels. It is because the computation was done inside a tensorflow graph, which was compiled before the computation. While compiling a graph, the different branches are entered, so you saw the warning. Once the real computation begins, i.e. your labels and logits used for computation, everything is fine.\r\n\r\nAn example to see such effect is (note that there is a `@tf.function` before `compute_loss`):\r\n\r\n import tensorflow as tf\r\n from typing import Dict, List, Optional, Union\r\n import warnings\r\n\r\n\r\n def shape_list(x: tf.Tensor) -> List[int]:\r\n \"\"\"\r\n Deal with dynamic shape in tensorflow cleanly.\r\n\r\n Args:\r\n x (:obj:`tf.Tensor`): The tensor we want the shape of.\r\n\r\n Returns:\r\n :obj:`List[int]`: The shape of the tensor as a list.\r\n \"\"\"\r\n static = x.shape.as_list()\r\n dynamic = tf.shape(x)\r\n return [dynamic[i] if s is None else s for i, s in enumerate(static)]\r\n\r\n\r\n @tf.function\r\n def compute_loss(labels, logits):\r\n\r\n loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(\r\n from_logits=True, reduction=tf.keras.losses.Reduction.NONE\r\n )\r\n # make sure only labels that are not equal to -100\r\n # are taken into account as loss\r\n if tf.math.reduce_any(labels == -1):\r\n warnings.warn(\"Using `-1` to mask the loss for the token is deprecated. Please use `-100` instead.\")\r\n active_loss = tf.reshape(labels, (-1,)) != -1\r\n print(f'During graph compiling - branch 1: {labels}')\r\n tf.print(f'Executed in graph - branch 1: {labels}')\r\n else:\r\n active_loss = tf.reshape(labels, (-1,)) != -100\r\n print(f'During graph compiling - branch 2: {labels}')\r\n tf.print(f'Executed in graph - branch 2: {labels}')\r\n\r\n reduced_logits = tf.boolean_mask(tf.reshape(logits, (-1, shape_list(logits)[2])), active_loss)\r\n\r\n labels = tf.boolean_mask(tf.reshape(labels, (-1,)), active_loss)\r\n\r\n return loss_fn(labels, reduced_logits)\r\n\r\n\r\n batch_size = 3\r\n seq_len = 5\r\n dim = 4\r\n labels = tf.constant(0, shape=[batch_size, seq_len])\r\n logits = tf.random.uniform(shape=[batch_size, seq_len, dim])\r\n\r\n loss = compute_loss(labels, logits)\r\n\r\n print(f'loss = {loss}')\r\n\r\nYou will see something like\r\n\r\n /home/imo/Desktop/venv/lib/python3.6/site-packages/tensorflow/python/autograph/impl/api.py:493: UserWarning: Using `-1` to mask the loss for the token is deprecated. Please use `-100` instead.\r\n return py_builtins.overload_of(f)(*args)\r\n During graph compiling - branch 1: Tensor(\"labels:0\", shape=(3, 5), dtype=int32)\r\n During graph compiling - branch 2: Tensor(\"labels:0\", shape=(3, 5), dtype=int32)\r\n Executed in graph - branch 2: Tensor(\"labels:0\", shape=(3, 5), dtype=int32)\r\n loss = [1.5343634 1.610856 1.433133 1.4082022 1.5018827 1.0152265 1.563687\r\n 1.2404382 1.1259079 1.7140993 1.4652599 1.6314502 1.5104814 1.389543\r\n 1.45472 ]\r\n\r\nIf you remove the `tf.function` above `compute_loss`, there is no graph compiled, and you won't see the warning you had.\r\n\r\n During graph compiling - branch 2: [[0 0 0 0 0] # There is no graph compiled, it is just our print statement.\r\n [0 0 0 0 0]\r\n [0 0 0 0 0]]\r\n Executed in graph - branch 2: [[0 0 0 0 0]\r\n [0 0 0 0 0]\r\n [0 0 0 0 0]]\r\n loss = [1.648417 1.2457228 1.5540932 1.7658947 1.4607204 1.529434 1.3607037\r\n 1.6142995 0.9669408 1.316714 1.3906621 1.689343 1.3678703 1.324768\r\n 1.5207067]\r\n\r\nWhen `TFTrainer` is used, the computation is done in graph model.", "Hi chiapas,\r\nThanks a lot for the detailed answer, it was really helpful!\r\nI am closing the issue." ]

