Datasets:

---

DanielPFlorian

/

Transformers-Github-Issues

like 0

[]

[]

[]

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

[]

[ "This can be worked around, for anyone hitting this issue, by setting `sentencepiece==0.1.91` explicitly.", "Maybe we could set `sentencepiece==0.1.91` in the setup.py to prevent this from happening, as we already had the issue with the 0.1.92.\r\n\r\nDo you want to open a PR for that?", "This should also be fixed by #8073 ", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "Had the same issue. Found a trim_batch implementation on the transformers repo, used it, didn't face any issues so far.\r\n\r\n[Link to implementation](https://github.com/huggingface/transformers/blob/783d7d2629e97c5f0c5f9ef01b8c66410275c204/examples/research_projects/rag/utils_rag.py#L35)\r\n\r\nCode for reference:\r\n\r\n```python\r\ndef trim_batch(\r\n input_ids,\r\n pad_token_id,\r\n attention_mask=None,\r\n):\r\n \"\"\"Remove columns that are populated exclusively by pad_token_id\"\"\"\r\n keep_column_mask = input_ids.ne(pad_token_id).any(dim=0)\r\n if attention_mask is None:\r\n return input_ids[:, keep_column_mask]\r\n else:\r\n return (input_ids[:, keep_column_mask], attention_mask[:, keep_column_mask])\r\n```" ]

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

[ { "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 } ]

[ "You're correct! @patrickvonplaten, git blame shows you as the author, want to fix?", "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" } ]

[]

[]

[]

[]

[]

[ { "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" } ]

[]

[]

[]

[]

[ "One probably related issue is that if I try to use `model.evaluate` I get an error with the following traceback:\r\n```\r\n\r\n---------------------------------------------------------------------------\r\nValueError Traceback (most recent call last)\r\n<ipython-input-125-3f03cbe29a62> in <module>\r\n----> 1 model.evaluate(test_dataset)\r\n\r\n~/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/keras/engine/training.py in _method_wrapper(self, *args, **kwargs)\r\n 106 def _method_wrapper(self, *args, **kwargs):\r\n 107 if not self._in_multi_worker_mode(): # pylint: disable=protected-access\r\n--> 108 return method(self, *args, **kwargs)\r\n 109 \r\n 110 # Running inside `run_distribute_coordinator` already.\r\n\r\n~/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/keras/engine/training.py in evaluate(self, x, y, batch_size, verbose, sample_weight, steps, callbacks, max_queue_size, workers, use_multiprocessing, return_dict)\r\n 1377 with trace.Trace('TraceContext', graph_type='test', step_num=step):\r\n 1378 callbacks.on_test_batch_begin(step)\r\n-> 1379 tmp_logs = test_function(iterator)\r\n 1380 if data_handler.should_sync:\r\n 1381 context.async_wait()\r\n\r\n~/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/eager/def_function.py in __call__(self, *args, **kwds)\r\n 778 else:\r\n 779 compiler = \"nonXla\"\r\n--> 780 result = self._call(*args, **kwds)\r\n 781 \r\n 782 new_tracing_count = self._get_tracing_count()\r\n\r\n~/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/eager/def_function.py in _call(self, *args, **kwds)\r\n 821 # This is the first call of __call__, so we have to initialize.\r\n 822 initializers = []\r\n--> 823 self._initialize(args, kwds, add_initializers_to=initializers)\r\n 824 finally:\r\n 825 # At this point we know that the initialization is complete (or less\r\n\r\n~/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/eager/def_function.py in _initialize(self, args, kwds, add_initializers_to)\r\n 695 self._concrete_stateful_fn = (\r\n 696 self._stateful_fn._get_concrete_function_internal_garbage_collected( # pylint: disable=protected-access\r\n--> 697 *args, **kwds))\r\n 698 \r\n 699 def invalid_creator_scope(*unused_args, **unused_kwds):\r\n\r\n~/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/eager/function.py in _get_concrete_function_internal_garbage_collected(self, *args, **kwargs)\r\n 2853 args, kwargs = None, None\r\n 2854 with self._lock:\r\n-> 2855 graph_function, _, _ = self._maybe_define_function(args, kwargs)\r\n 2856 return graph_function\r\n 2857 \r\n\r\n~/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/eager/function.py in _maybe_define_function(self, args, kwargs)\r\n 3211 \r\n 3212 self._function_cache.missed.add(call_context_key)\r\n-> 3213 graph_function = self._create_graph_function(args, kwargs)\r\n 3214 self._function_cache.primary[cache_key] = graph_function\r\n 3215 return graph_function, args, kwargs\r\n\r\n~/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/eager/function.py in _create_graph_function(self, args, kwargs, override_flat_arg_shapes)\r\n 3073 arg_names=arg_names,\r\n 3074 override_flat_arg_shapes=override_flat_arg_shapes,\r\n-> 3075 capture_by_value=self._capture_by_value),\r\n 3076 self._function_attributes,\r\n 3077 function_spec=self.function_spec,\r\n\r\n~/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/framework/func_graph.py in func_graph_from_py_func(name, python_func, args, kwargs, signature, func_graph, autograph, autograph_options, add_control_dependencies, arg_names, op_return_value, collections, capture_by_value, override_flat_arg_shapes)\r\n 984 _, original_func = tf_decorator.unwrap(python_func)\r\n 985 \r\n--> 986 func_outputs = python_func(*func_args, **func_kwargs)\r\n 987 \r\n 988 # invariant: `func_outputs` contains only Tensors, CompositeTensors,\r\n\r\n~/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/eager/def_function.py in wrapped_fn(*args, **kwds)\r\n 598 # __wrapped__ allows AutoGraph to swap in a converted function. We give\r\n 599 # the function a weak reference to itself to avoid a reference cycle.\r\n--> 600 return weak_wrapped_fn().__wrapped__(*args, **kwds)\r\n 601 weak_wrapped_fn = weakref.ref(wrapped_fn)\r\n 602 \r\n\r\n~/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/framework/func_graph.py in wrapper(*args, **kwargs)\r\n 971 except Exception as e: # pylint:disable=broad-except\r\n 972 if hasattr(e, \"ag_error_metadata\"):\r\n--> 973 raise e.ag_error_metadata.to_exception(e)\r\n 974 else:\r\n 975 raise\r\n\r\nValueError: in user code:\r\n\r\n /Users/rr48396/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/keras/engine/training.py:1224 test_function *\r\n return step_function(self, iterator)\r\n /Users/rr48396/opt/anaconda3/lib/python3.7/site-packages/transformers/modeling_tf_utils.py:142 compute_loss *\r\n return loss_fn(labels, logits)\r\n /Users/rr48396/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/keras/losses.py:149 __call__ **\r\n losses = ag_call(y_true, y_pred)\r\n /Users/rr48396/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/keras/losses.py:253 call **\r\n return ag_fn(y_true, y_pred, **self._fn_kwargs)\r\n /Users/rr48396/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/util/dispatch.py:201 wrapper\r\n return target(*args, **kwargs)\r\n /Users/rr48396/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/keras/losses.py:1567 sparse_categorical_crossentropy\r\n y_true, y_pred, from_logits=from_logits, axis=axis)\r\n /Users/rr48396/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/util/dispatch.py:201 wrapper\r\n return target(*args, **kwargs)\r\n /Users/rr48396/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/keras/backend.py:4783 sparse_categorical_crossentropy\r\n labels=target, logits=output)\r\n /Users/rr48396/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/util/dispatch.py:201 wrapper\r\n return target(*args, **kwargs)\r\n /Users/rr48396/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/nn_ops.py:4175 sparse_softmax_cross_entropy_with_logits_v2\r\n labels=labels, logits=logits, name=name)\r\n /Users/rr48396/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/util/dispatch.py:201 wrapper\r\n return target(*args, **kwargs)\r\n /Users/rr48396/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/nn_ops.py:4090 sparse_softmax_cross_entropy_with_logits\r\n logits.get_shape()))\r\n\r\n ValueError: Shape mismatch: The shape of labels (received (1,)) should equal the shape of logits except for the last dimension (received (512, 100)).\r\n```\r\n\r\nShouldn't be an error related to my dataset, as it's constructed the same way as in the tutorial...", "never mind, I didn't realize the model requires manually batching the dataset at prediction. closing :) " ]

[]

[]

[]

[ { "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" } ]

[]

[]

[]

[]

[ "### Solution\r\n\r\nIf you encounter the same, here is how to fix it:\r\n\r\n```\r\npip uninstall -y tensorflow-gpu tensorflow\r\npip install tensorflow-gpu -U \r\n```\r\n(I assume you want the gpu version - adjust if not)\r\n", "I think that for the last few versions, when installing `tensorflow` you get a `tensorflow` that can use your GPU out of the box, so there's no need to play with `tensorflow-gpu`/`tensorflow`!", "I think it's another package's dependency pulling in `tensorflow-gpu` - e.g. I see: `wandb/requirements.txt:tensorflow-gpu==2.3.1`", "In my case, after `conda install tensorflow-gpu` - to install `tensorflow` version 2.2,\r\nI then tried `pip install autokeras` (because conda does not have this package). \r\n`pip` would install **_another_** `tensorflow` (in this case, version 2.3).\r\n\r\nAnd this is when the problem happened.\r\n\r\n```\r\n>>> import tensorflow as tf\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"/home/longnv/.conda/envs/testENV/lib/python3.6/site-packages/tensorflow/__init__.py\", line 41, in <module>\r\n from tensorflow.python.tools import module_util as _module_util\r\n File \"/home/longnv/.conda/envs/testENV/lib/python3.6/site-packages/tensorflow/python/__init__.py\", line 84, in <module>\r\n from tensorflow.python import keras\r\n File \"/home/longnv/.conda/envs/testENV/lib/python3.6/site-packages/tensorflow/python/keras/__init__.py\", line 27, in <module>\r\n from tensorflow.python.keras import models\r\n File \"/home/longnv/.conda/envs/testENV/lib/python3.6/site-packages/tensorflow/python/keras/models.py\", line 24, in <module>\r\n from tensorflow.python.keras import metrics as metrics_module\r\n File \"/home/longnv/.conda/envs/testENV/lib/python3.6/site-packages/tensorflow/python/keras/metrics.py\", line 37, in <module>\r\n from tensorflow.python.keras.engine import base_layer\r\n File \"/home/longnv/.conda/envs/testENV/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py\", line 51, in <module>\r\n from tensorflow.python.keras import initializers\r\n File \"/home/longnv/.conda/envs/testENV/lib/python3.6/site-packages/tensorflow/python/keras/initializers/__init__.py\", line 127, in <module>\r\n populate_deserializable_objects()\r\n File \"/home/longnv/.conda/envs/testENV/lib/python3.6/site-packages/tensorflow/python/keras/initializers/__init__.py\", line 85, in populate_deserializable_objects\r\n generic_utils.populate_dict_with_module_objects(\r\nAttributeError: module 'tensorflow.python.keras.utils.generic_utils' has no attribute 'populate_dict_with_module_objects'\r\n>>> conda uninstall keras\r\n\r\n```\r\nIt seems like we have similar error.\r\nI am not sure what you have done with your env but your solution worked in my case too, obviously.", "Please add the snippet of the code into the program of general_utils.py\r\n\r\n**1. Find out your path**\r\n\r\nFor me, the path is listed as follows:\r\n\r\n/home/user/miniconda3/lib/python3.7/site-packages/tensorflow/python/keras/utils/generic_utils.py\r\n\r\n**2. Paste the code into the program of generic_utils.py**\r\n```\r\ndef populate_dict_with_module_objects(target_dict, modules, obj_filter):\r\n for module in modules:\r\n for name in dir(module):\r\n obj = getattr(module, name)\r\n if obj_filter(obj):\r\n target_dict[name] = obj\r\n```\r\n\r\n**3. Initialize the dev tool**\r\nIt will be working after initializing the dev tool (it is Terminal for me in Linux)" ]

