Datasets:

---

DanielPFlorian

/

Transformers-Github-Issues

like 0

[]

[]

[ "Hi! I believe this is a bug that was fixed on master. Could you try and install from source and tell me if it fixes your issue?\r\n\r\nYou can do so with the following command in your python environment:\r\n\r\n```\r\npip install git+https://github.com/huggingface/transformers\r\n```", "@LysandreJik Thanks for the hint. I used a workaround. I installed `transformers` using:\r\n\r\n```\r\nconda install -c conda-forge transformers\r\n```\r\n\r\nIn Python 3.7.4, then I added `DUMMY_INPUTS = [[7, 6, 0, 0, 1], [1, 2, 3, 0, 0], [0, 0, 0, 4, 5]]` after variable `logger` in `/opt/anaconda3/lib/python3.7/site-packages/transformers/modeling_tf_utils.py`, because it was missing.\r\n\r\nIt was interesting, because in GCP, transformers were not showing in `python3` , only in `sudo python3`", "@LysandreJik I updated the install from source but ran into the same issue. The \"solution\" was again as @RubensZimbres mentions: adding `DUMMY_INPUTS = [[7, 6, 0, 0, 1], [1, 2, 3, 0, 0], [0, 0, 0, 4, 5]]` after variable `logger`, however I needed it changed in `/opt/anaconda3/lib/python3.7/site-packages/transformers/modeling_tf_pytorch_utils.py`.\r\n", "I created a Pull Request at: https://github.com/huggingface/transformers/pull/1847", "Can't reproduce on master and the latest release (2.2.1).\r\nFeel free to reopen if the issue is still there.\r\n" ]

[]

[]

[ "Hi, some Transformers have activation in their heads, for instance, Bert.\r\nSee here: https://github.com/huggingface/transformers/blob/master/transformers/modeling_bert.py#L421\r\n\r\nThis is most likely a design choice with a minor effect for deep transformers as they learn to generate the current or next token along several of the output layers.\r\nSee the nice blog post and paper of Lena Voita for some intuition on this: https://lena-voita.github.io/posts/emnlp19_evolution.html", "Thank you! That link was very helpful." ]

[]

[]

[ "Hi, we've added some details on the multi-modal models here: https://huggingface.co/transformers/multilingual.html\r\nAnd an XNLI example here: https://github.com/huggingface/transformers/tree/master/examples#xnli" ]

[]

[]

[ "You're right, this is an error. The PR #1811 aims to fix that issue!", "It should be fixed now, thanks! Feel free to re-open if the error persists." ]

[]

[]

[]

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

[]