[]

[ { "login": "sshleifer", "id": 6045025, "node_id": "MDQ6VXNlcjYwNDUwMjU=", "avatar_url": "https://avatars.githubusercontent.com/u/6045025?v=4", "gravatar_id": "", "url": "https://api.github.com/users/sshleifer", "html_url": "https://github.com/sshleifer", "followers_url": "https://api.github.com/users/sshleifer/followers", "following_url": "https://api.github.com/users/sshleifer/following{/other_user}", "gists_url": "https://api.github.com/users/sshleifer/gists{/gist_id}", "starred_url": "https://api.github.com/users/sshleifer/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/sshleifer/subscriptions", "organizations_url": "https://api.github.com/users/sshleifer/orgs", "repos_url": "https://api.github.com/users/sshleifer/repos", "events_url": "https://api.github.com/users/sshleifer/events{/privacy}", "received_events_url": "https://api.github.com/users/sshleifer/received_events", "type": "User", "site_admin": false } ]

[ "The input argument for `BartConfig.__init__()` should be named `classif_dropout ` instead of `classifier_dropout`", "Great catch, thanks!" ]

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

[]

[]

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

[]

[ "This seems like more of a TPU issue than a `huggingface/transformers` issue. Do you mind copying the full output of your command? Maybe in a `pastebin` or a github gist if it doesn't fit here.", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

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

[ "The XLNet model outputs some past states called `mems` at index 2. Those can't be concatenated together because they have a sequence length that varies. You should pass along `--past_index 2` to your script so that:\r\n1. those `mems` are used\r\n2. they are discarded from the predictions, and thus evaluation should work.\r\n\r\nWe will have something easier to use in the future, but for now it should work around your problem.", "Thanks for your fast reply. Unfortunately ```--past_index 2``` doesn't work for me. \r\nNew error logs\r\n```bash\r\n10/05/2020 22:55:40 - INFO - filelock - Lock 140417916796544 acquired on /data/home/liusishun/glue/SST-2/cached_dev_XLNetTokenizer_64_sst-2.lock\r\n10/05/2020 22:55:41 - INFO - filelock - Lock 140417916796544 released on /data/home/liusishun/glue/SST-2/cached_dev_XLNetTokenizer_64_sst-2.lock\r\n10/05/2020 22:55:44 - INFO - __main__ - *** Evaluate ***\r\nEvaluation: 93%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▉ | 13/14 [00:09<00:00, 1.41it/s]\r\nTraceback (most recent call last):\r\n File \"/data/home/liusishun/applications/transformers/examples/text-classification/run_glue.py\", line 247, in <module>\r\n main()\r\n File \"/data/home/liusishun/applications/transformers/examples/text-classification/run_glue.py\", line 197, in main\r\n eval_result = trainer.evaluate(eval_dataset=eval_dataset)\r\n File \"/data/home/liusishun/.conda/envs/myenv/lib/python3.8/site-packages/transformers/trainer.py\", line 1297, in evaluate\r\n output = self.prediction_loop(eval_dataloader, description=\"Evaluation\")\r\n File \"/data/home/liusishun/.conda/envs/myenv/lib/python3.8/site-packages/transformers/trainer.py\", line 1377, in prediction_loop\r\n loss, logits, labels = self.prediction_step(model, inputs, prediction_loss_only)\r\n File \"/data/home/liusishun/.conda/envs/myenv/lib/python3.8/site-packages/transformers/trainer.py\", line 1459, in prediction_step\r\n outputs = model(**inputs)\r\n File \"/data/home/liusishun/.conda/envs/myenv/lib/python3.8/site-packages/torch/nn/modules/module.py\", line 532, in __call__\r\n result = self.forward(*input, **kwargs)\r\n File \"/data/home/liusishun/.conda/envs/myenv/lib/python3.8/site-packages/transformers/modeling_xlnet.py\", line 1499, in forward\r\n transformer_outputs = self.transformer(\r\n File \"/data/home/liusishun/.conda/envs/myenv/lib/python3.8/site-packages/torch/nn/modules/module.py\", line 532, in __call__\r\n result = self.forward(*input, **kwargs)\r\n File \"/data/home/liusishun/.conda/envs/myenv/lib/python3.8/site-packages/transformers/modeling_xlnet.py\", line 1226, in forward\r\n new_mems = new_mems + (self.cache_mem(output_h, mems[i]),)\r\n File \"/data/home/liusishun/.conda/envs/myenv/lib/python3.8/site-packages/transformers/modeling_xlnet.py\", line 1011, in cache_mem\r\n new_mem = torch.cat([prev_mem, curr_out], dim=0)[cutoff:]\r\nRuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 0. Got 40 and 64 in dimension 1 at /opt/conda/conda-bld/pytorch_1579061855666/work/aten/src/THC/generic/THCTensorMath.cu:71\r\n```\r\n\r\ncurrent script\r\n```bash\r\nGLUE_DIR=~/glue\r\nCUDA_VISIBLE_DEVICES=0\r\nTASK_NAME=SST-2\r\n\r\npython3 ~/applications/transformers/examples/text-classification/run_glue.py \\\r\n --model_name_or_path ~/xlnet \\\r\n --task_name $TASK_NAME \\\r\n --do_eval \\\r\n --data_dir $GLUE_DIR/$TASK_NAME \\\r\n --max_seq_length 64 \\\r\n --per_device_train_batch_size 32 \\\r\n --per_device_eval_batch_size 64 \\\r\n --past_index 2 \\\r\n --learning_rate 2e-5 \\\r\n --num_train_epochs 3.0 \\\r\n --output_dir ~/result/$TASK_NAME/ \\\r\n --overwrite_output_dir \\\r\n --eval_steps 100 \\\r\n```\r\nAny idea?", "Asking for the XLNet specialists on our internal slack. I think the main problem is that the model returns those mems that can't be used for anything (and can't be concatenated). The fact you have an error with `past_index` show they can't really be used to speed up sequence classification.", "Thanks for your response. Could you have any temporary workarounds or further actions about this problem?", "Use another model...", "Hi @StepinSilence and @sgugger ! Any updates on this issue?\r\n@StepinSilence were able to find a work around to use XLNet?", "Hi, @adhithyaarun. I remember that this issue occurred when batch size couldn't divide the dataset size, so if you set the batch size a factor of the size of your dataset it may work. However, I can't confirm this right now because our server data disk died several days ago.", "Hello. I encountered the same problem using a Camembert Model with transformers 3.4.0. This issue seems to rise when using dynamic padding. Any workaround for this other than padding to max length?", "You should update to 3.5.0, which contains a fix for this in `Trainer`, to be able to do evaluation with dynamic padding.", "From reading the paper (especilally the experiment part about SQuad, RACE, ...) I originally thought that the cached memory was also used during fine-tuning and not just during pre-training, but from this description here: https://github.com/zihangdai/xlnet/issues/41#issuecomment-505102587 it seems like the cached memory is actually not used during fine-tuning. So I'd suggest that we disable it for all models except `XLNetLMHeadModel` where it obviously makes sense to use it. I'll add a PR to fix it", "Really thank all of you for fixing 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 } ]