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

[]

[ "is self.generate?", "Hello! Could you provide all the information relative to your environment as asked in the template, as well as the code that generates the error? Thanks!", "Thanks for your replying!\r\n\r\noutputs = self(**model_inputs)\r\nThis code appear in function _generate_beam_search and _generate_no_beam_search.\r\nThe 'self' is instance of class, it is not callable, so it can be used like this? There is not a function name like self.generate?\r\n\r\nThus, i just test the genaration_tf_util.py, so i add the test code like this. And i delete unrelated code just for test this before 'while cur_len < max_length: model_inputs = self.prepare_inputs_for_generation(\r\n input_ids, past=past, attention_mask=attention_mask, use_cache=use_cache\r\n )':\r\n\r\nMy environment is:\r\nwindows\r\ntensorflow 2.0.0rc0\r\nnumpy 1.17.2\r\n\r\n\r\n```\r\n# my add\r\n# if __name__ == '__main__':\r\n# a = TFGenerationMixin()\r\n# a._generate_beam_search(input_ids = None,\r\n# cur_len = 10,\r\n# max_length = 100,\r\n# min_length = 5,\r\n# do_sample = None,\r\n# early_stopping = None,\r\n# # num_beams = None,\r\n# temperature = None,\r\n# top_k = None,\r\n# top_p = None,\r\n# repetition_penalty = None,\r\n# no_repeat_ngram_size=None,\r\n# bad_words_ids = None,\r\n# # bos_token_id = None,\r\n# pad_token_id = None,\r\n# eos_token_id = None,\r\n# batch_size=4,\r\n# num_return_sequences=None,\r\n# length_penalty = None,\r\n# num_beams=None,\r\n# vocab_size=None,\r\n# # no_repeat_ngram_size = None,\r\n# # num_return_sequences = None,\r\n# encoder_outputs=None,\r\n# attention_mask = None,\r\n# # decoder_start_token_id = None,\r\n# use_cache = None,\r\n# )\r\n\r\n```\r\n\r\n> Hello! Could you provide all the information relative to your environment as asked in the template, as well as the code that generates the error? Thanks!\r\n\r\n", "The `self` is the instance of the class, calling it as `self(...)` results in calling the `__call__` method.", "see https://www.geeksforgeeks.org/callable-in-python/", "> The `self` is the instance of the class, calling it as `self(...)` results in calling the `__call__` method.\r\n\r\nBut where is __call__ method of class TFGenerationMixin?", "Ah, my bad, I hadn't taken a close enough look at your code. You can't initialize a `TFGenerationMixin` like this, as it is an abstract class. It is there to have TF classes inherit from that abstract class, not to be used as-is.\r\n\r\nCould you tell me what you're trying to do so I could guide you towards the classes you should use?", "TFPreTrainedModel inherit from TFGenerationMixin, and TFPreTrainedModel is base class for all TF models. Some TF models have __call__ method, leading to self(..) in TFGenerationMixin can be callable, that's right?\r\n\r\nThis question appear just when I read the code, not for do something. \r\n\r\n", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[]

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

[]

[ "Hi! Could you put the whole stack trace here? I fear it might be an internal sentencepiece error, for which we'll be unable to help and you would have more luck opening an issue on the sentencepiece repo directly.", "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" ]

[]

[]

[ "`tests/test_tokenization_fsmt.py::FSMTTokenizationTest::test_match_encode_decode` fixed in https://github.com/huggingface/transformers/pull/8031\r\n\r\nI checked it off in your list.", "@patrickvonplaten can you take a look at the failing TF-Longformer and TF-T5 tests?", "> @patrickvonplaten can you take a look at the failing TF-Longformer and TF-T5 tests?\r\n\r\nWill take a look at the TF-Longformer Test :-) I tried a bit unsuccessfully to debug the `TF-T5` test for 2h a week ago and I didn't manage to get rid of `cast_bool_....` for `TFT5` at the moment. I think the hacky `cast_bool_...` function is the reason for the T5 Test Failure. Not sure if it's worth it spending a lot of time here again. :-/ I could comment out the test for now? Or @jplu @LysandreJik do you have a nice TF insight to solve it? ", "For now I propose to comment them out. I will do a pass over it later in a couple of weeks. I planned to go through each TF model anyway.", "I commented both TF T5 tests out ATM, same for the longformer. I think the Longformer test can be fixed, but the T5 tests cannot, at least not without introducing breaking changes.\r\n\r\nI think that we would need to have an additional layer between the `TFT5Model` and the encoder/decoder models, same as we do with every other model (the xxxMainLayer), otherwise we won't be able to use the saved model. Even with the `cast_bool_....` they're not currently usable as saved models.\r\n\r\nIt's the same issue with BART, and why those two tests are commented in BART as well imo.", "All tests are passing now. The only error on the scheduled is because of something on the hub with one tiny models, which I'm fixing right now." ]

[]

[]

[]

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

[ { "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 } ]