[ "Are you in a multi-GPU setup ?", "I did not use multi-GPU setup, I used to `model.cuda()`, it ocuurs \"RuntimeError: CUDA error: device-side assert triggered\", so I changed to `model.cuda(0)`, but the error still occurs.", "Do you mind showing how you initialize BERT and the code surrounding the error?", "My project structure is as follows:\r\n\r\nAnd `dataset.py` is as follows:\r\n```\r\n# -*- coding: utf-8 -*-\r\n\"\"\"\r\n@time: 2019/8/6 19:47\r\n@author: wangjiawei\r\n\"\"\"\r\n\r\n\r\nfrom torch.utils.data import Dataset\r\nimport torch\r\nimport csv\r\n\r\n\r\nclass MyDataset(Dataset):\r\n\r\n def __init__(self, data_path, tokenizer):\r\n super(MyDataset, self).__init__()\r\n\r\n self.tokenizer = tokenizer\r\n texts, labels = [], []\r\n with open(data_path, 'r', encoding='utf-8') as csv_file:\r\n reader = csv.reader(csv_file, quotechar='\"')\r\n for idx, line in enumerate(reader):\r\n text = \"\"\r\n for tx in line[1:]:\r\n text += tx\r\n text += \" \"\r\n text = self.tokenizer.tokenize(text)\r\n if len(text) > 512:\r\n text = text[:512]\r\n text = self.tokenizer.encode(text, add_special_tokens=True)\r\n text_id = torch.tensor(text)\r\n texts.append(text_id)\r\n label = int(line[0]) - 1\r\n labels.append(label)\r\n\r\n self.texts = texts\r\n self.labels = labels\r\n self.num_class = len(set(self.labels))\r\n\r\n def __len__(self):\r\n return len(self.labels)\r\n\r\n def __getitem__(self, item):\r\n label = self.labels[item]\r\n text = self.texts[item]\r\n\r\n return {'text': text, 'label': torch.tensor(label).long()}\r\n```\r\n\r\n`main.py` is as follows:\r\n```\r\n# -*- coding: utf-8 -*-\r\n\"\"\"\r\n@time: 2019/7/17 20:37\r\n@author: wangjiawei\r\n\"\"\"\r\n\r\nimport os\r\nimport torch\r\nimport torch.nn as nn\r\nfrom torch.utils.data import DataLoader, random_split\r\nfrom utils import get_default_folder, my_collate_fn\r\nfrom dataset import MyDataset\r\nfrom model import TextClassify\r\nfrom torch.utils.tensorboard import SummaryWriter\r\nimport argparse\r\nimport shutil\r\nimport numpy as np\r\nimport random\r\nimport time\r\nfrom train import train_model, evaluate\r\nfrom transformers import BertModel, BertTokenizer\r\n\r\nseed_num = 123\r\nrandom.seed(seed_num)\r\ntorch.manual_seed(seed_num)\r\nnp.random.seed(seed_num)\r\n\r\nif __name__ == \"__main__\":\r\n parser = argparse.ArgumentParser(\"self attention for Text Categorization\")\r\n parser.add_argument(\"--batch_size\", type=int, default=64)\r\n parser.add_argument(\"--num_epoches\", type=int, default=20)\r\n parser.add_argument(\"--lr\", type=float, default=0.001)\r\n parser.add_argument(\"--kernel_size\", type=int, default=3)\r\n parser.add_argument(\"--word_dim\", type=int, default=768)\r\n parser.add_argument(\"--out_dim\", type=int, default=300)\r\n parser.add_argument(\"--dropout\", default=0.5)\r\n parser.add_argument(\"--es_patience\", type=int, default=3)\r\n parser.add_argument(\"--dataset\", type=str,\r\n choices=[\"agnews\", \"dbpedia\", \"yelp_review\", \"yelp_review_polarity\", \"amazon_review\",\r\n \"amazon_polarity\", \"sogou_news\", \"yahoo_answers\"], default=\"yelp_review\")\r\n parser.add_argument(\"--log_path\", type=str, default=\"tensorboard/classify\")\r\n\r\n args = parser.parse_args()\r\n\r\n input, output = get_default_folder(args.dataset)\r\n train_path = input + os.sep + \"train.csv\"\r\n\r\n if not os.path.exists(output):\r\n os.makedirs(output)\r\n\r\n # with open(input + os.sep + args.vocab_file, 'rb') as f1:\r\n # vocab = pickle.load(f1)\r\n\r\n # emb_begin = time.time()\r\n # pretrain_word_embedding = build_pretrain_embedding(args.embedding_path, vocab, args.word_dim)\r\n # emb_end = time.time()\r\n # emb_min = (emb_end - emb_begin) % 3600 // 60\r\n # print('build pretrain embed cost {}m'.format(emb_min))\r\n\r\n model_class, tokenizer_class, pretrained_weights = BertModel, BertTokenizer, 'bert-base-uncased'\r\n tokenizer = tokenizer_class.from_pretrained(pretrained_weights)\r\n bert = model_class.from_pretrained(pretrained_weights)\r\n\r\n\r\n train_dev_dataset = MyDataset(input + os.sep + \"train.csv\", tokenizer)\r\n len_train_dev_dataset = len(train_dev_dataset)\r\n dev_size = int(len_train_dev_dataset * 0.1)\r\n train_size = len_train_dev_dataset - dev_size\r\n train_dataset, dev_dataset = random_split(train_dev_dataset, [train_size, dev_size])\r\n test_dataset = MyDataset(input + os.sep + \"test.csv\", tokenizer)\r\n\r\n train_dataloader = DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True, collate_fn=my_collate_fn)\r\n dev_dataloader = DataLoader(dev_dataset, batch_size=args.batch_size, shuffle=False, collate_fn=my_collate_fn)\r\n test_dataloader = DataLoader(test_dataset, batch_size=args.batch_size, shuffle=False, collate_fn=my_collate_fn)\r\n\r\n model = TextClassify(bert, args.kernel_size, args.word_dim, args.out_dim, test_dataset.num_class, args.dropout)\r\n\r\n log_path = \"{}_{}\".format(args.log_path, args.dataset)\r\n if os.path.isdir(log_path):\r\n shutil.rmtree(log_path)\r\n os.makedirs(log_path)\r\n writer = SummaryWriter(log_path)\r\n\r\n if torch.cuda.is_available():\r\n model.cuda(0)\r\n\r\n criterion = nn.CrossEntropyLoss()\r\n optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)\r\n best_acc = -1\r\n early_stop = 0\r\n model.train()\r\n batch_num = -1\r\n\r\n train_begin = time.time()\r\n\r\n for epoch in range(args.num_epoches):\r\n epoch_begin = time.time()\r\n print('train {}/{} epoch'.format(epoch + 1, args.num_epoches))\r\n batch_num = train_model(model, optimizer, criterion, batch_num, train_dataloader, writer)\r\n dev_acc = evaluate(model, dev_dataloader)\r\n writer.add_scalar('dev/Accuracy', dev_acc, epoch)\r\n print('dev_acc:', dev_acc)\r\n if dev_acc > best_acc:\r\n early_stop = 0\r\n best_acc = dev_acc\r\n print('new best_acc on dev set:', best_acc)\r\n torch.save(model.state_dict(), \"{}{}\".format(output, 'best.pt'))\r\n else:\r\n early_stop += 1\r\n\r\n epoch_end = time.time()\r\n cost_time = epoch_end - epoch_begin\r\n print('train {}th epoch cost {}m {}s'.format(epoch + 1, int(cost_time / 60), int(cost_time % 60)))\r\n print()\r\n\r\n # Early stopping\r\n if early_stop > args.es_patience:\r\n print(\r\n \"Stop training at epoch {}. The best dev_acc achieved is {}\".format(epoch - args.es_patience, best_acc))\r\n break\r\n\r\n train_end = time.time()\r\n train_cost = train_end - train_begin\r\n hour = int(train_cost / 3600)\r\n min = int((train_cost % 3600) / 60)\r\n second = int(train_cost % 3600 % 60)\r\n print()\r\n print()\r\n print('train end', '-' * 50)\r\n print('train total cost {}h {}m {}s'.format(hour, min, second))\r\n print('-' * 50)\r\n\r\n model_name = \"{}{}\".format(output, 'best.pt')\r\n\r\n model.load_state_dict(torch.load(model_name))\r\n test_acc = evaluate(model, test_dataloader)\r\n print('test acc on test set:', test_acc)\r\n```\r\n\r\n`model.py` is as follows:\r\n```\r\n# -*- coding: utf-8 -*-\r\n\"\"\"\r\n@time: 2019/8/6 19:04\r\n@author: wangjiawei\r\n\"\"\"\r\n\r\nimport torch.nn as nn\r\nimport torch\r\nimport torch.nn.functional as F\r\n\r\n\r\nclass TextClassify(nn.Module):\r\n def __init__(self, bert, kernel_size, word_dim, out_dim, num_class, dropout=0.5):\r\n super(TextClassify, self).__init__()\r\n self.bert = bert\r\n self.cnn = nn.Sequential(nn.Conv1d(word_dim, out_dim, kernel_size=kernel_size, padding=1), nn.ReLU(inplace=True))\r\n self.drop = nn.Dropout(dropout, inplace=True)\r\n self.fc = nn.Linear(out_dim, num_class)\r\n\r\n def forward(self, word_input):\r\n batch_size = word_input.size(0)\r\n represent = self.bert(word_input)[0]\r\n represent = self.drop(represent)\r\n represent = represent.transpose(1, 2)\r\n contexts = self.cnn(represent)\r\n feature = F.max_pool1d(contexts, contexts.size(2)).contiguous().view(batch_size, -1)\r\n feature = self.drop(feature)\r\n feature = self.fc(feature)\r\n return feature\r\n```\r\n\r\n`train.py` is as follows:\r\n```\r\n# -*- coding: utf-8 -*-\r\n\"\"\"\r\n@time: 2019/8/6 20:14\r\n@author: wangjiawei\r\n\"\"\"\r\n\r\n\r\nimport torch\r\n\r\n\r\ndef train_model(model, optimizer, criterion, batch_num, dataloader, writer):\r\n model.train()\r\n\r\n for batch in dataloader:\r\n model.zero_grad()\r\n batch_num += 1\r\n text = batch['text']\r\n label = batch['label']\r\n if torch.cuda.is_available():\r\n text = text.cuda(0)\r\n label = label.cuda(0)\r\n\r\n feature = model(text)\r\n loss = criterion(feature, label)\r\n writer.add_scalar('Train/Loss', loss, batch_num)\r\n loss.backward()\r\n optimizer.step()\r\n\r\n return batch_num\r\n\r\n\r\ndef evaluate(model, dataloader):\r\n model.eval()\r\n correct_num = 0\r\n total_num = 0\r\n\r\n for batch in dataloader:\r\n text = batch['text']\r\n label = batch['label']\r\n if torch.cuda.is_available():\r\n text = text.cuda(0)\r\n label = label.cuda(0)\r\n\r\n with torch.no_grad():\r\n predictions = model(text)\r\n _, preds = torch.max(predictions, 1)\r\n correct_num += torch.sum((preds == label)).item()\r\n total_num += len(label)\r\n\r\n acc = (correct_num / total_num) * 100\r\n return acc\r\n```\r\n\r\n`utils.py` is as follows:\r\n```\r\n# -*- coding: utf-8 -*-\r\n\"\"\"\r\n@time: 2019/8/6 19:54\r\n@author: wangjiawei\r\n\"\"\"\r\n\r\n\r\nimport csv\r\nimport