[ { "login": "patrickvonplaten", "id": 23423619, "node_id": "MDQ6VXNlcjIzNDIzNjE5", "avatar_url": "https://avatars.githubusercontent.com/u/23423619?v=4", "gravatar_id": "", "url": "https://api.github.com/users/patrickvonplaten", "html_url": "https://github.com/patrickvonplaten", "followers_url": "https://api.github.com/users/patrickvonplaten/followers", "following_url": "https://api.github.com/users/patrickvonplaten/following{/other_user}", "gists_url": "https://api.github.com/users/patrickvonplaten/gists{/gist_id}", "starred_url": "https://api.github.com/users/patrickvonplaten/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/patrickvonplaten/subscriptions", "organizations_url": "https://api.github.com/users/patrickvonplaten/orgs", "repos_url": "https://api.github.com/users/patrickvonplaten/repos", "events_url": "https://api.github.com/users/patrickvonplaten/events{/privacy}", "received_events_url": "https://api.github.com/users/patrickvonplaten/received_events", "type": "User", "site_admin": false }, { "login": "lhoestq", "id": 42851186, "node_id": "MDQ6VXNlcjQyODUxMTg2", "avatar_url": "https://avatars.githubusercontent.com/u/42851186?v=4", "gravatar_id": "", "url": "https://api.github.com/users/lhoestq", "html_url": "https://github.com/lhoestq", "followers_url": "https://api.github.com/users/lhoestq/followers", "following_url": "https://api.github.com/users/lhoestq/following{/other_user}", "gists_url": "https://api.github.com/users/lhoestq/gists{/gist_id}", "starred_url": "https://api.github.com/users/lhoestq/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/lhoestq/subscriptions", "organizations_url": "https://api.github.com/users/lhoestq/orgs", "repos_url": "https://api.github.com/users/lhoestq/repos", "events_url": "https://api.github.com/users/lhoestq/events{/privacy}", "received_events_url": "https://api.github.com/users/lhoestq/received_events", "type": "User", "site_admin": false } ]