[ "If you look at\r\n\r\nhttps://s3.amazonaws.com/models.huggingface.co/bert/facebook/bart-large/config.json\r\nand \r\nhttps://s3.amazonaws.com/models.huggingface.co/bert/facebook/bart-base/config.json\r\n(how to do this for any model: go to [model hub](https://s3.amazonaws.com/models.huggingface.co/) and click see raw config file)\r\n\r\nyou will see different `task_specific_params`. These are used for fine-tuning by default so bart-large \r\nis forced to generate at least 56 tokens.\r\n\r\nThere are many ways to fix. Easiest is to comment out this line https://github.com/huggingface/transformers/blob/master/examples/seq2seq/finetune.py#L66\r\n\r\nMore involved would be to make a local copy of the config and insert the generation parameters you want. You can pass it to finetune.py with `--config_name`.\r\n\r\n\r\nI will think about how to update bart-base and bart-large to have more reasonable task_specific_params.", "cc @patil-suraj @stas00 @patrickvonplaten for awareness of a very sneaky bug.", "@sshleifer , thank you very much for your reply. Indeed, I have checked those configurations. So I changed the parameters for the `generate` method to consider min_length=0:\r\n\r\n```\r\n generated_ids = self.model.generate(\r\n batch[\"input_ids\"],\r\n attention_mask=batch[\"attention_mask\"],\r\n use_cache=True,\r\n decoder_start_token_id=self.decoder_start_token_id,\r\n num_beams=self.eval_beams,\r\n no_repeat_ngram_size=0,\r\n min_length=0,\r\n max_length=self.eval_max_length,\r\n length_penalty=1.0\r\n )\r\n```\r\n\r\nI used this code for both `facebook/bart-base` and `facebook/bart-large`. And the outputs for `bart-large` are as I mentioned. I have been trying to figure out the reason in the last days without success. Maybe I'm doing some wrong, but I could not discover what it is yet.\r\n\r\nAnother point is that the generation for `bart-large` is much slower than `bart-base`. Maybe it is because the model is generating tokens until the limit (max_length).", "How did you call `generate` to produce the outputs in your Issue Description?\r\nYour change to finetune.py will not change the config.\r\n", "This is my `_generative_step` method: \r\n```\r\n def _generative_step(self, batch: dict) -> dict:\r\n t0 = time.time()\r\n\r\n generated_ids = self.model.generate(\r\n batch[\"input_ids\"],\r\n attention_mask=batch[\"attention_mask\"],\r\n use_cache=True,\r\n decoder_start_token_id=self.decoder_start_token_id,\r\n num_beams=self.eval_beams,\r\n no_repeat_ngram_size=0,\r\n min_length=0,\r\n max_length=self.eval_max_length,\r\n length_penalty=1.0\r\n )\r\n gen_time = (time.time() - t0) / batch[\"input_ids\"].shape[0]\r\n preds: List[str] = self.ids_to_clean_text(generated_ids)\r\n target: List[str] = self.ids_to_clean_text(batch[\"labels\"])\r\n\r\n a = self.tokenizer.batch_decode(batch[\"input_ids\"].tolist())\r\n b = self.tokenizer.batch_decode(batch[\"labels\"].tolist())\r\n c = self.tokenizer.batch_decode(generated_ids)\r\n pad_token_id = self.tokenizer.pad_token_id\r\n tgt_ids = batch[\"labels\"]\r\n if isinstance(self.model, T5ForConditionalGeneration):\r\n decoder_input_ids = self.model._shift_right(tgt_ids)\r\n else:\r\n decoder_input_ids = shift_tokens_right(tgt_ids, pad_token_id)\r\n e = self.tokenizer.batch_decode(decoder_input_ids.tolist())\r\n\r\n loss_tensors = self._step(batch)\r\n base_metrics = {name: loss for name, loss in zip(self.loss_names, loss_tensors)}\r\n rouge: Dict = self.calc_generative_metrics(preds, target)\r\n summ_len = np.mean(lmap(len, generated_ids))\r\n base_metrics.update(gen_time=gen_time, gen_len=summ_len, preds=preds, target=target, a=a, b=b, c=c, e=e, **rouge)\r\n return base_metrics\r\n```\r\n\r\n`_step` method:\r\n\r\n```\r\n def _step(self, batch: dict) -> Tuple:\r\n pad_token_id = self.tokenizer.pad_token_id\r\n src_ids, src_mask = batch[\"input_ids\"], batch[\"attention_mask\"]\r\n tgt_ids = batch[\"labels\"]\r\n if isinstance(self.model, T5ForConditionalGeneration):\r\n decoder_input_ids = self.model._shift_right(tgt_ids)\r\n else:\r\n decoder_input_ids = shift_tokens_right(tgt_ids, pad_token_id)\r\n if not self.already_saved_batch: # This would be slightly better if it only happened on rank zero\r\n batch[\"decoder_input_ids\"] = decoder_input_ids\r\n self.save_readable_batch(batch)\r\n\r\n outputs = self(src_ids, attention_mask=src_mask, decoder_input_ids=decoder_input_ids, use_cache=False)\r\n lm_logits = outputs[0]\r\n if self.hparams.label_smoothing == 0:\r\n # Same behavior as modeling_bart.py, besides ignoring pad_token_id\r\n ce_loss_fct = torch.nn.CrossEntropyLoss(ignore_index=pad_token_id)\r\n\r\n assert lm_logits.shape[-1] == self.vocab_size\r\n loss = ce_loss_fct(lm_logits.view(-1, lm_logits.shape[-1]), tgt_ids.view(-1))\r\n else:\r\n lprobs = torch.nn.functional.log_softmax(lm_logits, dim=-1)\r\n loss, nll_loss = label_smoothed_nll_loss(\r\n lprobs, tgt_ids, self.hparams.label_smoothing, ignore_index=pad_token_id\r\n )\r\n return (loss,)\r\n```\r\n\r\nThis is my validation_epoch_end:\r\n\r\n```\r\n def validation_epoch_end(self, outputs, prefix=\"val\") -> Dict:\r\n self.step_count += 1\r\n losses = {k: torch.stack([x[k] for x in outputs]).mean() for k in self.loss_names}\r\n loss = losses[\"loss\"]\r\n generative_metrics = {\r\n k: np.array([x[k] for x in outputs]).mean() for k in self.metric_names + [\"gen_time\", \"gen_len\"]\r\n }\r\n metric_val = (\r\n generative_metrics[self.val_metric] if self.val_metric in generative_metrics else losses[self.val_metric]\r\n )\r\n metric_tensor: torch.FloatTensor = torch.tensor(metric_val).type_as(loss)\r\n generative_metrics.update({k: v.item() for k, v in losses.items()})\r\n losses.update(generative_metrics)\r\n all_metrics = {f\"{prefix}_avg_{k}\": x for k, x in losses.items()}\r\n all_metrics[\"step_count\"] = self.step_count\r\n self.metrics[prefix].append(all_metrics) # callback writes this to self.metrics_save_path\r\n preds = flatten_list([x[\"preds\"] for x in outputs])\r\n\r\n val_outputs_folder = \"val_outputs\"\r\n os.system(\"mkdir -p \" + os.path.join(self.hparams.output_dir, val_outputs_folder))\r\n\r\n if \"preds\" in outputs[0]:\r\n tb_all = {}\r\n idx_tb = 0\r\n for output_batch in outputs:\r\n a,b,c,e = output_batch[\"a\"], output_batch[\"b\"], output_batch[\"c\"], output_batch[\"e\"]\r\n\r\n\r\n for aa,bb,ee,cc in zip(a,b,e,c):\r\n tb_all[idx_tb] = {}\r\n tb_all[idx_tb]['input_ids'] = aa\r\n tb_all[idx_tb]['labels'] = bb\r\n tb_all[idx_tb]['decoder_input_ids'] = ee\r\n tb_all[idx_tb]['generated_ids'] = cc\r\n idx_tb += 1\r\n\r\n file_debug = os.path.join(self.hparams.output_dir, val_outputs_folder,\r\n \"debug_\" +\r\n str(self.step_count) + \".json\")\r\n save_json(tb_all, file_debug)\r\n\r\n return {\r\n \"log\": all_metrics,\r\n \"preds\": preds,\r\n f\"{prefix}_loss\": loss,\r\n f\"{prefix}_{self.val_metric}\": metric_tensor,\r\n }\r\n```\r\n\r\n\r\nSo I use the `debug_k.json` file to check the outputs. Sorry for the variable names.\r\n\r\nOne example for `bart-base`:\r\n\r\n```\r\n \"1366\": {\r\n \"input_ids\": \"<s> ( report :ARG1 ( station :ARG1 ( troop :mod ( country :wiki Russia :name ( name :op1 Russia ) ) :ARG0-of ( withdraw :ARG2 ( country :quant 3 :location ( sea :wiki Baltic_Sea :name ( name :op1 Baltic :op2 Sea ) ) ) ) ) :ARG2 ( and :op1 ( state :wiki - :name ( name :op1 Jalininggele ) :location country ) :op2 ( state :wiki - :name ( name :op1 Simolingsike ) ) :op3 ( city :wiki - :name ( name :op1 Yelinia ) :location ( relative-position :op1 ( city :wiki Moscow :name ( name :op1 Moscow ) ) :quant ( distance-quantity :quant 300 :unit ( kilometer ) ) ) ) ) :mod ( respective ) ) )</s><pad><pad><pad>\",\r\n \"labels\": \"<s> It is reported that the Russian troops that withdrew from the three Baltic Sea countries will be stationed respectively in the Russian state of Jalininggele, the state of Simolingsike and Yelinia city which is 300 kilometers away from Moscow.</s>\",\r\n \"decoder_input_ids\": \"</s><s> It is reported that the Russian troops that withdrew from the three Baltic Sea countries will be stationed respectively in the Russian state of Jalininggele, the state of Simolingsike and Yelinia city which is 300 kilometers away from Moscow.\",\r\n \"generated_ids\": \"</s><s> Russian troops withdrawing from 3 Baltic Sea countries are reported to have respectively been stationed in the Baltic Sea states of Jalininggele,Simolingsike and Yelinia 300 kilometers away from Moscow.</s>\"\r\n },\r\n```\r\n\r\none example for `bart-large`:\r\n```\r\n \"1366\": {\r\n \"input_ids\": \"<s> ( report :ARG1 ( station :ARG1 ( troop :mod ( country :wiki Russia :name ( name :op1 Russia ) ) :ARG0-of ( withdraw :ARG2 ( country :quant 3 :location ( sea :wiki Baltic_Sea :name ( name :op1 Baltic :op2 Sea ) ) ) ) ) :ARG2 ( and :op1 ( state :wiki - :name ( name :op1 Jalininggele ) :location country ) :op2 ( state :wiki - :name ( name :op1 Simolingsike ) ) :op3 ( city :wiki - :name ( name :op1 Yelinia ) :location ( relative-position :op1 ( city :wiki Moscow :name ( name :op1 Moscow ) ) :quant ( distance-quantity :quant 300 :unit ( kilometer ) ) ) ) ) :mod ( respective ) ) )</s><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad>\",\r\n \"labels\": \"<s> It is reported that the Russian troops that withdrew from the three Baltic Sea countries will be stationed respectively in the Russian state of Jalininggele, the state of Simolingsike and Yelinia city which is 300 kilometers away from Moscow.</s>\",\r\n \"decoder_input_ids\": \"</s><s> It is reported that the Russian troops that withdrew from the three Baltic Sea countries will be stationed respectively in the Russian state of Jalininggele, the state of Simolingsike and Yelinia city which is 300 kilometers away from Moscow.\",\r\n \"generated_ids\": \"</s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s></s>\"\r\n },\r\n```\r\n\r\n\r\n\r\n", "@sshleifer I have changed the code (`3.3.1`) version in order to use the same processed decoder input for the model as the one used in transformer version `2.11.0` and it worked for both BARTs! Both BARTs (`facebook/bart-base` and `facebook/bart-large`) give good BLEU scores and generate good outputs!\r\n\r\nThe changed code:\r\n```\r\n def _step(self, batch: dict) -> Tuple:\r\n pad_token_id = self.tokenizer.pad_token_id\r\n src_ids, src_mask = batch[\"input_ids\"], batch[\"attention_mask\"]\r\n if isinstance(self.model, T5ForConditionalGeneration):\r\n tgt_ids = batch[\"labels\"]\r\n decoder_input_ids = self.model._shift_right(tgt_ids)\r\n else:\r\n #decoder_input_ids = shift_tokens_right(tgt_ids, pad_token_id)\r\n y = batch[\"labels\"]\r\n decoder_input_ids = y[:, :-1].contiguous()\r\n tgt_ids = y[:, 1:].clone()\r\n if not self.already_saved_batch: # This would be slightly better if it only happened on rank zero\r\n batch[\"decoder_input_ids\"] = decoder_input_ids\r\n self.save_readable_batch(batch)\r\n\r\n outputs = self(src_ids, attention_mask=src_mask, decoder_input_ids=decoder_input_ids, use_cache=False)\r\n lm_logits = outputs[0]\r\n if self.hparams.label_smoothing == 0:\r\n # Same behavior as modeling_bart.py, besides ignoring pad_token_id\r\n ce_loss_fct = torch.nn.CrossEntropyLoss(ignore_index=pad_token_id)\r\n\r\n assert lm_logits.shape[-1] == self.vocab_size\r\n loss = ce_loss_fct(lm_logits.view(-1, lm_logits.shape[-1]), tgt_ids.view(-1))\r\n else:\r\n lprobs = torch.nn.functional.log_softmax(lm_logits, dim=-1)\r\n loss, nll_loss = label_smoothed_nll_loss(\r\n lprobs, tgt_ids, self.hparams.label_smoothing, ignore_index=pad_token_id\r\n )\r\n return (loss,)\r\n```\r\n\r\nan example generated by `facebook/bart-base` using the new code:\r\n```\r\n \"1366\": {\r\n \"input_ids\": \"<s> ( report :ARG1 ( station :ARG1 ( troop :mod ( country :wiki Russia :name ( name :op1 Russia ) ) :ARG0-of ( withdraw :ARG2 ( country :quant 3 :location ( sea :wiki Baltic_Sea :name ( name :op1 Baltic :op2 Sea ) ) ) ) ) :ARG2 ( and :op1 ( state :wiki - :name ( name :op1 Jalininggele ) :location country ) :op2 ( state :wiki - :name ( name :op1 Simolingsike ) ) :op3 ( city :wiki - :name ( name :op1 Yelinia ) :location ( relative-position :op1 ( city :wiki Moscow :name ( name :op1 Moscow ) ) :quant ( distance-quantity :quant 300 :unit ( kilometer ) ) ) ) ) :mod ( respective ) ) )</s><pad><pad><pad>\",\r\n \"labels\": \" It is reported that the Russian troops that withdrew from the three Baltic Sea countries will be stationed respectively in the Russian state of Jalininggele, the state of Simolingsike and Yelinia city which is 300 kilometers away from Moscow.</s>\",\r\n \"decoder_input_ids\": \"<s> It is reported that the Russian troops that withdrew from the three Baltic Sea countries will be stationed respectively in the Russian state of Jalininggele, the state of Simolingsike and Yelinia city which is 300 kilometers away from Moscow.\",\r\n \"generated_ids\": \"</s> Russian troops withdrawing from 3 Baltic Sea countries have been reported to be stationed respectively in Jalininggele, Simolingsike and Yelinia 300 kilometers (200 miles) from Moscow.</s><pad><pad>\"\r\n },\r\n```\r\n\r\nan example generated by `facebook/bart-large` using the new code:\r\n```\r\n \"1366\": {\r\n \"input_ids\": \"<s> ( report :ARG1 ( station :ARG1 ( troop :mod ( country :wiki Russia :name ( name :op1 Russia ) ) :ARG0-of ( withdraw :ARG2 ( country :quant 3 :location ( sea :wiki Baltic_Sea :name ( name :op1 Baltic :op2 Sea ) ) ) ) ) :ARG2 ( and :op1 ( state :wiki - :name ( name :op1 Jalininggele ) :location country ) :op2 ( state :wiki - :name ( name :op1 Simolingsike ) ) :op3 ( city :wiki - :name ( name :op1 Yelinia ) :location ( relative-position :op1 ( city :wiki Moscow :name ( name :op1 Moscow ) ) :quant ( distance-quantity :quant 300 :unit ( kilometer ) ) ) ) ) :mod ( respective ) ) )</s><pad><pad><pad>\",\r\n \"labels\": \" It is reported that the Russian troops that withdrew from the three Baltic Sea countries will be stationed respectively in the Russian state of Jalininggele, the state of Simolingsike and Yelinia city which is 300 kilometers away from Moscow.</s>\",\r\n \"decoder_input_ids\": \"<s> It is reported that the Russian troops that withdrew from the three Baltic Sea countries will be stationed respectively in the Russian state of Jalininggele, the state of Simolingsike and Yelinia city which is 300 kilometers away from Moscow.\",\r\n \"generated_ids\": \"</s> The Russian troop stations were respectively located in Jalininggele, Simolingsike and Yelinia located 300 kilometers (250 miles) away from Moscow in 3 countries on the Baltic Sea.</s><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad><pad>\"\r\n },\r\n```\r\n\r\nWhat I don't understand is why the previous version only works for `bart-base`, in my experiments. Another question is what is the correct/better way to use the model (to use `shift_tokens_right` or another approach?)\r\n\r\n", "Interesting.\r\n`shift_tokens_right` has always done better on my datasets, but it's interesting that you have the opposite experience. The old code `tgt_ids = y[:, 1:].clone()` doesn't work well for tokenizers (Marian, Pegasus, T5) that don't add a `<s>` token to the beginning of the sequence, because it deletes a token.\r\n\r\nIf you can replicate the results on a small/shareable dataset I would be happy to try to understand what's going on more deeply.", "I can see a changing behavior of `bart-large` between v3.0.2 and v3.1.0, which seems to be linked to your findings. Here's a minimal example for language generation:\r\n```py\r\nimport transformers\r\n\r\nfrom transformers import (\r\n BartTokenizer,\r\n BartForConditionalGeneration,\r\n)\r\n\r\nprint(f'** transformers v{transformers.__version__} **')\r\n\r\ntokenizer = BartTokenizer.from_pretrained('facebook/bart-large')\r\nmodel = BartForConditionalGeneration.from_pretrained('facebook/bart-large')\r\n\r\ninput_txt = 'This is <mask> sentence.'\r\nprint(f'Input: \"{input_txt}\"')\r\n\r\ninputs = tokenizer.encode(input_txt, return_tensors='pt')\r\noutputs = model.generate(inputs)\r\noutput_txt = tokenizer.decode(outputs[0], skip_special_tokens=True)\r\n\r\nprint(f'Output: \"{output_txt}\"')\r\n```\r\nFor v3.0.2, it correctly produces\r\n```bash\r\n** transformers v3.0.2 **\r\nSome weights of BartForConditionalGeneration were not initialized from the model checkpoint at facebook/bart-large and are newly initialized: ['final_logits_bias']\r\nYou should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\r\nInput: \"This is <mask> sentence.\"\r\nOutput: \"This is a partial sentence.\"\r\n```\r\nwhile v3.1.0 repeats the first token:\r\n```bash\r\n** transformers v3.1.0 **\r\nInput: \"This is <mask> sentence.\"\r\nOutput: \"ThisThis is a sentence.\"\r\n```", "Digging a bit deeper, I can trace the issue back to this line https://github.com/huggingface/transformers/blob/4b3ee9cbc53c6cf6cee6bfae86cc2c6ec0778ee5/src/transformers/modeling_bart.py#L1114\r\nand, in turn, the default value of `force_bos_token_to_be_generated`:\r\nhttps://github.com/huggingface/transformers/blob/4b3ee9cbc53c6cf6cee6bfae86cc2c6ec0778ee5/src/transformers/configuration_bart.py#L140\r\n\r\nTo restore behavior from v3.0.2, we can change that value manually\r\n```py\r\n...\r\nconfig = BartConfig.from_pretrained('facebook/bart-large')\r\nconfig.force_bos_token_to_be_generated = True\r\n\r\ntokenizer = BartTokenizer.from_pretrained('facebook/bart-large')\r\nmodel = BartForConditionalGeneration.from_pretrained('facebook/bart-large', config=config)\r\n...\r\n```\r\nwhich gives\r\n```bash\r\n** transformers v3.1.0 **\r\nInput: \"This is <mask> sentence.\"\r\nOutput: \"This is a partial sentence.\"\r\n```\r\nand even\r\n```bash\r\n** transformers v3.4.0 **\r\nInput: \"This is <mask> sentence.\"\r\nOutput: \"This is a partial sentence.\"\r\n```\r\n@sshleifer What's the best approach to fix this? Modify bart-large's config.json?", "Your solution is awesome, great catch!\r\n\r\nI think the right fix is to\r\n+ Update the docs\r\n+ add `task_specific_params : {'fill_mask': {'force_bos_token_to_be_generated': 'true'}` to `bart-base` and `bart-large` configs.\r\n\r\nI am hesitant to change the default because `force_bos_token_to_be_generated = False` seems to be optimal for many fine-tuning tasks.", "Added a mask filling example to the docs in #8421 .", ":+1: Brilliant, thanks a lot @sshleifer !", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "Hello \r\nI'm using transformers 4.8.2 but there's still issue about same problem.\r\nI changed config.force_bos_token_to_be_generated=True.\r\n********The Result**************\r\ninput_txt = 'This is <mask> sentence.'\r\noutput_txt = 'ThisThis is a sentence.'\r\n\r\nanyone experience this??? ", "> Hello I'm using transformers 4.8.2 but there's still issue about same problem. I changed config.force_bos_token_to_be_generated=True. ********The Result************** input_txt = 'This is sentence.' output_txt = 'ThisThis is a sentence.'\r\n> \r\n> anyone experience this???\r\n\r\nHi @yeonsookKwak I have the same issue. Would you please share the solution if any? 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" } ]