nltk\r\nimport numpy as np\r\nfrom torch.nn.utils.rnn import pad_sequence\r\nimport torch\r\n\r\n\r\ndef load_pretrain_emb(embedding_path, embedd_dim):\r\n embedd_dict = dict()\r\n with open(embedding_path, 'r', encoding=\"utf8\") as file:\r\n for line in file:\r\n line = line.strip()\r\n if len(line) == 0:\r\n continue\r\n tokens = line.split()\r\n if not embedd_dim + 1 == len(tokens):\r\n continue\r\n embedd = np.empty([1, embedd_dim])\r\n embedd[:] = tokens[1:]\r\n first_col = tokens[0]\r\n embedd_dict[first_col] = embedd\r\n return embedd_dict\r\n\r\n\r\ndef build_pretrain_embedding(embedding_path, vocab, embedd_dim=300):\r\n embedd_dict = dict()\r\n if embedding_path is not None:\r\n embedd_dict = load_pretrain_emb(embedding_path, embedd_dim)\r\n alphabet_size = vocab.size()\r\n scale = 0.1\r\n pretrain_emb = np.empty([vocab.size(), embedd_dim])\r\n perfect_match = 0\r\n case_match = 0\r\n not_match = 0\r\n for word, index in vocab.items():\r\n if word in embedd_dict:\r\n pretrain_emb[index, :] = embedd_dict[word]\r\n perfect_match += 1\r\n elif word.lower() in embedd_dict:\r\n pretrain_emb[index, :] = embedd_dict[word.lower()]\r\n case_match += 1\r\n else:\r\n pretrain_emb[index, :] = np.random.uniform(-scale, scale, [1, embedd_dim])\r\n not_match += 1\r\n\r\n pretrain_emb[0, :] = np.zeros((1, embedd_dim))\r\n pretrained_size = len(embedd_dict)\r\n print('pretrained_size:', pretrained_size)\r\n print(\"Embedding:\\n pretrain word:%s, prefect match:%s, case_match:%s, oov:%s, oov%%:%s\" % (\r\n pretrained_size, perfect_match, case_match, not_match, (not_match + 0.) / alphabet_size))\r\n return pretrain_emb\r\n\r\n\r\ndef get_default_folder(dataset):\r\n if dataset == \"agnews\":\r\n input = \"data/ag_news_csv\"\r\n output = \"output/ag_news/\"\r\n elif dataset == \"dbpedia\":\r\n input = \"data/dbpedia_csv\"\r\n output = \"output/dbpedia/\"\r\n elif dataset == \"yelp_review\":\r\n input = \"data/yelp_review_full_csv\"\r\n output = \"output/yelp_review_full/\"\r\n elif dataset == \"yelp_review_polarity\":\r\n input = \"data/yelp_review_polarity_csv\"\r\n output = \"output/yelp_review_polarity/\"\r\n elif dataset == \"amazon_review\":\r\n input = \"data/amazon_review_full_csv\"\r\n output = \"output/amazon_review_full/\"\r\n elif dataset == \"amazon_polarity\":\r\n input = \"data/amazon_review_polarity_csv\"\r\n output = \"output/amazon_review_polarity/\"\r\n elif dataset == \"sogou_news\":\r\n input = \"data/sogou_news_csv\"\r\n output = \"output/sogou_news/\"\r\n elif dataset == \"yahoo_answers\":\r\n input = \"data/yahoo_answers_csv\"\r\n output = \"output/yahoo_answers/\"\r\n return input, output\r\n\r\n\r\ndef my_collate(batch_tensor, key):\r\n if key == 'text':\r\n batch_tensor = pad_sequence(batch_tensor, batch_first=True, padding_value=0)\r\n else:\r\n batch_tensor = torch.stack(batch_tensor)\r\n return batch_tensor\r\n\r\n\r\ndef my_collate_fn(batch):\r\n return {key: my_collate([d[key] for d in batch], key) for key in batch[0]}\r\n\r\n\r\nclass Vocabulary(object):\r\n def __init__(self, filename):\r\n self._id_to_word = []\r\n self._word_to_id = {}\r\n self._pad = -1\r\n self._unk = -1\r\n self.index = 0\r\n\r\n self._id_to_word.append('<PAD>')\r\n self._word_to_id['<PAD>'] = self.index\r\n self._pad = self.index\r\n self.index += 1\r\n self._id_to_word.append('<UNK>')\r\n self._word_to_id['<UNK>'] = self.index\r\n self._unk = self.index\r\n self.index += 1\r\n\r\n word_num = dict()\r\n\r\n with open(filename, 'r', encoding='utf-8') as f1:\r\n reader = csv.reader(f1, quotechar='\"')\r\n for line in reader:\r\n text = \"\"\r\n for tx in line[1:]:\r\n text += tx\r\n text += \" \"\r\n\r\n text = nltk.word_tokenize(text)\r\n for word in text:\r\n if word not in word_num:\r\n word_num[word] = 0\r\n word_num[word] += 1\r\n\r\n for word, num in word_num.items():\r\n if num >= 3:\r\n self._id_to_word.append(word)\r\n self._word_to_id[word] = self.index\r\n self.index += 1\r\n\r\n def unk(self):\r\n return self._unk\r\n\r\n def pad(self):\r\n return self._pad\r\n\r\n def size(self):\r\n return len(self._id_to_word)\r\n\r\n def word_to_id(self, word):\r\n if word in self._word_to_id:\r\n return self._word_to_id[word]\r\n elif word.lower() in self._word_to_id:\r\n return self._word_to_id[word.lower()]\r\n return self.unk()\r\n\r\n def id_to_word(self, cur_id):\r\n return self._id_to_word[cur_id]\r\n\r\n def items(self):\r\n return self._word_to_id.items()\r\n```\r\n", "when I use Google Cloud, it works. Before I used Google Colab. It is strange.", "It's probably because your token embeddings size (vocab size) doesn't match with pre-trained model. Do `model.resize_token_embeddings(len(tokenizer))` before training. Please check #1848 and #1849 ", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "I also faced the same issue while encoding 20_newspaper dataset, After further investigation, I found there are some sentences which are not in the English language, for example :\r\n\r\n`'subject romanbmp pa of from pwisemansalmonusdedu cliff replyto pwisemansalmonusdedu cliff distribution usa organization university of south dakota lines maxaxaxaxaxaxaxaxaxaxaxaxaxaxax maxaxaxax39f8z51 wwizbhjbhjbhjbhjgiz mgizgizbhjgizm 1tei0lv9f9f9fra 5z46q0475vo4 mu34u34u m34w 084oo aug 0y5180 mc p8v5555555555965hwgv 7uqgy gp452gvbdigiz maxaxaxaxaxaxaxaxaxaxaxaxaxaxax maxaxax34t2php2app12a34u34tm11 6um6 mum8 zbjf0kvlvc9vde5e9e5g9vg9v38vc o3o n5 mardi 24y2 g92li6e0q8axaxaxaxaxaxaxax maxaxaxaxaxaxaxas9ne1whjn 1tei4pmf9l3u3 mr hjpm75u4u34u34u nfyn 46uo m5ug 0y4518hr8y3m15556tdy65c8u 47y m7 hsxgjeuaxaxaxaxaxaxaxaxaxaxaxax maxaxaxaxaxaxaxax34t2pchp2a2a2a234u m34u3 zexf3w21fp4 wm3t f9d uzi0mf8axaxaxaxaxaxrxkrldajj mkjznbbwp0nvmkvmkvnrkmkvmhi axaxaxaxax maxaxaxaxaxaxaxks8vc9vfiue949h g9v38v6un5 mg83q3x w5 3t pi0wsr4c362l zkn2axaxaxaxaxaxaxaxaxaxax maxaxaxaxaxaxaxaxaxaxaxaxaxaxax maxax39f9fpli6e1t1'`\r\n\r\n**This was causing the model to produce this error If you are facing this issue follow these steps:** \r\n\r\n1. Catch Exception and store the sentences in the new list to see if the sentence is having weird characters \r\n2. Reduce the batch size, If your batch size is too high (if you are using batch encoding)\r\n3. Check if your sentence length is too long for the model to encode, trim the sentence length ( try 200 first then 128 )", "I am continuously getting the runtime error: CUDA error: device-side assert triggered, I am new to transformer library.\r\n\r\nI am creating the longformer classifier in the below format:\r\n\r\nmodel = LongformerForSequenceClassification.from_pretrained('allenai/longformer-base-4096', gradient_checkpointing = True, num_labels = 5)\r\n\r\nand tokenizer as :\r\n\r\ntokenizer = LongformerTokenizer.from_pretrained('allenai/longformer-base-4096')\r\n\r\nencoded_data_train = tokenizer.batch_encode_plus(\r\ndf[df.data_type=='train'].content.values,\r\nadd_special_tokens = True,\r\nreturn_attention_mask=True,\r\npadding = True,\r\nmax_length = 3800,\r\nreturn_tensors='pt'\r\n)\r\n\r\nCan anyone please help? I am using latest version of transformer library and using google colab GPU for building my classifier.", "> I also faced the same issue while encoding 20_newspaper dataset, After further investigation, I found there are some sentences which are not in the English language, for example :\r\n> \r\n> `'subject romanbmp pa of from pwisemansalmonusdedu cliff replyto pwisemansalmonusdedu cliff distribution usa organization university of south dakota lines maxaxaxaxaxaxaxaxaxaxaxaxaxaxax maxaxaxax39f8z51 wwizbhjbhjbhjbhjgiz mgizgizbhjgizm 1tei0lv9f9f9fra 5z46q0475vo4 mu34u34u m34w 084oo aug 0y5180 mc p8v5555555555965hwgv 7uqgy gp452gvbdigiz maxaxaxaxaxaxaxaxaxaxaxaxaxaxax maxaxax34t2php2app12a34u34tm11 6um6 mum8 zbjf0kvlvc9vde5e9e5g9vg9v38vc o3o n5 mardi 24y2 g92li6e0q8axaxaxaxaxaxaxax maxaxaxaxaxaxaxas9ne1whjn 1tei4pmf9l3u3 mr hjpm75u4u34u34u nfyn 46uo m5ug 0y4518hr8y3m15556tdy65c8u 47y m7 hsxgjeuaxaxaxaxaxaxaxaxaxaxaxax maxaxaxaxaxaxaxax34t2pchp2a2a2a234u m34u3 zexf3w21fp4 wm3t f9d uzi0mf8axaxaxaxaxaxrxkrldajj mkjznbbwp0nvmkvmkvnrkmkvmhi axaxaxaxax maxaxaxaxaxaxaxks8vc9vfiue949h g9v38v6un5 mg83q3x w5 3t pi0wsr4c362l zkn2axaxaxaxaxaxaxaxaxaxax maxaxaxaxaxaxaxaxaxaxaxaxaxaxax maxax39f9fpli6e1t1'`\r\n> \r\n> **This was causing the model to produce this error If you are facing this issue follow these steps:**\r\n> \r\n> 1. Catch Exception and store the sentences in the new list to see if the sentence is having weird characters\r\n> \r\n> 2. Reduce the batch size, If your batch size is too high (if you are using batch encoding)\r\n> \r\n> 3. Check if your sentence length is too long for the model to encode, trim the sentence length ( try 200 first then 128 )\r\n\r\nWhat a bizzare dataset. \r\n\r\nSo, this error does occur if you use a an english tokenizer on some random garbage input! \r\n\r\n```python3\r\nfrom langdetect import detect\r\ndef detect_robust(x):\r\n try:\r\n out = detect(x) # 20 news managed to break this\r\n except Exception:\r\n out = 'not-en'\r\n return out\r\n\r\nlang_ = train_df['message'].map(detect_robust)\r\ntrain_df = train_df[lang_=='en']\r\n```\r\n^ this is ridiculously slow, but after the cleaning the error is gone!\r\n" ]