[ "tagging @patrickvonplaten who will be more suitable to help", "Hey @josemlopez - thanks for the issue. @lhoestq - I think we should add an argument to the `RagRetriever.from_pretrained(...)` that passes the cache dir to the `load_dataset` function, no? What do you think? ", "Thanks for your work guys.\r\n\r\nBTW, in case this can be helpful. \r\nI've move my things so I can have enough room for the dataset in \"/root/.cache/huggingface/datasets/\".\r\n\r\nDoing that, I've suffered this error. I can't say if it is related or not:\r\n\r\n```\r\n---------------------------------------------------------------------------\r\nUnpicklingError Traceback (most recent call last)\r\n/opt/conda/lib/python3.7/site-packages/numpy/lib/npyio.py in load(file, mmap_mode, allow_pickle, fix_imports, encoding)\r\n 459 try:\r\n--> 460 return pickle.load(fid, **pickle_kwargs)\r\n 461 except Exception:\r\n\r\nUnpicklingError: pickle data was truncated\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nOSError Traceback (most recent call last)\r\n/opt/conda/lib/python3.7/site-packages/datasets/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)\r\n 552 # Prepare split will record examples associated to the split\r\n--> 553 self._prepare_split(split_generator, **prepare_split_kwargs)\r\n 554 except OSError:\r\n\r\n/opt/conda/lib/python3.7/site-packages/datasets/builder.py in _prepare_split(self, split_generator)\r\n 840 for key, record in utils.tqdm(\r\n--> 841 generator, unit=\" examples\", total=split_info.num_examples, leave=False, disable=not_verbose\r\n 842 ):\r\n\r\n/opt/conda/lib/python3.7/site-packages/tqdm/notebook.py in __iter__(self, *args, **kwargs)\r\n 217 try:\r\n--> 218 for obj in super(tqdm_notebook, self).__iter__(*args, **kwargs):\r\n 219 # return super(tqdm...) will not catch exception\r\n\r\n/opt/conda/lib/python3.7/site-packages/tqdm/std.py in __iter__(self)\r\n 1128 try:\r\n-> 1129 for obj in iterable:\r\n 1130 yield obj\r\n\r\n~/.cache/huggingface/modules/datasets_modules/datasets/wiki_dpr/14b973bf2a456087ff69c0fd34526684eed22e48e0dfce4338f9a22b965ce7c2/wiki_dpr.py in _generate_examples(self, data_file, vectors_files)\r\n 131 break\r\n--> 132 vecs = np.load(open(vectors_files.pop(0), \"rb\"), allow_pickle=True)\r\n 133 vec_idx = 0\r\n\r\n/opt/conda/lib/python3.7/site-packages/numpy/lib/npyio.py in load(file, mmap_mode, allow_pickle, fix_imports, encoding)\r\n 462 raise IOError(\r\n--> 463 \"Failed to interpret file %s as a pickle\" % repr(file))\r\n 464 finally:\r\n\r\nOSError: Failed to interpret file <_io.BufferedReader name='/root/.cache/huggingface/datasets/downloads/f34d5f091294259b4ca90e813631e69a6ded660d71b6cbedf89ddba50df94448'> as a pickle\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nOSError Traceback (most recent call last)\r\n<ipython-input-6-f28df370ac47> in <module>\r\n 1 # ln -s /workspace/notebooks/POCs/cache /root/.cache/huggingface/datasets\r\n----> 2 retriever = RagRetriever.from_pretrained(\"facebook/rag-token-nq\", index_name=\"exact\", use_dummy_dataset=False)\r\n\r\n/opt/conda/lib/python3.7/site-packages/transformers/retrieval_rag.py in from_pretrained(cls, retriever_name_or_path, **kwargs)\r\n 307 generator_tokenizer = rag_tokenizer.generator\r\n 308 return cls(\r\n--> 309 config, question_encoder_tokenizer=question_encoder_tokenizer, generator_tokenizer=generator_tokenizer\r\n 310 )\r\n 311 \r\n\r\n/opt/conda/lib/python3.7/site-packages/transformers/retrieval_rag.py in __init__(self, config, question_encoder_tokenizer, generator_tokenizer)\r\n 298 self.config = config\r\n 299 if self._init_retrieval:\r\n--> 300 self.init_retrieval()\r\n 301 \r\n 302 @classmethod\r\n\r\n/opt/conda/lib/python3.7/site-packages/transformers/retrieval_rag.py in init_retrieval(self)\r\n 324 \r\n 325 logger.info(\"initializing retrieval\")\r\n--> 