[]

[ "Closed in favor of #8002 " ]

[ { "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" } ]

[]

[]

[]

[]

[ "Oh! I'll take a look, thanks for the report @tholor ", "Thanks for the fast fix @thomwolf ! Very much appreciated!" ]

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

[]

[ "Hello! In order to use the xlm-roberta-large model, why don't you use the `from_pretrained` method?\r\n\r\n```py\r\nfrom transformers import XLMRobertaModel\r\n\r\nmodel = XLMRobertaModel.from_pretrained(\"xlm-roberta-large\")\r\n```", "> Hello! In order to use the xlm-roberta-large model, why don't you use the `from_pretrained` method?\r\n> \r\n> ```python\r\n> from transformers import XLMRobertaModel\r\n> \r\n> model = XLMRobertaModel.from_pretrained(\"xlm-roberta-large\")\r\n> ```\r\nThanks for your answer!\r\nI tried do it like what you said, but I couldn't linked the URL to download the model, so I try to download the model、cofig、tokenizer to local and load it. \r\nso, The question what I said was I have not find the vocab.txt to generate the tokenizer.", "You don't need the URL to download the model, you can just use the identifier as its shown in my message. Or is there a reason why you want to have the URLs? ", "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" } ]

[]

[]

[]

[]

[ "Let's wait for Victor to answer on that issue before merging.", "looks good!" ]

[ { "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" } ]

[]

[ "@jcblaisecruz02 looks like your config.json files do not contain a `architectures` field nor a `model_type`, so your models might be incorrectly categorized – any way you could add those? Thank you!", "> @jcblaisecruz02 looks like your config.json files do not contain a `architectures` field nor a `model_type`, so your models might be incorrectly categorized – any way you could add those? Thank you!\r\n\r\nAh gotcha! I'll add those. Thanks!", "(easiest way should be to just call `.save_pretrained` again)" ]

[]

[]