[]

[]

[ "Indeed, good catch, thanks @ronakice " ]

[]

[]

[ "This looks good, do you want to add your command and the results you mention in the README of the examples in `examples/README.md`?", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1803?src=pr&el=h1) Report\n> Merging [#1803](https://codecov.io/gh/huggingface/transformers/pull/1803?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/8618bf15d6edc8774cedc0aae021d259d89c91fc?src=pr&el=desc) will **increase** coverage by `1.08%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1803?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1803 +/- ##\n==========================================\n+ Coverage 78.91% 79.99% +1.08% \n==========================================\n Files 131 131 \n Lines 19450 19450 \n==========================================\n+ Hits 15348 15559 +211 \n+ Misses 4102 3891 -211\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1803?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tests/utils.py](https://codecov.io/gh/huggingface/transformers/pull/1803/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL3V0aWxzLnB5) | `91.17% <0%> (-2.95%)` | :arrow_down: |\n| [transformers/pipelines.py](https://codecov.io/gh/huggingface/transformers/pull/1803/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3BpcGVsaW5lcy5weQ==) | `67.94% <0%> (+0.58%)` | :arrow_up: |\n| [transformers/modeling\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1803/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3V0aWxzLnB5) | `94.1% <0%> (+1.07%)` | :arrow_up: |\n| [transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/1803/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX29wZW5haS5weQ==) | `81.51% <0%> (+1.32%)` | :arrow_up: |\n| [transformers/modeling\\_ctrl.py](https://codecov.io/gh/huggingface/transformers/pull/1803/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2N0cmwucHk=) | `96.47% <0%> (+2.2%)` | :arrow_up: |\n| [transformers/modeling\\_xlnet.py](https://codecov.io/gh/huggingface/transformers/pull/1803/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3hsbmV0LnB5) | `74.54% <0%> (+2.32%)` | :arrow_up: |\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/1803/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `72.57% <0%> (+12%)` | :arrow_up: |\n| [transformers/tests/modeling\\_tf\\_common\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1803/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2NvbW1vbl90ZXN0LnB5) | `94.39% <0%> (+17.24%)` | :arrow_up: |\n| [transformers/modeling\\_tf\\_pytorch\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1803/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3B5dG9yY2hfdXRpbHMucHk=) | `90.06% <0%> (+80.79%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1803?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1803?src=pr&el=footer). Last update [8618bf1...8a2be93](https://codecov.io/gh/huggingface/transformers/pull/1803?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "ok, I have updated the readme.", "@importpandas thanks for contributing this fix, works great! My results running this mod:\r\n```\r\n{\r\n \"exact\": 82.07698138633876,\r\n \"f1\": 85.898874470488,\r\n \"total\": 11873,\r\n \"HasAns_exact\": 79.60526315789474,\r\n \"HasAns_f1\": 87.26000954590184,\r\n \"HasAns_total\": 5928,\r\n \"NoAns_exact\": 84.54163162321278,\r\n \"NoAns_f1\": 84.54163162321278,\r\n \"NoAns_total\": 5945,\r\n \"best_exact\": 83.22243746315169,\r\n \"best_exact_thresh\": -11.112004280090332,\r\n \"best_f1\": 86.88541353813282,\r\n \"best_f1_thresh\": -11.112004280090332\r\n}\r\n```\r\nDistributed processing for batch size 48 using gradient accumulation consumes 10970 MiB on each of 2x NVIDIA 1080Ti.\r\n```", "@ahotrod As for your first results, I will suggest you watch my changes on file `run_squad.py` in this pr. It just added a another input to the model and only a few lines. Without this changing, I got the same results with you, which is nearly to 0 on unanswerable questions. I think Distributed processing isn't responsible for that. When it comes to the running script, it doesn't matter since I had hardly changed it.", "@importpandas re-run just completed (36 hours) and results are posted above.\r\nWorks great, thanks again!", "> @importpandas re-run just completed (36 hours) and results are posted above.\r\n> Works great, thanks again!\r\n\r\nokay, pleasure", "LGTM, merging cc @LysandreJik (the `run_squad` master)" ]