326 self.index.init_index()\r\n 327 \r\n 328 def postprocess_docs(self, docs, input_strings, prefix, n_docs, return_tensors=None):\r\n\r\n/opt/conda/lib/python3.7/site-packages/transformers/retrieval_rag.py in init_index(self)\r\n 238 split=self.dataset_split,\r\n 239 index_name=self.index_name,\r\n--> 240 dummy=self.use_dummy_dataset,\r\n 241 )\r\n 242 self.dataset.set_format(\"numpy\", columns=[\"embeddings\"], output_all_columns=True)\r\n\r\n/opt/conda/lib/python3.7/site-packages/datasets/load.py in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, save_infos, script_version, **config_kwargs)\r\n 609 download_config=download_config,\r\n 610 download_mode=download_mode,\r\n--> 611 ignore_verifications=ignore_verifications,\r\n 612 )\r\n 613 \r\n\r\n/opt/conda/lib/python3.7/site-packages/datasets/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)\r\n 474 if not downloaded_from_gcs:\r\n 475 self._download_and_prepare(\r\n--> 476 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n 477 )\r\n 478 # Sync info\r\n\r\n/opt/conda/lib/python3.7/site-packages/datasets/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)\r\n 553 self._prepare_split(split_generator, **prepare_split_kwargs)\r\n 554 except OSError:\r\n--> 555 raise OSError(\"Cannot find data file. \" + (self.manual_download_instructions or \"\"))\r\n 556 \r\n 557 if verify_infos:\r\n\r\nOSError: Cannot find data file. \r\n```\r\n\r\nWhen running this: \r\n\r\n`retriever = RagRetriever.from_pretrained(\"facebook/rag-token-nq\", index_name=\"exact\", use_dummy_dataset=False)`", "> Hey @josemlopez - thanks for the issue. @lhoestq - I think we should add an argument to the `RagRetriever.from_pretrained(...)` that passes the cache dir to the `load_dataset` function, no? What do you think?\r\n\r\nSure we can add `cache_dir=...` to `RagRetriever.from_pretrained`.\r\nIn the meantime you can specify `HF_DATASETS_CACHE` to tell where to store the dataset used by RAG for retrieval\r\n\r\n> Thanks for your work guys.\r\n> \r\n> BTW, in case this can be helpful.\r\n> I've move my things so I can have enough room for the dataset in \"/root/.cache/huggingface/datasets/\".\r\n> \r\n> Doing that, I've suffered this error. I can't say if it is related or not:\r\n> ...\r\n> When running this:\r\n> \r\n> `retriever = RagRetriever.from_pretrained(\"facebook/rag-token-nq\", index_name=\"exact\", use_dummy_dataset=False)`\r\n\r\nCould you create an issue on the `datasets` repo ? this seems unrelated ", "Hi @lhoestq , \r\n\r\n>In the meantime you can specify HF_DATASETS_CACHE to tell where to store the dataset used by RAG for retrieval\r\n\r\nHF_DATASETS_CACHE works fine: \r\n\r\n```\r\nretriever = RagRetriever.from_pretrained(\"facebook/rag-token-nq\", index_name=\"exact\", use_dummy_dataset=False)\r\n\r\nUsing custom data configuration psgs_w100.nq.no_index\r\nReusing dataset wiki_dpr (/my_cache/cache/wiki_dpr/psgs_w100.nq.no_index/0.0.0/14b973bf2a456087ff69c0fd34526684eed22e48e0dfce4338f9a22b965ce7c2)\r\nUsing custom data configuration psgs_w100.nq.exact\r\n\r\nDownloading and preparing dataset wiki_dpr/psgs_w100.nq.exact (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /my_cache/cache/wiki_dpr/psgs_w100.nq.exact/0.0.0/14b973bf2a456087ff69c0fd34526684eed22e48e0dfce4338f9a22b965ce7c2...\r\n```\r\n\r\n>Could you create an issue on the datasets repo ? this seems unrelated\r\n\r\nsure, I'll post the other issue in the datasets repo.\r\n\r\nThanks!", "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 see that tests are failing, but shouldn't `min_length` and `top_k` simply not be allowed to go to zero?", "`min_length` defaults to 0 which is expteced behavior. `top_k` is 0 if it is not used => so I don't think we should do these changes.", "We can fix the typo though ;-) ", "My bad, that makes sense 😄 " ]