[ "This looks good but I only see `test_output.txt` in the artifacts, for some reason?", "you must be looking at the wrong job? As I said I only did it for one job at the moment - this one:\r\nhttps://app.circleci.com/pipelines/github/huggingface/transformers/14359/workflows/b38c8e8c-2867-4366-a907-a202da9bc9ee/jobs/104410/steps\r\n```\r\n ~/transformers/output.txt\r\n ~/transformers/tests_durations.txt\r\n ~/transformers/tests_failures.txt\r\n ~/transformers/tests_passes.txt\r\n ~/transformers/tests_short_summary.txt\r\n ~/transformers/tests_stats.txt\r\n```", "Ah my bad, I miscklicked!", "@sshleifer or @sgugger - I configured github artifacts in `self-push.yaml` of this PR - would one of you be able to start that job for me as I have no perms to do so. Thank you very much!\r\n\r\nI hope I did it right, I added:\r\n```\r\n - name: test suite reports artifacts\r\n uses: actions/upload-artifact@v2\r\n with:\r\n name: tests_results\r\n path: tests_*\r\n```\r\nI'm not sure whether this should be `path: ~/transformers/tests_*` like it was on circle_ci config - it should pick it up from the cwd.\r\n\r\nI currently added it only to `run_tests_torch_and_tf_gpu` - so in theory it should upload the reports to the workflow results.\r\n\r\nFor reference, the information on this setup is at this 2 pages:\r\n* https://docs.github.com/en/free-pro-team@latest/actions/guides/storing-workflow-data-as-artifacts\r\n* https://github.com/actions/upload-artifact#usage\r\n", "I can't figure out how to run a github actions workflow against a branch. It looks good enough that we I'm happy to just acknowledge that this could break on merge, in which case we'd send a follow up PR.", "Thank you for trying, @sshleifer \r\n\r\nAh, it's not finished yet, merge-wise - it's very rough on edges.\r\n* I just want to figure out how to make the results available on github actions in parallel with\r\n* waiting on you guys to hear what reports do you want and which not before finalizing this.\r\n\r\nCan you suggest a different way of testing this? This was your recommendation in first place - to test it on a PR branch - except I can't test it since I don't have permissions to access the runners. Surely there must be a way of testing this?\r\n\r\nAlternatively, we could go as simple as creating a new github workflow job that simply runs a job of `echo test > tests_1.txt; echo test2 > tests_2.txt` and then uploads `tests_*` as an artifact and checking that it is what you want. It should just work, since the docs suggest that as an example. Once we know it's working then the rest is easy.\r\n\r\nEarlier you were talking about some possible problems with this - something about the job being always successful, I can't find that comment - but I am pretty sure there is no such issue with the approach I implemented - where `pytest` generates all the report files and we don't need to do anything about its log parsing.", "> waiting on you guys to hear what reports do you want and which not before finalizing this.\r\n\r\nDon't wait, just make a sensible choice that's easy to change. Lean towards fewer reports.\r\n\r\n> Can you suggest a different way of testing this? \r\n\r\nI don't know a good way of testing github actions. [act](https://github.com/nektos/act) looks promising, but I've never used it. The issue is not permissions it is that github workflows, afaict, cannot be run against arbitrary branches. There is a \"rerun all jobs\" button, but it will just rerun on master. Would be incredibly valuable if you figured out how to test github actions locally.\r\n\r\nHere is everything I can see for self-push at https://github.com/huggingface/transformers/actions/runs/326336555/workflow\r\n\r\n", "I agree with Sam that we can merge to test and iterate if the reports look wrong (as soon as we're sure that the circleCI part is good to go, which we can test on this PR). From what I understand, the PR adds a new job, so it does not break the existing ones/reports.", "I will work on completing this and we can put it in for one circle-ci and one github workflow and see how it goes - thank you for your feedback, @sshleifer and @sgugger ", "This is good to merge." ]

[]

[]

[ "It seems not work until I save the new tokenizer. I close the issue.\r\n```\r\ntokenizer.save_pretrained('/path/to/tokenizer')\r\ntokenizer = BertTokenizer.from_pretrained('/path/to/tokenizer')\r\n```", "looks like tokenizer.vocab_size does not update after add tokens. but len(tokenizer) shows correct number ", "Yes, the `vocab_size` only contains the information relative to the initial vocabulary. You can find the added tokens either in `tokenizer.get_added_vocab()`, which returns the dictionary, or `tokenizer.added_tokens_encoder`, which returns the amount of added tokens." ]

[]

[]

[ "One followup is to update the `test_trainer_distributed` to work with pytest. Then ideally, if we could have one command to run all those tests, that would be awesome (maybe we can use a pytest marker to mark all distributed-specific tests so it's easy to select them all?)", "I will port `test_trainer_distributed` - thank you for flagging that, @sgugger \r\nTracking it here: https://github.com/huggingface/transformers/issues/8058" ]

[ { "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" } ]

[]

[]

[]

[]

[]

[]

[]

[ "cc @patil-suraj @patrickvonplaten @xixiaoyao" ]

[]

[]

[ "Thank you for the heads up - I will be working on all these related issues shortly - too much data indeed, but not just that.", "I'd say remove them complete for now and also -rA - I need to experiment and see how to make this data available w/o making logs unusable. ", "More context on github actions:\r\n\r\n\r\nif we can somehow catch the return value of \r\nbash\r\n```\r\nx= python -m pytest -n 1 --dist=loadfile -s examples --durations=50 | tee test_output.txt\r\nsave test_output.txt # always succeeds\r\nsys.exit(x)\r\n```\r\nor something like that, we can make huge progress on the github actions issue and start making artifacts files.\r\n\r\nThe reason artifacts files broke was that even in the below code, even if line 1 raises an error, line 2 succeeds so github actions thinks the job succeeded\r\n```bash\r\npython -m pytest -n 1 --dist=loadfile -s examples --durations=50 | tee test_output.txt\r\nsave test_output.txt # always succeeds\r\n```", "oh, unless I'm missing something, we don't need any of the workarounds. \r\n\r\nI already have the first requested component (failures) working, see: https://github.com/huggingface/transformers/pull/7995\r\n\r\nCheck out the resulting artifacts:\r\nhttps://app.circleci.com/pipelines/github/huggingface/transformers/14354/workflows/1ccd616e-218f-4ae1-b413-91d2faa0e942/jobs/104363/artifacts\r\n\r\nthis is what we want right?\r\n\r\n`pytest` provides hooks for doing this kind of work, so it's just figuring out which hooks to call.\r\n\r\nIn your example instead of `x = cmd` what you need to save is `$?` which is the exit status of the command.\r\n", "That's great, but note that this is all much easier in circleci. My ask is to make it work in github actions.\r\nThe failures are already pretty easy to find in circleci.\r\n\r\n\r\n2) you mean\r\n```bash\r\nx= python -m pytest -n 1 --dist=loadfile -s examples --durations=50 | tee test_output.txt\r\nsave test_output.txt # always succeeds\r\nsys.exit($x)\r\n```\r\n?", "`test_failures.txt` is really nice!", "Ah, good point. let me see what other handy reports I can squeese in circle-ci and then I will move to github actions.\r\n\r\nI'm not following your question yet, let me get to github actions and then it'll probably make sense, but yes I'm referring to that example when I said:\r\n> In your example instead of `x = cmd` what you need to save is `$?` which is the exit status of the command.\r\n\r\ni.e. `sys.exit($?)` but you must save it right away upon `pytest` completion, since the next command will overwrite it." ]

[ { "id": 1834067346, "node_id": "MDU6TGFiZWwxODM0MDY3MzQ2", "url": "https://api.github.com/repos/huggingface/transformers/labels/Documentation", "name": "Documentation", "color": "77cc3b", "default": false, "description": "" }, { "id": 1990918270, "node_id": "MDU6TGFiZWwxOTkwOTE4Mjcw", "url": "https://api.github.com/repos/huggingface/transformers/labels/Good%20First%20Issue", "name": "Good First Issue", "color": "bbf794", "default": false, "description": "" } ]

[]

[ "Hi, has anyone picked this up? I can give it a go in the coming week or two when I have some let off from releases at work if no-one is doing this. ", "Nobody has picked this up yet, would love to see such a contribution!", "Awesome, have our last release till year end freeze this week at work. I'll get in there afterwards. Would like to learn about some of the libraries involved in this project this seems like a good intro. " ]

[]

[]

[ "@patrickvonplaten \r\n+ I deleted the `_force_token_id` function, replaced with faster `tf.where` one-liner. (+ added regression test). \r\n+ Replaced unneeded `TFSharedEmbedding` with `tf.keras.layers.Embedding`\r\n+ switched all `.shape` to `shape_list`\r\n\r\nWDYT?" ]

[]

[]

[ "your 2nd option is correct here: \r\n\r\n```python\r\n# pad_token prepended, eos_token unmasked in attention\r\ndecoder_input_ids: tensor([[0, 2018, 55, 1, 0, 0, 0]])\r\ndecoder_attention_mask: tensor([[1, 1, 1, 1, 0, 0, 0]])\r\n```\r\n\r\n1) You have to start with `decoder_start_token_id = pad_token_id` in T5 and \r\n2) the last EOS token should be attended to because the model \"should learn\" when the sentence is finished.", "@patrickvonplaten Thanks, Patrick! That makes perfect sense. You're awesome!", "@patrickvonplaten I noticed that when you pass this to the model:\r\n\r\n```\r\ndecoder_input_ids: tensor([[2018, 55, 1, 0]])\r\ndecoder_attention_mask: tensor([[1, 1, 1, 0])\r\n```\r\n`T5ConditionalGeneration` changes it to this before passing it to the decoder:\r\n\r\n```\r\n# Masks the eos_token\r\n# Correctly prepends an extra pad_id to inputs BUT appends a pad_token to attention_mask\r\ndecoder_input_ids: tensor([[0, 2018, 55, 1, 0]])\r\ndecoder_attention_mask: tensor([[1, 1, 1, 0, 0])\r\n```\r\nSo that you have to actually pass this initially:\r\n\r\n```\r\n# Pass this to model\r\ndecoder_input_ids: tensor([[2018, 55, 1, 0]])\r\ndecoder_attention_mask: tensor([[1, 1, 1, 1])\r\n\r\n# Which turns into this\r\ndecoder_input_ids: tensor([[0, 2018, 55, 1, 0]])\r\ndecoder_attention_mask: tensor([[1, 1, 1, 1, 0])\r\n```", "Hey @alexorona - sorry I don't quite follow here...could you provide a code example that I can run to see what you mean? :-) " ]

[ { "id": 2796628563, "node_id": "MDU6TGFiZWwyNzk2NjI4NTYz", "url": "https://api.github.com/repos/huggingface/transformers/labels/WIP", "name": "WIP", "color": "234C99", "default": false, "description": "Label your PR/Issue with WIP for some long outstanding Issues/PRs that are work in progress" } ]