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

[]

[ "This looks like an error related to Google SentencePiece and in particular this issue: https://github.com/google/sentencepiece/issues/411", "https://github.com/google/sentencepiece/issues/411#issuecomment-557596691\r\n\r\n```\r\npip install https://github.com/google/sentencepiece/releases/download/v0.1.84/sentencepiece-0.1.84-cp38-cp38-manylinux1_x86_64.whl\r\n```", "I've got problem installing matplotlib in python. While installing these error massage is shown. What to do now?\r\n\r\n\r\nCommand \"C:\\Users\\tawfiq\\PycharmProjects\\untitled4\\venv\\Scripts\\python.exe -u -c \"import setuptools, tokenize;__file__='C:\\\\Users\\\\tawfiq\\\\AppData\\\\Local\\\\Temp\\\\pip-install-yfe_bqlr\\\\matplotlib\\\\setup.py';f=getatt\r\nr(tokenize, 'open', open)(__file__);code=f.read().replace('\\r\\n', '\\n');f.close();exec(compile(code, __file__, 'exec'))\" install --record C:\\Users\\tawfiq\\AppData\\Local\\Temp\\pip-record-vmiaih4n\\install-record.txt -\r\n-single-version-externally-managed --compile --install-headers C:\\Users\\tawfiq\\PycharmProjects\\untitled4\\venv\\include\\site\\python3.8\\matplotlib\" failed with error code 1 in C:\\Users\\tawfiq\\AppData\\Local\\Temp\\pip-i\r\nnstall-yfe_bqlr\\matplotlib\\\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", "This is still an issue on MacOS: https://github.com/google/sentencepiece/issues/411#issuecomment-578509088", "> google/sentencepiece#411 (comment)\r\n\r\nThis error comes up: ``` ERROR: sentencepiece-0.1.84-cp38-cp38-manylinux1_x86_64.whl is not a supported wheel on this platform.```\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 } ]

[]