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

[]

[ "Add model card for https://huggingface.co/akhooli/xlm-r-large-arabic-toxic" ]

[]

[]

[]

[]

[]

[ "yes, the fix is trivial. Which tool is complaining?", "I was running `make modified_only_fixup` before merging master and `black` was complaining.\r\nBut after merging master, there is no complaining.\r\nAnd I should be using `make fixup`, so this might just be a UserError.\r\n\r\nShould we merge the linked PR anyways or wait to see if I run into this again?\r\n", "Indeed, `black` isn't smooth - it picks .py files when you give it a dir, but doesn't do the same if you give it explicit files:\r\n```\r\n$ black Makefile\r\nerror: cannot format Makefile: Cannot parse: 1:1: .PHONY: modified_only_fixup extra_quality_checks quality style fixup fix-copies test test-examples docs\r\nOh no! 💥 💔 💥\r\n1 file failed to reformat.\r\n```\r\nSo yes, please merge the linked PR.", "`make fixup` is just `make modified_only_fixup` + `make extra_quality_checks` so no user error" ]

[ { "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" ]

[]

[]

[ "@patrickvonplaten I have made the required changes. Please review", "@patrickvonplaten I have made the required changes. Please review and merge" ]

[]

[ { "login": "sgugger", "id": 35901082, "node_id": "MDQ6VXNlcjM1OTAxMDgy", "avatar_url": "https://avatars.githubusercontent.com/u/35901082?v=4", "gravatar_id": "", "url": "https://api.github.com/users/sgugger", "html_url": "https://github.com/sgugger", "followers_url": "https://api.github.com/users/sgugger/followers", "following_url": "https://api.github.com/users/sgugger/following{/other_user}", "gists_url": "https://api.github.com/users/sgugger/gists{/gist_id}", "starred_url": "https://api.github.com/users/sgugger/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/sgugger/subscriptions", "organizations_url": "https://api.github.com/users/sgugger/orgs", "repos_url": "https://api.github.com/users/sgugger/repos", "events_url": "https://api.github.com/users/sgugger/events{/privacy}", "received_events_url": "https://api.github.com/users/sgugger/received_events", "type": "User", "site_admin": false } ]

[ "You did not provide any metrics to your `Trainer` and it looks like your dataset has no labels. `Trainer.evaluate` thus can't return anything useful." ]

[]

[]

[]

[]

[]

[ "Hello! I recommend you read [this doc](https://huggingface.co/transformers/task_summary.html) first to get an understanding of different tasks.\r\n\r\nWhat the warning you got means:\r\n\r\n- The model checkpoint (`\"gpt2\"`) was trained on a specific task (here, causal language modelling, or CLM)\r\n- You're loading that checkpoint in an architecture that has an additional head on top of it. This means there are a few more layers on top of the existing model.\r\n- The warning tells you: The base model (the GPT-2 architecture) is correctly initialized from the checkpoint. The additional head **is not**.\r\n- It cannot be initialized from that checkpoint as the multiple-choice head requires to be trained on a multiple-choice task. The CLM task mentioned earlier doesn't require this head and doesn't train it.\r\n- If you want to leverage that checkpoint with multiple-choice, it means that you should train these few layers on a multiple-choice task. Similarly to sequence classification or token classification, there are multiple different multiple-choice tasks, so you should find/create a dataset close to your use-case. You can find an example script showcasing that [here](https://github.com/huggingface/transformers/tree/master/examples/multiple-choice).\r\n\r\nI hope I helped answer your queries. Feel free to re-open if you have additional questions." ]

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

[]

[]