[]

[ "@patrickvonplaten @sshleifer @sgugger Could we solve the errors here to have `torch>=1.3` instead?", "May be it's simpler to set it to 1.4+ and only if someone asks for it to bother with 1.3?", "After discussion with the team, we'll take a look at supporting v1.3+ in the coming days, and if it requires too many efforts we'll stick with v1.4+. We'll take this as an opportunity to test the versions we say we support as well (1.3, 1.4, 1.5, 1.6, 1.7) so that the README isn't full of empty promises :slightly_smiling_face:.\r\n\r\nCould you keep your PR as-is for the coming days, and let me come back to you when we've reached a consensus?", "That's an excellent and clear proposition, @LysandreJik - thank you!", "> We'll take this as an opportunity to test the versions we say we support as well \r\n\r\nIf I may propose a scheduled CI that runs all tests for each of the supported versions, say, once a week or so. Probably `tf` too.\r\n\r\nI trust you will have the best plan. ", "ping", "will try to work on it today - are we sticking to torch 1.3 @LysandreJik ?\r\n\r\nMaybe we could discuss also whether we can do some more general optimizations in the lib then (I think we can safely change the attention masks to bools then)", "Yes, we are! There's a branch in progress here: https://github.com/huggingface/transformers/tree/previous-torch\r\nFeel free to push fixes onto it directly. I've been planning on doing so right after the TAPAS merge.", "The branch tests out torch versions going back to v1.3. It's not setup for slow tests right now, and it tests on every commit. I haven't really thought about if this is the best way to do so, but it's certainly easier to debug the failing tests this way.", "I don't suppose there is a point at resolving the conflict, right? ", "ping", "I'd like to get to it as soon as I have some availability.", "This issue has been automatically marked as stale and been closed because it has not had recent activity. Thank you for your contributions.\n\nIf you think this still needs to be addressed please comment on this thread.", "ping", "It's on the roadmap!", "It seems wasteful trying to keep up with the conflicts for more than 6 months. Since it's on the roadmap I think it's safe to close this one now." ]

[]

[]

[]

[ { "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" } ]

[]

[]

[]

[]

[ "I would fix this by training on much more data (like 1000 obs) and getting the loss down much further.", "I used more iterations - 6 was enough for 1 gpu, 10 for 2, so I went with 10.\r\n\r\nThis issue will be resolved by https://github.com/huggingface/transformers/pull/7965" ]

[]

[]

[]

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

[]

[ "Could you provide the information related to your environment, as well as the code that outputs this error, like it is asked in the issue template?", "I am facing same issue as the given issu 'DistributedDataParallel' is custom class created by coder that is having base model available in Transformer repo\r\n\r\nWhere in below code that class is \"SentimentClassifier\"\r\n\r\n class SentimentClassifier(nn.Module):\r\n \r\n def __init__(self, n_classes):\r\n super(SentimentClassifier, self).__init__()\r\n self.bert = BertModel.from_pretrained(\"bert-base-multilingual-cased\")\r\n self.drop = nn.Dropout(p=0.3)\r\n self.out = nn.Linear(self.bert.config.hidden_size, n_classes)\r\n \r\n def forward(self, input_ids, attention_mask):\r\n _, pooled_output = self.bert(\r\n input_ids=input_ids,\r\n attention_mask=attention_mask\r\n )\r\n output = self.drop(pooled_output)\r\n return self.out(output)`\r\n\r\nthat is why it is giving error -\r\n\r\n SentimentClassifier object has no attribute 'save_pretrained'\r\n\r\nwhich is correct but I also want to know how can I save that model with my trained weights just like the base model so that I can Import it in few lines and use it.\r\n\r\nonly thing I am able to obtaine from this finetuning is a .bin file \r\nand I am not able to load state dict also\r\n\r\nI am looking for way to save my finetuned model with \"save_pretrained\"", "Instead of inheriting from `nn.Module` you could inherit from `PreTrainedModel`, which is the abstract class we use for all models, that contains `save_pretrained`. Can you try that?", "fine-tuning codes I seen on hugging face repo itself shows the same way to do that...so I did that...\r\nbdw I will try as you said and will update here\r\n\r\nhere is the link i refered that from\r\n\r\nhttps://huggingface.co/transformers/notebooks.html\r\n\r\n", "Hey, My code just like this\r\n\r\n```\r\nfrom transformers import EncoderDecoderModel, BertTokenizer\r\nimport torch\r\nimport argparse\r\nimport os\r\nimport argparse\r\nimport torch.multiprocessing as mp\r\nimport torchvision\r\nimport torchvision.transforms as transforms\r\nimport torch.nn as nn\r\nimport torch.distributed as dist\r\n\r\ndef main():\r\n parser = argparse.ArgumentParser()\r\n args = parser.parse_args()\r\n args.max_src_len = 512\r\n args.max_dst_len = 128\r\n args.gpus = 4\r\n args.world_size = args.gpus\r\n args.epoches = 30\r\n mp.spawn(train, nprocs=args.gpus, args=(args,))\r\n\r\ndef train(gpu, args):\r\n rank = gpu\r\n dist.init_process_group( \r\n \tbackend='nccl', \r\n \t\tinit_method='tcp://127.0.0.1:23456', \r\n \tworld_size=args.world_size, \r\n \trank=rank \r\n ) \r\n torch.manual_seed(0)\r\n model = EncoderDecoderModel.from_pretrained(\"bert2bert\")\r\n torch.cuda.set_device(gpu)\r\n model = model.to(gpu)\r\n optimizer = torch.optim.Adam(model.parameters(), lr=1e-3)\r\n model = nn.parallel.DistributedDataParallel(model, device_ids=[gpu])\r\n dataset_path = 'dataset/example.json'\r\n vocab_path = 'dataset/vocab.txt'\r\n dataset = CNNDataset(dataset_path, vocab_path, args)\r\n train_sampler = torch.utils.data.distributed.DistributedSampler(\r\n \tdataset,\r\n \tnum_replicas=args.world_size,\r\n \trank=rank\r\n )\r\n dataloader = DataLoader(dataset, batch_size=32, shuffle=False, \r\n num_workers=0,\r\n pin_memory=True,\r\n sampler=train_sampler)\r\n cnt = 0\r\n for epoch in range(args.epoches):\r\n for src, dst in dataloader:\r\n\r\n src = torch.stack(src).to(gpu)\r\n dst = torch.stack(dst).to(gpu)\r\n mask = (src!=0)\r\n mask = mask.long()\r\n outputs = model(input_ids=src, attention_mask=mask, decoder_input_ids=dst, labels=dst, return_dict=True)\r\n loss, logits = outputs.loss, outputs.logits\r\n optimizer.zero_grad()\r\n\r\n loss.backward()\r\n\r\n optimizer.step()\r\n\r\n if cnt % 1000 == 0 and gpu == 0 :\r\n model.save_pretrained(\"bert2bert\")\r\n cnt = cnt + 1\r\n\r\n\r\nif __name__ == '__main__':\r\n\r\n main()\r\n\r\n```\r\n@LysandreJik ,@ganeshkharad2", "I can save this with state_dict. But how can I load it again with from_pretrained method ? ", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "> I can save this with state_dict. But how can I load it again with from_pretrained method ?\r\n\r\nHi, i meet the same problem, have you solved this problem? or?", "> I can save this with state_dict. But how can I load it again with from_pretrained method ?\r\n\r\nHi, Did you find any workaround for this? Thanks in advance.", "Any solution for this?" ]

[]

[]

[ "Hey @yhznb, \r\n\r\nWe try to mainly use the github issues for bugs in the library. For more customized questions it would be great if you could use https://discuss.huggingface.co/ instead.\r\n\r\nRegarding your question I would just add a layer to `BertLMHeadModel` wherever you want to and then build your `EncoderDecoderModel` from `BertModel` (encoder) & your use-case speciifc `BertLMHeadModel` (decoder).", "Hey, @patrickvonplaten, I have the same question. Can you provide a example of building the EncoderDecoderModel from BertModel (encoder) & use-case speciifc BertLMHeadModel ? I can't find this in the official document. Thank you very much .", "I think the model(EncoderDecoderModel) outputs all the hidden states at once . And I want to control it step by step. For example , I want to change the LMhead of Decoder by concatenating another vector. The problem is that the DecoderModel outputs all the hidden states at once. I want to control it for step by step decoding. In other words. I want to use the concatenated vector as the hidden state for generation and use the generated word vector for next step's input. How can I change the model or call the interface properly ? Is it possible under the framework of huggingface ? \r\nThank you very much ! @patrickvonplaten", "I also raised this in the forum. Does this issue need to be closed ?\r\nThe link is here :\r\nhttps://discuss.huggingface.co/t/control-encoderdecodermodel-to-generate-tokens-step-by-step/1756", "thank you very much ! @patrickvonplaten ", "Have you solved your question ？ @AI678 I think it is all about changing the LMhaed and the calculation of logits. But I don't know how to change it .", "Yes , you are right. @yhznb", "> Hey @yhznb,\r\n> \r\n> We try to mainly use the github issues for bugs in the library. For more customized questions it would be great if you could use https://discuss.huggingface.co/ instead.\r\n> \r\n> Regarding your question I would just add a layer to `BertLMHeadModel` wherever you want to and then build your `EncoderDecoderModel` from `BertModel` (encoder) & your use-case speciifc `BertLMHeadModel` (decoder).\r\n\r\nSorry, I misunderstood what you meant. This is a feature to be developed. So, how long can this feature be developed ? thank you for your response.", "Hey , I have similar demands. Because I think using only vanilla bert2bert or roberta2roberta is not sufficient for abstractive summarization . For fluency and information richness, we should consider to change the top layer of decoder for further learning.", "Hey, @patrickvonplaten, when do you want to release that ? ", "@nlpLover123 , you can control it step by step. But I think it is too slow for a large dataset like cnn-dailymail.\r\nAnd I also want to ask when do you want to release that ? @patrickvonplaten \r\nIf that needs too much time, maybe I would write a encoder_decoder_model from scratch. Because I have little time to wait for that. \r\nThank you very much .\r\n", "that is too difficult @AI678 .Maybe it is slower that step by step generation.", "so I just want to make a specific change at the LMhead layer @moonlightarc ", "@AI678 , I don't think we are planning on releasing such a feature into the library. It's a very specific request and I'd suggest that you try to fork the repo and make the changes according to your needs" ]