[ "This problem comes out from multiple GPUs usage. The error you have reported says that you have parameters or the buffers of the model in **two different locations**. \r\nSaid this, it's probably related to #1504 issue. Reading comments in the #1504 issue, i saw that @h-sugi suggests 4 days ago to modify the source code in `run_**.py` like this:\r\n\r\nBEFORE: `device = torch.device(\"cuda\" if torch.cuda.is_available() and not args.no_cuda else \"cpu\")`\r\nAFTER: `device = torch.device(\"cuda:0\" if torch.cuda.is_available() and not args.no_cuda else \"cpu\").`\r\n\r\nMoreover, the person who opened the issue @ahotrod says this: _Have had many successful SQuAD fine-tuning runs on PyTorch 1.2.0 with Pytorch-Transformers 1.2.0, maybe even Transformers 2.0.0, and Apex 0.1. New environment built with the latest versions (Pytorch 1.3.0, Transformers 2.1.1) spawns data parallel related error above_\r\n\r\nPlease, keep us updated on this topic!\r\n\r\n> ## Bug\r\n> Model I am using (Bert, XLNet....): Bert\r\n> \r\n> Language I am using the model on (English, Chinese....): English\r\n> \r\n> The problem arise when using:\r\n> \r\n> * [ ] the official example scripts: (give details) : transformers/examples/run_glue.py\r\n> * [ ] my own modified scripts: (give details)\r\n> \r\n> The tasks I am working on is:\r\n> \r\n> * [ ] an official GLUE/SQUaD task: (give the name) : MRPC\r\n> * [ ] my own task or dataset: (give details)\r\n> \r\n> ## To Reproduce\r\n> Steps to reproduce the behavior:\r\n> \r\n> \r\n> I've tested using\r\n> python -m pytest -sv ./transformers/tests/\r\n> python -m pytest -sv ./examples/\r\n> and it works fine without couple of tesks.\r\n> \r\n> \r\n> after test, i downloaded glue datafile via\r\n> https://gist.github.com/W4ngatang/60c2bdb54d156a41194446737ce03e2e\r\n> and tried run_glue.py\r\n> \r\n> pip install -r ./examples/requirements.txt\r\n> export GLUE_DIR=/path/to/glue\r\n> export TASK_NAME=MRPC\r\n> \r\n> \r\n> python ./examples/run_glue.py \r\n> --model_type bert \r\n> --model_name_or_path bert-base-uncased \r\n> --task_name $TASK_NAME \r\n> --do_train \r\n> --do_eval \r\n> --do_lower_case \r\n> --data_dir $GLUE_DIR/$TASK_NAME \r\n> --max_seq_length 128 \r\n> --per_gpu_eval_batch_size=8 \r\n> --per_gpu_train_batch_size=8 \r\n> --learning_rate 2e-5 \r\n> --num_train_epochs 3.0 \r\n> --output_dir /tmp/$TASK_NAME/\r\n> \r\n> and i got this error.\r\n> \r\n> `11/11/2019 21:10:50 - INFO - __main__ - Total optimization steps = 345 Epoch: 0%| | 0/3 [00:00<?, ?it/sTraceback (most recent call last): | 0/115 [00:00<?, ?it/s] File \"./examples/run_glue.py\", line 552, in <module> main() File \"./examples/run_glue.py\", line 503, in main global_step, tr_loss = train(args, train_dataset, model, tokenizer) File \"./examples/run_glue.py\", line 146, in train outputs = model(**inputs) File \"/home/insublee/anaconda3/envs/py_torch4/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__ result = self.forward(*input, **kwargs) File \"/home/insublee/anaconda3/envs/py_torch4/lib/python3.7/site-packages/torch/nn/parallel/data_parallel.py\", line 146, in forward \"them on device: {}\".format(self.src_device_obj, t.device)) RuntimeError: module must have its parameters and buffers on device cuda:0 (device_ids[0]) but found one of them on device: cuda:3`\r\n> \r\n> ## Environment\r\n> * OS: ubuntu16.04LTS\r\n> * Python version: 3.7.5\r\n> * PyTorch version: 1.2.0\r\n> * PyTorch Transformers version (or branch): 2.1.1\r\n> * Using GPU ? 4-way 2080ti\r\n> * Distributed of parallel setup ? cuda10.0 cudnn 7.6.4\r\n> * Any other relevant information:\r\n> \r\n> ## Additional context\r\n> thank you.", "thanks a lot. it works!!!! :)", "@TheEdoardo93 After the change of `cuda` to `cuda:0`, will we still have multiple GPU usage for the jobs?", "As stated in the official [docs](https://pytorch.org/tutorials/beginner/blitz/data_parallel_tutorial.html), if you use `torch.device('cuda:0')` you will use **only** a GPU. If you want to use **multiple** GPUs, you can easily run your operations on multiple GPUs by making your model run parallelly using DataParallel: `model = nn.DataParallel(model)`\r\n\r\nYou can read more information [here](https://discuss.pytorch.org/t/run-pytorch-on-multiple-gpus/20932/38) and [here](https://pytorch.org/tutorials/beginner/former_torchies/parallelism_tutorial.html).\r\n\r\n> @TheEdoardo93 After the change of `cuda` to `cuda:0`, will we still have multiple GPU usage for the jobs?", "@TheEdoardo93 I tested `run_glue.py` on a multi-gpu machine. Even, after changing `\"cuda\"` to `\"cuda:0\"` in this line https://github.com/huggingface/transformers/blob/b0ee7c7df3d49a819c4d6cef977214bd91f5c075/examples/run_glue.py#L425 , the training job will still use both GPUs relying on `torch.nn.DataParallel`. It means that `torch.nn.DataParallel` is smart enough even if you are defining `torch.device` to be `cuda:0`, if there are several gpus available, it will use all of them. ", "> @TheEdoardo93 I tested `run_glue.py` on a multi-gpu machine. Even, after changing `\"cuda\"` to `\"cuda:0\"` in this line\r\n> \r\n> https://github.com/huggingface/transformers/blob/b0ee7c7df3d49a819c4d6cef977214bd91f5c075/examples/run_glue.py#L425\r\n> \r\n> , the training job will still use both GPUs relying on `torch.nn.DataParallel`. It means that `torch.nn.DataParallel` is smart enough even if you are defining `torch.device` to be `cuda:0`, if there are several gpus available, it will use all of them.\r\n\r\nmanually set 'args.n_gpu = 1' works for me", "Just a simple `os.environ['CUDA_VISIBLE_DEVICES'] = 'GPU_NUM'` at the beginning of the script should work.", "> > @TheEdoardo93 I tested `run_glue.py` on a multi-gpu machine. Even, after changing `\"cuda\"` to `\"cuda:0\"` in this line\r\n> > https://github.com/huggingface/transformers/blob/b0ee7c7df3d49a819c4d6cef977214bd91f5c075/examples/run_glue.py#L425\r\n> > \r\n> > , the training job will still use both GPUs relying on `torch.nn.DataParallel`. It means that `torch.nn.DataParallel` is smart enough even if you are defining `torch.device` to be `cuda:0`, if there are several gpus available, it will use all of them.\r\n> \r\n> manually set 'args.n_gpu = 1' works for me\r\n\r\nbut then you are not able to use more than 1 gpu, right?", "> @TheEdoardo93 I tested `run_glue.py` on a multi-gpu machine. Even, after changing `\"cuda\"` to `\"cuda:0\"` in this line\r\n> \r\n> https://github.com/huggingface/transformers/blob/b0ee7c7df3d49a819c4d6cef977214bd91f5c075/examples/run_glue.py#L425\r\n> \r\n> , the training job will still use both GPUs relying on `torch.nn.DataParallel`. It means that `torch.nn.DataParallel` is smart enough even if you are defining `torch.device` to be `cuda:0`, if there are several gpus available, it will use all of them.\r\n\r\nI have trid but it doesn't work for me.", "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", "duplicate #1799 " ]

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

[]

[ "I haven't yet tried XLM for SQUad but I did try to finetune it on a similar (private) dataset. I managed to make it converge however the F1 is much worse than that of BERT (~0.75 vs ~0.80). XLM training seems to be very learning rate sensitive so you may want to tinker with that a bit.", "Thank you for your answer. \r\nDid you have any changes to the code released in pytorch-transformers when you test on your own dataset? Could you tell me if you did any changes? Or cloud you share your parameter settings when you test on your own data set?\r\nThank you.", "> Thank you for your answer.\r\n> Did you have any changes to the code released in pytorch-transformers when you test on your own dataset? Could you tell me if you did any changes? Or cloud you share your parameter settings when you test on your own data set?\r\n> Thank you.\r\n\r\nI think the only changes I made was adding language embeddings since I was finetuning `xlm-mlm-tlm-xnli15-1024` which requires that.", "> > Thank you for your answer.\r\n> > Did you have any changes to the code released in pytorch-transformers when you test on your own dataset? Could you tell me if you did any changes? Or cloud you share your parameter settings when you test on your own data set?\r\n> > Thank you.\r\n> \r\n> I think the only changes I made was adding language embeddings since I was finetuning `xlm-mlm-tlm-xnli15-1024` which requires that.\r\n\r\nHi, I've tried all the XLM pre-training models and got random results. loss does not converge. can you help me with this problem? thanks @suicao @ZhengWeiH ", "> > Thank you for your answer.\r\n> > Did you have any changes to the code released in pytorch-transformers when you test on your own dataset? Could you tell me if you did any changes? Or cloud you share your parameter settings when you test on your own data set?\r\n> > Thank you.\r\n> \r\n> I think the only changes I made was adding language embeddings since I was finetuning `xlm-mlm-tlm-xnli15-1024` which requires that.\r\n\r\nThanks @suicao !\r\nWhere did you obtain language embeddings? Doesn't pre-trained `XLM` already come with a multilingual BPE embedding dictionary?", "Ok, I think I may have found the issue in this particular model: it requires language IDs to be specified, but the default is `0`, which for `xlm-mlm-tlm-xnli15-1024` is Arabic, not English (which SQuAD is in). English is `4`, based on the `XLMConfig` class.\r\nI think this will require passing an additional input tensor to all training and evaluation batches, that's all `4` `int64`s.", "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 @johnahug, I would recommend checking out how the TFBertFor* models work and trying a similar method to add the desired layers via Keras, for example: https://github.com/huggingface/transformers/blob/155c782a2ccd103cf63ad48a2becd7c76a7d2115/transformers/modeling_tf_bert.py#L834\r\n\r\nI might be able to help out more at a later time, but I'm not sure exactly when just yet. So it might be worthwhile to go ahead and try it out :)", "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 @watkinsm any update on this?", "hey @johnahug have you got what you were looking for? I am also trying to add CNN to my bert embedding but have no idea how to" ]