[ { "id": 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, if you want to implement a custom loss, you should not pass `labels` to the model, but instead retrieve the `hidden_states` and compute them as you would with any other PyTorch model.", "Thanks for your fast answer. I found out that nn.CrossEntropyLoss expects class indices and does not take one-hot encoded tensors as target labels. So there is no difference between your implementation and using one-hot encoded labels." ]

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

[]

[ "Hi, please fill-in the template or we won't be able to help you.", "Hello, I have updated it. Thanks", "This part is the most important part, please complete it:\r\n\r\n\r\n> ```\r\n> transformers version:\r\n> Platform:\r\n> Python version:\r\n> PyTorch version (GPU?):\r\n> Tensorflow version (GPU?):\r\n> Using GPU in script?:\r\n> Using distributed or parallel set-up in script?:\r\n> ```", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[]

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

[]

[]

[]

[]

[ "Distillation is not supported (yet) on TPU. You can check the status of different example scripts [here](https://github.com/huggingface/transformers/tree/master/examples). Those supported by `Trainer` or `TFTrainer` or `pytorch-lightning` can be run on TPU, others cannot." ]

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

[]

[ "I am unsure on why you think this test is wrong: if this test is true, we import APEX. So if we check with `<=`, we will then try to import APEX which is exactly what we are trying to avoid.", "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": 1834088753, "node_id": "MDU6TGFiZWwxODM0MDg4NzUz", "url": "https://api.github.com/repos/huggingface/transformers/labels/Tests", "name": "Tests", "color": "a6fcca", "default": false, "description": "Related to tests" } ]

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

[]

[]

[]

[ "Pinging @joeddav ", "Hey @sachinruk, thanks for taking the time to contribute 🤗\r\n\r\nI take your point and the case that you've described can certainly happen with the current method, but I maintain what I said in the other thread that there's not really a single correct way of doing this. We're highjacking the outputs of a model trained on a completely different distribution (NLI data) for own own purposes, so it's just a game of figuring out what makes intuitive sense and what empirically works the best.\r\n\r\nI ran a quick benchmark on the AG's News topic classification dataset comparing the current method with the one you've proposed, and they performed similarly. I got a weighted F1 of ~70 with the entailment-only method used in the pipeline and around 68 with the method you've proposed.\r\n\r\nIf you can show that your proposed method empirically does significantly better, we could look into changing it or adding it as an additional kwarg to specify the method. But as is, I don't think it makes sense to change the behavior of the pipeline under people's feet." ]

[]

[]

[ "Hello! How did you train your GPT-2 from scratch? Which framework did you use?", "> Hello! How did you train your GPT-2 from scratch? Which framework did you use?\r\n\r\nUsed PyTorch framework.", "Used this script\r\n```\r\nimport os\r\nimport gc\r\nimport glob\r\nimport torch\r\nimport pickle\r\nimport joblib\r\nfrom tqdm.auto import tqdm\r\nfrom pathlib import Path\r\nfrom tokenizers import ByteLevelBPETokenizer\r\nfrom transformers import GPT2Tokenizer\r\nimport torch\r\nfrom transformers import GPT2TokenizerFast\r\nfrom transformers import GPT2LMHeadModel\r\nfrom transformers import DataCollatorForLanguageModeling\r\nfrom transformers import TextDataset\r\nfrom transformers import GPT2Config\r\nfrom transformers import Trainer, TrainingArguments\r\n\r\n\r\ntokenizer = GPT2Tokenizer.from_pretrained('hindi/')\r\nvocab_size = tokenizer.vocab_size \r\nprint(vocab_size)\r\nprint(torch.cuda.is_available())\r\n\r\n\r\nconfig = GPT2Config(\r\n vocab_size=vocab_size \r\n)\r\n\r\ntokenizer = GPT2TokenizerFast.from_pretrained(\"hindi/\", max_len=512)\r\nmodel = GPT2LMHeadModel(config=config)\r\nprint(model.num_parameters())\r\n\r\nprint(\"Now let's build our training Dataset\")\r\n\r\ndataset = TextDataset(\r\n tokenizer=tokenizer,\r\n file_path=\"data/train.txt\",\r\n block_size=132,\r\n)\r\n\r\nprint(\"Start Training\")\r\n\r\ndata_collator = DataCollatorForLanguageModeling(\r\n tokenizer=tokenizer, mlm=True, mlm_probability=0.15\r\n)\r\n\r\nprint(\"Trainer Classes\")\r\n\r\ntraining_args = TrainingArguments(\r\n output_dir=\"hindi/\",\r\n overwrite_output_dir=True,\r\n num_train_epochs=1,\r\n per_gpu_train_batch_size=64,\r\n save_steps=10_000,\r\n save_total_limit=2,\r\n)\r\n\r\ntrainer = Trainer(\r\n model=model,\r\n args=training_args,\r\n data_collator=data_collator,\r\n train_dataset=dataset,\r\n prediction_loss_only=True,\r\n)\r\n\r\ntrainer.train()\r\n\r\ntrainer.save_model(\"hindi/\")\r\nprint(\"done\")\r\n```\r\nfor training @LysandreJik " ]