[ { "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" } ]

[]

[]

[]

[]

[]

[]

[]

[ "Pinging @sgugger " ]

[]

[]

[ "But If 1 models implies 1 task, how should we cope with models that are able to do multiple things (like `t5-base`) ? \r\n\r\nI think `datasets` does it correctly in that it does not make any choice on your behalf, but instead raises an Exception with your available choices.", "> But If 1 models implies 1 task, how should we cope with models that are able to do multiple things (like `t5-base`) ?\r\n> \r\n> I think `datasets` does it correctly in that it does not make any choice on your behalf, but instead raises an Exception with your available choices.\r\n\r\nI think T5Base should default to a `ForConditionalPipeline` (which is more or less the `Text2TextPipeline` we have right now). Then the user could either provide a pipeline config that makes sure \"summarization\" or \"translation\" params are used for the pipeline. Note: In the end of the day all Seq2Seq pipelines are exactly the same -> they are all based on `.generate()` and they only differ on which params (`max_length`, `prefix`, ...) are used. Or/And we create a very shallow \"alias\" pipeline named `class TranslationPipeline(ConditionalGenerationPipeline)` that only overwrites the config params similar to what we do here: https://github.com/huggingface/transformers/blob/901e9b8eda2fe88af717f960ddc05cac1803679b/src/transformers/pipelines.py#L568 right now. \r\n\r\nMaybe @mfuntowicz can also give a bit more context on the Pipeline v2 vision here." ]

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

[]

[ "It was recently added, so you may need to upgrade your version of transformers.", "> It was recently added, so you may need to upgrade your version of transformers.\r\n\r\nThanks,it works!\r\nI use transformers in \r\n\r\n> kaggle notebook\r\n\r\n , maybe there is some bug in such online applications. I once tried to update transformer to see whether this error would not appear, but it was not work. \r\n\r\nThis time I close my browser and restart the kaggle notebook. Everything going well ! ", "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've just successfully solved this problem using this command.\r\n\r\n`pip install transformers --upgrade`\r\n" ]

[]

[]

[]

[]

[]

[ "As far as I can see, the model you are trying load doesn't have a compliant format. We cannot help more without the full error stack.", "Edit, added the full error stack.\r\nAlso, this is fairly easy to reproduce." ]

[]

[]

[ "Ok so should be ready to merge after a quick review @LysandreJik @sgugger!" ]

[]

[]

[ "Tagging a few people working on the pipelines and @sshleifer because I've added additional imports of TF Bart in `modeling_tf_auto` to make the pipelines happy.", "Ok. great, thanks @sgugger and @sshleifer.\r\n\r\nI took the occasion to speed up the CI test suite (cc @stas00) by:\r\n- spinning out the pipeline test in a separate job\r\n- reducing the dual framework (tf+pt) tests overhead by focusing them on the PT+TF cross interactions (adding new tests at the same time) and removing the double testing with the tf and pytorch standalone tests.\r\n\r\nReady to merge imo.", "Whoah! this is amazing - the slowest job is now the fastest! Thank you!!!\r\n\r\nI think there is only one potential issue with it - the half-baked until now codecov report is now completely useless since it no longer covers all tests so just as well remove it completely.", "@stas00 Did just that https://github.com/huggingface/transformers/commit/829b9f8cc321aa28396e6203e0f21eed26b132f7\r\n\r\nRemoved codecov from the repo as well.", "It's still there ;)\r\n```\r\n.circleci/config.yml: - run: pip install codecov pytest-cov\r\n.circleci/config.yml: - run: RUN_PT_TF_CROSS_TESTS=1 python -m pytest -n 8 --dist=loadfile -rA -s ./tests/ -m is_pt_tf_cross_test --cov --durations=0 | tee output.txt\r\n.circleci/config.yml: - run: codecov\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" } ]

[]

[ "I think we should also add some meta-information (see [here](https://github.com/huggingface/model_card)) to the models:\r\n\r\n```\r\n---\r\nlanguage: de\r\nlicense: mit\r\ndatasets:\r\n- wikipedia\r\n---\r\n```\r\n\r\nAnd maybe we need to add `masked-lm` to the `tags` array, so that we can use the inference widget on the model page to do some nice masking experiments :)", "Awesome collaboration btw :heart: :hugs: ", "> And maybe we need to add `masked-lm` to the `tags` array, so that we can use the inference widget on the model page to do some nice masking experiments :)\r\n\r\nShouldn't need to (in theory)", " ```json\r\n\"architectures\": [\r\n \"BertForMaskedLM\"\r\n ],\r\n```\r\n\r\nis currently missing in our BERT configs -> @brandenchan would it be possible that you add it :hugs: ", "> ```json\r\n> \"architectures\": [\r\n> \"BertForMaskedLM\"\r\n> ],\r\n> ```\r\n> \r\n> is currently missing in our BERT configs -> @brandenchan would it be possible that you add it 🤗\r\n\r\n@stefan-it Done! Out of interest, what's the difference between BertForMaskedLM and BertForPretraining?", "If I remember correctly BertForPretraining loads a LM head and a NSP head (both heads trained during pretraining). Am I correct @LysandreJik?" ]

[]

[]

[]

[]

[]

[ "@VictorSanh what do you think?" ]

[]

[]

[ "Hey @rabeehk - for more specific cases we recommend that you fork master and tweak the model however you would like to so that it fits your purpose. \r\nI didn't understand 100% what kind of script you are looking for, but here you can browse some of the T5 scripts, we have collected: https://github.com/huggingface/transformers/tree/master/notebooks#-transformers-notebooks", "Hi\nthanks for getting back to me. I am looking for a script showing to train\nT5 with multiple tasks. this is when they create a T5 registery mixture\ndataset in their original code and train one model on a mixture of several\ndataset.\nDo you know if huggingface version works fine for handling multiple\ndatasets? and do you know how performance is different from JAX\nimplementation? is this more or less the same? can one train the base T5\nwith a mixture of datasets with huggingface code?\n\nthank you very much.\nBest\nRabeeh\n\nOn Fri, Oct 23, 2020, 8:20 AM Patrick von Platen <[email protected]>\nwrote:\n\n> Hey @rabeehk <https://github.com/rabeehk> - for more specific cases we\n> recommend that you fork master and tweak the model however you would like\n> to so that it fits your purpose.\n> I didn't understand 100% what kind of script you are looking for, but here\n> you can browse some of the T5 scripts, we have collected:\n> https://github.com/huggingface/transformers/tree/master/notebooks#-transformers-notebooks\n>\n> —\n> You are receiving this because you were mentioned.\n> Reply to this email directly, view it on GitHub\n> <https://github.com/huggingface/transformers/issues/7966#issuecomment-714945934>,\n> or unsubscribe\n> <https://github.com/notifications/unsubscribe-auth/ABP4ZCEVU5IEVMEYHK4VLSTSMEOELANCNFSM4S2XT3OA>\n> .\n>\n" ]

[]

[]

[ "@sshleifer, the slow test isn't working for me prior to this PR with 0 or 1 gpu:\r\n\r\n```\r\nCUDA_VISIBLE_DEVICES=\"\" RUN_SLOW=1 pytest -sv examples/seq2seq/test_finetune_trainer.py::TestFinetuneTrainer::test_finetune_trainer_slow \r\n[...]\r\n\r\nself = <seq2seq.test_finetune_trainer.TestFinetuneTrainer testMethod=test_finetune_trainer_slow>\r\n\r\n @slow\r\n def test_finetune_trainer_slow(self):\r\n # There is a missing call to __init__process_group somewhere\r\n> output_dir = self.run_trainer(eval_steps=2, max_len=\"128\", model_name=MARIAN_MODEL, num_train_epochs=3)\r\n\r\nexamples/seq2seq/test_finetune_trainer.py:35: \r\n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \r\nexamples/seq2seq/test_finetune_trainer.py:113: in run_trainer\r\n main()\r\nexamples/seq2seq/finetune_trainer.py:199: in main\r\n model = AutoModelForSeq2SeqLM.from_pretrained(\r\nsrc/transformers/modeling_auto.py:1118: in from_pretrained\r\n return MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING[type(config)].from_pretrained(\r\nsrc/transformers/modeling_utils.py:947: in from_pretrained\r\n model = cls(config, *model_args, **model_kwargs)\r\nsrc/transformers/modeling_bart.py:964: in __init__\r\n base_model = BartModel(config)\r\nsrc/transformers/modeling_bart.py:843: in __init__\r\n self.encoder = BartEncoder(config, self.shared)\r\nsrc/transformers/modeling_bart.py:315: in __init__\r\n self.embed_positions = SinusoidalPositionalEmbedding(\r\nsrc/transformers/modeling_bart.py:1331: in __init__\r\n self.weight = self._init_weight(self.weight)\r\n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _\r\n\r\nout = Parameter containing:\r\ntensor([[ 2.2244e+00, -1.2380e+00, -3.5307e-01, ..., -1.0924e+00,\r\n -1.3130e+00, 1.7737... [-3.8906e-01, 9.2203e-01, 1.7887e-01, ..., -1.7493e-01,\r\n -1.6993e+00, 2.0896e-01]], requires_grad=True)\r\n\r\n @staticmethod\r\n def _init_weight(out: nn.Parameter):\r\n \"\"\"Identical to the XLM create_sinusoidal_embeddings except features are not interleaved.\r\n The cos features are in the 2nd half of the vector. [dim // 2:]\r\n \"\"\"\r\n n_pos, dim = out.shape\r\n position_enc = np.array(\r\n [[pos / np.power(10000, 2 * (j // 2) / dim) for j in range(dim)] for pos in range(n_pos)]\r\n )\r\n> out[:, 0 : dim // 2] = torch.FloatTensor(np.sin(position_enc[:, 0::2])) # This line breaks for odd n_pos\r\nE RuntimeError: a view of a leaf Variable that requires grad is being used in an in-place operation\r\n```\r\n**edit**: After messing around with my conda env due to ever-breaking tf, this went away, but a new thing came instead: https://github.com/huggingface/transformers/issues/7982\r\n", "interesting, I can't reproduce that. What's your `transformers-cli env`? Does it fail after the change?", "No, it fails on master.\r\n\r\n```\r\n- `transformers` version: 3.4.0\r\n- Platform: Linux-4.15.0-118-generic-x86_64-with-glibc2.10\r\n- Python version: 3.8.5\r\n- PyTorch version (GPU?): 1.8.0.dev20201020 (True)\r\n- Tensorflow version (GPU?): 2.3.1 (True)\r\n- Using GPU in script?: <fill in>\r\n- Using distributed or parallel set-up in script?: <fill in>\r\n```", "I'd open an issue \"sinusoidal positional embedding broken on torch 1.8\".\r\nReasoing: [these](https://github.com/huggingface/transformers/blob/master/tests/test_modeling_bart.py#L606) fast tests pass CI and I can't replicate on on torch 1.5.\r\n\r\nIs this ready to merge otherwise?", "Need to sort out this first: https://github.com/huggingface/transformers/issues/7982\r\n\r\nIt's mostly ready otherwise, but the slow test will fail as it has nothing to do with this PR. \r\n\r\n**edit** resolved in this PR.", "Out of curiosity, how did you resolve the bleu issue?", "Great work, btw! This is awesome. Now we can tell people to run these tests before they break things :) \r\nApparently there is multi-gpu ci running src/ tests at some frequency FYI, I think through gh actions.", "I replied in the other issue: I used more iterations - 6 was enough for 1 gpu, 10 for 2, so I went with 10.\r\n\r\nI think if someone tries it on more than 2 gpus it might need even more iterations - could probably codify this with a factor of n_gpus.\r\n", "Documenting it now https://github.com/huggingface/transformers/pull/7993\r\nIf you think anything needs to be added please let me know.\r\nIt will get better over time.", "> Apparently there is multi-gpu ci running src/ tests at some frequency FYI, I think through gh actions.\r\n\r\nOnce a day yes:\r\nhttps://github.com/huggingface/transformers/blob/master/.github/workflows/self-scheduled.yml#L74", "Later I want to add the non-interactive IO pipe options - in case it hangs for someone - by default it could be non-interactive - always works, and only make it interactive for debug purposes." ]