[]

[]

[]

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

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1796?src=pr&el=h1) Report\n> Merging [#1796](https://codecov.io/gh/huggingface/transformers/pull/1796?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/b5d330d11820f4ac2cc8c909b1a6a77e0cd961e0?src=pr&el=desc) will **decrease** coverage by `0.01%`.\n> The diff coverage is `0%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1796?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1796 +/- ##\n==========================================\n- Coverage 84.03% 84.02% -0.02% \n==========================================\n Files 94 94 \n Lines 14032 14034 +2 \n==========================================\n Hits 11792 11792 \n- Misses 2240 2242 +2\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1796?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_gpt2.py](https://codecov.io/gh/huggingface/transformers/pull/1796/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2dwdDIucHk=) | `83.91% <0%> (-0.54%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1796?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1796?src=pr&el=footer). Last update [b5d330d...60ebaa5](https://codecov.io/gh/huggingface/transformers/pull/1796?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Looks good to me!", "Are you loading from a TF 1.0 or a TF 2.0 model?", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "By any chance, did you open the port number you're using?", "Yes, it is established. What is the master_addr and master_port when submitting a job with single node and multigpu config in GCP?", "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, @kamalravi were you able to solve this issue?? I am facing the same issue right now where my `init_process_group` freezes and doesn't move forward for training. Could you take a look at this and suggest any relevant solutions if you have found them??\r\n\r\nThanks!\r\n\r\nhttps://github.com/pytorch/pytorch/issues/53395#issuecomment-954103393" ]

[]

[]

[ "For 2: It's next word prediction: https://openai.com/blog/better-language-models/", "> For 2: It's next word prediction: https://openai.com/blog/better-language-models/\r\n\r\nThank you for your reply. From my perspective, I believe the next word prediction is what the language model does, which means it is the \"unsupervised\" part in the first head. \r\n\r\nThe evidence is that gpt-2 is based on gpt. Based on figure 1 in gpt paper, the left head is for \"text prediction\"(next word prediction), and the right head is for \"task prediction\".", "Hi, the GPT2DoubleHeadsModel, as defined in [the documentation](https://huggingface.co/transformers/model_doc/gpt2.html#gpt2doubleheadsmodel), is: \"_The GPT2 Model transformer with a language modeling and a multiple-choice classification head on top e.g. for RocStories/SWAG tasks. The two heads are two linear layers. The language modeling head has its weights tied to the input embeddings, the classification head takes as input the input of a specified classification token index in the input sequence)._\".\r\n\r\nYou use a double head model when you want to have two separate losses. In this case the two linear layers compute two different losses: one of the heads is a language modeling loss, with a linear layer of size `hidden_size` x `vocab_size` and the other is a classification loss with a linear layer of size `hidden size` x `number_of_choices`.\r\n\r\nYou can read [this blog post](https://medium.com/huggingface/how-to-build-a-state-of-the-art-conversational-ai-with-transfer-learning-2d818ac26313) by @thomwolf which uses GPT2DoubleHeadsModel with a previous version of this library (pytorch-pretrained-BERT).", "@LysandreJik \r\nHi, thank you for your help!" ]

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

[]

[ "I'm assuming you did multiple runs with ≠ seeds?", "Yes -- I have tried with multiple different seeds.", "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", "> Yes -- I have tried with multiple different seeds.\r\n\r\n@evanweissburg Have you found a solution? I'm also having issues with MNLI data. ", "@anthonyfuller7 Nope. I think one resolution was to not use BERT large.", "I too am facing the same issue with BERT and Albert as well. The model does not converge on fine-tuning and the loss does not decrease even over 5 epochs. Did someone manage to solve this issue?" ]

[]

[]

[ "This is great, thanks @stefan-it!", "This is great, thanks a lot @stefan-it.\r\nI've added your quick benchmark in the readme.", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1792?src=pr&el=h1) Report\n> Merging [#1792](https://codecov.io/gh/huggingface/transformers/pull/1792?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/b5d330d11820f4ac2cc8c909b1a6a77e0cd961e0?src=pr&el=desc) will **increase** coverage by `0.03%`.\n> The diff coverage is `97.29%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1792?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1792 +/- ##\n==========================================\n+ Coverage 84.03% 84.07% +0.03% \n==========================================\n Files 94 94 \n Lines 14032 14069 +37 \n==========================================\n+ Hits 11792 11828 +36 \n- Misses 2240 2241 +1\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1792?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tests/modeling\\_distilbert\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1792/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX2Rpc3RpbGJlcnRfdGVzdC5weQ==) | `99.18% <100%> (+0.08%)` | :arrow_up: |\n| [transformers/modeling\\_distilbert.py](https://codecov.io/gh/huggingface/transformers/pull/1792/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2Rpc3RpbGJlcnQucHk=) | `95.87% <96.15%> (+0.02%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1792?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1792?src=pr&el=footer). Last update [b5d330d...05db5bc](https://codecov.io/gh/huggingface/transformers/pull/1792?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "I just wanted to say thanks for this PR, it was just what I was looking for at the time." ]

[]

[]