[]

[]

[ "We are actually having a longer internal discussion about the general handling of different attentions - this might still take a couple of days to be decided.", "> Cool, wonderful that you added so many tests!\r\n> \r\n> @patrickvonplaten, you say:\r\n> \r\n> > if global attention was used the attentions were set to the global attentions and the local attentions were discarded\r\n> \r\n> In this case, wouldn't the best model output be one where there are both `local_attentions`, `global_attentions` **as well as** `attentions` that are kept simply for backwards compatibility?\r\n\r\nThe previous design led to errors as shown here: https://github.com/huggingface/transformers/issues/5646 -> so I think it's fine to break backwards compatibility here. `local_attentions` would arguably be a better name than `attentions`, but for consistency with other models and for a standard case where `global_attention_mask=None`, so that `local_attentions` == (all) `attentions`, I would prefer to keep the name `attentions` here. \r\n\r\n> \r\n> Other than that and the docstrings, LGTM!\r\n\r\nDocstrings will be corrected!\r\n\r\n", "Works for me!", "@gui11aume great work again - understanding longformer's attention is not straightforward and your doc string was spot-on! Looking forward to your next contribution ;-) Hope you're fine with the small changes I made", "@patrickvonplaten @lalitpagaria this merge PR breaks longformer training with gradient checkpointing True\r\nplease fix as i am unable to train models with latest models. \r\n\r\n\r\nerror comes on this line \r\nhttps://github.com/huggingface/transformers/blob/master/src/transformers/modeling_longformer.py#L1072\r\n\r\nmodel expect 2 to 6 positional arguments, but 7 where giving. ", "> @patrickvonplaten @lalitpagaria this merge PR breaks longformer training with gradient checkpointing True\r\n> please fix as i am unable to train models with latest models.\r\n> \r\n> error comes on this line\r\n> https://github.com/huggingface/transformers/blob/master/src/transformers/modeling_longformer.py#L1072\r\n> \r\n> model expect 2 to 6 positional arguments, but 7 where giving.\r\n\r\nHey @manishiitg,\r\n\r\nThanks a lot for message! This did indeed break gradient checkpointing for longformer - sorry! This PR fixes it: https://github.com/huggingface/transformers/pull/8415" ]

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

[]

[ "Thanks for sharing. If you'd like you can contribute sample inputs for Sinhala at https://github.com/huggingface/widgets-server/blob/master/DefaultWidget.ts – Thanks!" ]

[]

[]

[ "Hello!\r\n\r\nThanks for you PR. The reason we used `tf.gradients` is because it handles `None` gradients while `tape.gradients` don't, so for now unless we can find a better way to handle the `None` values for gradients, we will keep it like this.\r\n\r\nAlso the training is forced to be done in graph compilation with `tf.function` so the eager mode is anyway disactivated.", "Thanks for detailed answer.\r\nAnother reason I did the change is to deal with following error that I getting after upgrading to the laters version of transformers library.\r\n`ValueError: distributed_training_steps() should not modify its Python input arguments. Check if it modifies any lists or dicts passed as arguments. Modifying a copy is allowed.`\r\n\r\nThis error disappears after my change.\r\ntensorflow==2.3.1\r\ntransformers==3.3.1", "Ok, can you open an issue with the details on how to reproduce the error please.", "I am closing this pull request as I found workaround regarding my issue issue with **parameter modification** .\r\n" ]

[]

[]

[ "It's hard to know what's going on without seeing your code. The error indicates `args.eval_steps = 0` and this argument is not modified by the `Trainer` itself, so you should make sure you did set it to something >0.", "You are right. I had `eval_steps = 0`. Thanks for the feedback!" ]