[]

[]

[]

[]

[]

[ " Instead of doing `self.base = DistilBertModel(DistilBertConfig())` you can do `self.base = DistilBertModel(DistilBertConfig(vocab_size=119547))`.\r\n" ]

[]

[]

[ "The only difference I'm seeing between our script and yours is the data loading, here using `datasets`. Are we aware of TPU slowdown when using `datasets` @lhoestq, @thomwolf?", "I haven't tested `datasets` with TPU yet. I know @sgugger tried once, did you notice slowdowns ?", "Not for the new run_glue script introduced in #7917", "Is there any metrics/debug information I can provide?", "I have similar issues:\r\n* mxu utilization mostly 0%\r\n* around 150 s/it\r\n\r\nsetup:\r\n* n1-highmem-16\r\n* TPU v2-8\r\n\r\nI tried it both with a map-style dataset using `datasets`' wiki dump and an iterable-style dataset with the `Trainer` adapted. Same result. The slowdown is not on behalf of the data loading but of the forward-pass, loss computation and backpropagation on the tpu.", "Are you sure all your batches of inputs have the exact same shape? I tested thoroughly the script on TPUs with the datasets library and:\r\n- when inputs are not all of the same size, the training is excruciatingly slow (which is expected because XLA recompiles the code at each training step in this case)\r\n- when inputs are all of the same size, it runs smoothly and fast.\r\n\r\nThis is independent of using the datasets library or not, and this is expected behavior on TPUs, as XLA does not handle well dynamic shapes. ", "How can I check whether inputs are the same size? Or how can I pad the inputs so it has fixed size?", "Your dataset is hidden inside the `load_dataset` function, so I can't advise you on how to add padding. There are examples of this in the new `run_mlm.py` script.\r\nAs for checking your inputs are all of the same size, it's just a pass through your dataset:\r\n```\r\nshapes = []\r\nfor x in dataset:\r\n shapes.append(x[\"input_ids\"].shape)\r\nprint(set(shapes))\r\n``` ", "Keeping shapes constant fixed the speed issue for me. After few iteration (~ 5), the graph stabilized and the iteration speed went down from 150s/it to few seconds per batch.\r\n\r\nFurther readings:\r\n* https://github.com/pytorch/xla/blob/master/TROUBLESHOOTING.md#known-performance-caveats\r\n* https://github.com/pytorch/xla/issues/2383\r\n* https://github.com/pytorch/xla/issues/2368", "Fixed it, closing." ]

[]

[]

[ "It's correct, copied from fairseq (authors code).\r\nyou can think of `shift_tokens_right` as `shift_tokens_right_and_wrap_eos_to_position0`.\r\nIf you have empirical evidence that there is a change that improves fine-tuning, I'd be happy to incorporate it.\r\n", "After I updated my version for this one, I'm also confused by the decoder input. \r\n\r\nWith this new encoding, after finetuning, bart-large is outputting:\r\n\r\nExample 1:\r\n\r\n```\r\n\"labels\": \"<s> There are too many traitors among our compatriots,</s><pad><pad><pad>\",\r\n \"decoder_input_ids\": \"</s><s> There are too many traitors among our compatriots,</s><pad><pad>\",\r\n \"generated_ids\": \"</s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s></s>\"\r\n```\r\n\r\nExample 2:\r\n\r\n```\r\n\"labels\": \"<s> Freedom of speech is relative. We need take national conditions into account.</s>\",\r\n \"decoder_input_ids\": \"</s><s> Freedom of speech is relative. We need take national conditions into account.\",\r\n \"generated_ids\": \"</s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s><s></s>\"\r\n```\r\n\r\n\r\n\r\nI'm trying to figure out why.\r\n\r\n I'm using the code provided by the examples/seq2seq folder. T5 works fine, but bart does not.", "@leoribeiro if you are reporting a bug, could you explain what you did and what you expected more clearly in a separate issue?\r\nOtherwise, I don't understand why you say the encoding is new. `shift_tokens_right` hasn't changed.", "@sshleifer thank you for your reply. What I mean by new encoding is adding `</s>` at the beginning of the decoder inputs. I think that in a previous transformer version (2.11.0), the code for BART did not use `</s>` at the beginning of the decoder inputs, correct? In the 2.11.0 version, my experiments with `facebook/bart-large` were working with the following code:\r\n\r\n```\r\n def _step(self, batch):\r\n pad_token_id = self.tokenizer.pad_token_id\r\n source_ids, source_mask, y = batch[\"source_ids\"], batch[\"source_mask\"], batch[\"target_ids\"]\r\n y_ids = y[:, :-1].contiguous()\r\n lm_labels = y[:, 1:].clone()\r\n lm_labels[y[:, 1:] == pad_token_id] = -100\r\n outputs = self(source_ids, attention_mask=source_mask, decoder_input_ids=y_ids, lm_labels=lm_labels,)\r\n\r\n loss = outputs[0]\r\n\r\n return loss\r\n```\r\n\r\nBut now, with the following code, my experiments with `facebook/bart-large` are not working:\r\n```\r\n def _step(self, batch: dict) -> Tuple:\r\n pad_token_id = self.tokenizer.pad_token_id\r\n src_ids, src_mask = batch[\"input_ids\"], batch[\"attention_mask\"]\r\n tgt_ids = batch[\"labels\"]\r\n if isinstance(self.model, T5ForConditionalGeneration):\r\n decoder_input_ids = self.model._shift_right(tgt_ids)\r\n else:\r\n decoder_input_ids = shift_tokens_right(tgt_ids, pad_token_id)\r\n if not self.already_saved_batch: # This would be slightly better if it only happened on rank zero\r\n batch[\"decoder_input_ids\"] = decoder_input_ids\r\n self.save_readable_batch(batch)\r\n outputs = self(src_ids, attention_mask=src_mask, decoder_input_ids=decoder_input_ids, use_cache=False)\r\n lm_logits = outputs[0]\r\n if self.hparams.label_smoothing == 0:\r\n # Same behavior as modeling_bart.py, besides ignoring pad_token_id\r\n ce_loss_fct = torch.nn.CrossEntropyLoss(ignore_index=pad_token_id)\r\n\r\n assert lm_logits.shape[-1] == self.vocab_size\r\n loss = ce_loss_fct(lm_logits.view(-1, lm_logits.shape[-1]), tgt_ids.view(-1))\r\n else:\r\n lprobs = torch.nn.functional.log_softmax(lm_logits, dim=-1)\r\n loss, nll_loss = label_smoothed_nll_loss(\r\n lprobs, tgt_ids, self.hparams.label_smoothing, ignore_index=pad_token_id\r\n )\r\n return (loss,)\r\n```\r\n\r\nI'm trying to understand if those things are related. The weird thing is that the exact same code works for `facebook/bart-base`. Please, see #8005. ", "Moved to #8005, problem seems to be config related." ]

[]

[]

[ "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 } ]

[]

[ "For OOM errors, I guess you need to reduce batch_size or parallelize over more GPUs.", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "Thanks Patrick :)\r\n\r\nThank you for sharing this @zolekode !", "@patrickvonplaten thanks for the correction" ]

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

[]

[ "Hi, I'm sorry but I really don't understand what the issue is. Could you clarify?", "I think this comes from the original Albert tokenization, maybe you want to ask on google repository about this?\r\nhttps://github.com/google-research/albert/blob/master/tokenization.py#L67", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "pinging the T5 master @patrickvonplaten ", "Hey @rabeehk - not sure about that. You can check the T5 notebooks we provide or `https://discuss.huggingface.co/`.", "Hi\nI could not find notebooks in the link you said.\nI am looking for a way to train multiple tasks at once in T5, similar to\nthis script:\nhttps://github.com/google-research/text-to-text-transfer-transformer/blob/master/t5/models/hf_model.py\nthanks for your help.\nBest\nRabeeh\n\nOn Thu, Oct 22, 2020 at 10:51 PM Patrick von Platen <\[email protected]> wrote:\n\n> Hey @rabeehk <https://github.com/rabeehk> - not sure about that. You can\n> check the T5 notebooks we provide or https://discuss.huggingface.co/.\n>\n> —\n> You are receiving this because you were mentioned.\n> Reply to this email directly, view it on GitHub\n> <https://github.com/huggingface/transformers/issues/7956#issuecomment-714754836>,\n> or unsubscribe\n> <https://github.com/notifications/unsubscribe-auth/ABP4ZCFCMFQC5USZT3FRAFLSMCLNZANCNFSM4S2MCL7Q>\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" ]