[ "This means you're encoding a sequence that is larger than the max sequence the model can handle (which is 512 tokens). This is not an error but a warning; if you pass that sequence to the model it will crash as it cannot handle such a long sequence.\r\n\r\nYou can truncate the sequence: `seq = seq[:512]` or use the `max_length` tokenizer parameter so that it handles it on its own.", "Thank you. I truncate the sequence and it worked. But I use the parameter `max_length` of the method \"encode\" of the class of Tokenizer , it do not works.", "Hi, could you show me how you're using the `max_length` parameter?\r\n\r\nEdit:\r\n\r\nThe recommended way is to call the tokenizer directly instead of using the `encode` method, so the following is the recommended way of handling it:\r\n\r\n```py\r\nfrom transformers import GPT2Tokenizer\r\n\r\ntext = \"This is a sequence\"\r\n\r\ntokenizer = GPT2Tokenizer.from_pretrained(\"gpt2\")\r\nx = tokenizer(text, truncation=True, max_length=2)\r\n\r\nprint(len(x)) # 2\r\n```\r\n\r\n--- \r\n\r\nPrevious answer:\r\n\r\nIf you use it as such it should truncate your sequences:\r\n\r\n```py\r\nfrom transformers import GPT2Tokenizer\r\n\r\ntext = \"This is a sequence\"\r\n\r\ntokenizer = GPT2Tokenizer.from_pretrained(\"gpt2\")\r\nx = tokenizer.encode(text, max_length=2)\r\n\r\nprint(len(x)) # 2\r\n```", "I use `max_length` is as follows:\r\n```\r\nmodel_class, tokenizer_class, pretrained_weights = BertModel, BertTokenizer, 'bert-base-uncased'\r\ntokenizer = tokenizer_class.from_pretrained(pretrained_weights)\r\nmodel = model_class.from_pretrained(pretrained_weights)\r\n\r\ntext = \"After a morning of Thrift Store hunting, a friend and I were thinking of lunch, and he ... \" #this sentence is very long, Its length is more than 512. In order to save space, not all of them are shown here\r\n# text = tokenizer.tokenize(text)\r\n# if len(text) > 512:\r\n# text = text[:512]\r\n#text = \"After a morning of Thrift Store hunting, a friend and I were thinking of lunch\"\r\n\r\ntext = tokenizer.encode(text, add_special_tokens=True, max_length=10)\r\nprint(text)\r\nprint(len(text))\r\n```\r\nIt works. I previously set `max_length` to 512, just output the encoded list, so I didn't notice that the length has changed. But the warning still occurs:\r\n\r\n\r\n", "Glad it works! Indeed, we should do something about this warning, it shouldn't appear when a max length is specified.", "Thank you very much!", "What if I need the sequence length to be longer than 512 (e.g., to retrieve the answer in a QA model)?", "Hi @LukasMut this question might be better suited to Stack Overflow.", "I have the same doubt as @LukasMut . Did you open a Stack Overflow question?", "did you got the solution @LukasMut @paulogaspar ", "Not really. All solutions point to using only the 512 tokens, and choosing what to place in those tokens (for example, picking which part of the text)", "Having the same issue @paulogaspar any update on this? I'm having sequences with more than 512 tokens.", "> Having the same issue @paulogaspar any update on this? I'm having sequences with more than 512 tokens.\r\n\r\nTake a look at my last answer, that's the point I'm at.", "Also dealing with this issue and thought I'd post what's going through my head, correct me if I'm wrong but I think the maximum sequence length is determined when the model is first trained? In which case training a model with a larger sequence length is the solution? And I'm wondering if fine-tuning can be used to increase the sequence length.", "Same question. What to do if text is long?", "That's a research questions guys", "This might help people looking for further details https://github.com/pytorch/fairseq/issues/1685 & https://github.com/google-research/bert/issues/27", "Hi,\r\nThe question i have is almost the same.\r\nBert has some configuration options. As far as i know about transformers, it's not constrain by sequence length at all. \r\nCan I change the config to have more than 512 tokens ?", "Most transformers are unfortunately completely constrained, which is the case for BERT (512 tokens max).\r\n\r\nIf you want to use transformers without being limited to a sequence length, you should take a look at Transformer-XL or XLNet.", "@LysandreJik \r\n\r\nI thought [XLNet](https://arxiv.org/pdf/1906.08237.pdf) has a max length of 512 as well. \r\n\r\n[Transformer-XL](https://arxiv.org/pdf/1906.08237.pdf) is still is a mystery to me because it seems like the length is still 512 for downstream tasks, unlike language modeling (pre-training).\r\n\r\nPlease let me know if my understanding is incorrect.\r\n\r\nThanks!", "XLNet was pre-trained/fine-tuned with a maximum length of 512, indeed. However, the model is not limited to such a length:\r\n\r\n```py\r\nfrom transformers import XLNetLMHeadModel, XLNetTokenizer\r\n\r\ntokenizer = XLNetTokenizer.from_pretrained(\"xlnet-base-cased\")\r\nmodel = XLNetLMHeadModel.from_pretrained(\"xlnet-base-cased\")\r\n\r\nencoded = tokenizer.encode_plus(\"Alright, let's do this\" * 500, return_tensors=\"pt\")\r\nprint(encoded[\"input_ids\"].shape) # torch.Size([1, 3503])\r\nprint(model(**encoded)[0].shape) # torch.Size([1, 3503, 32000])\r\n```\r\n\r\nThe model is not limited to a specific length because it doesn't leverage absolute positional embeddings, instead leveraging the same relative positional embeddings that Transformer-XL used. Please note that since the model isn't trained on larger sequences thant 512, no results are guaranteed on larger sequences, even if the model can still handle them.", "I was going to try this out, but after reading this out few times now, I still have no idea how I'm supposed to truncate the token stream for the pipeline.\r\n\r\nI got some results by combining @cswangjiawei 's advice of running the tokenizer, but it returns a truncated sequence that is slightly longer than the limit I set.\r\n\r\nOtherwise the results are good, although they come out slow and I may have to figure how to activate cuda on py torch.\r\n\r\nUpdate: There is an article that shows how to run the summarizer on large texts, I got it to work with this one: https://www.thepythoncode.com/article/text-summarization-using-huggingface-transformers-python", "> What if I need the sequence length to be longer than 512 (e.g., to retrieve the answer in a QA model)?\r\n\r\nYou can go for BigBird as it takes a input token size of 4096 tokens(but can take upto 16K size)", "Let me help with what I have understood. Correct me if I am wrong. The reason you can't use sequence length more than max_length is because of the positional encoding. Let's have a look at the positional encoding in the original Transformer paper\r\n\r\n\r\nSo the _pos_ in the formula is the index of the words, and they have set 10000 as the scale to cover the usual length of most of the sentences. Now, if you look at the visualization of these functions, you will notice until the _pos_ value is less than 10000 we will get a unique temporal representation of each word. But once it's length is more than 10000 representation won't be unique for each word (e.g. 1st and 10001 will have the same representation). So if max_length = scale (512 as discussed here) and sequence_length > max_length positional encoding will not work.\r\nI didn't check what scale value (you can check it by yourself) BERT uses, but probably this may be the reason.\r\n\r\n.", "> > What if I need the sequence length to be longer than 512 (e.g., to retrieve the answer in a QA model)?\r\n> \r\n> You can go for BigBird as it takes a input token size of 4096 tokens(but can take upto 16K size)\r\n\r\nThe code and weights for BigBird haven't been published yet, am I right?", "> > > What if I need the sequence length to be longer than 512 (e.g., to retrieve the answer in a QA model)?\r\n> > \r\n> > \r\n> > You can go for BigBird as it takes a input token size of 4096 tokens(but can take upto 16K size)\r\n> \r\n> The code and weights for BigBird haven't been published yet, am I right?\r\n\r\nYes and in that case you have Longformers, Reformers which can handle the long sequences.", "My model was pretrained with max_seq_len of 128 and max_posi_embeddings of 512 using the original BERT code release.\r\nI am having the same problem here. I have tried a couple of fixes, but none of them is working for me.\r\n\r\n```\r\nexport MAX_LENGTH=120\r\nexport MODEL=\"./bert-1.5M\"\r\n\r\npython3 preprocess.py ./data/train.txt $MODEL $MAX_LENGTH > train.txt\r\npython3 preprocess.py ./data/dev.txt $MODEL $MAX_LENGTH > dev.txt\r\npython3 preprocess.py ./data/test.txt $MODEL $MAX_LENGTH > test.txt\r\n```\r\n\r\n```export MAX_LENGTH=512 # I have tried 128, 256 \r\n\r\nI am running run_ner_old.py file.\r\n\r\nCan anyone help.", "> I use `max_length` is as follows:\r\n> \r\n> ```\r\n> model_class, tokenizer_class, pretrained_weights = BertModel, BertTokenizer, 'bert-base-uncased'\r\n> tokenizer = tokenizer_class.from_pretrained(pretrained_weights)\r\n> model = model_class.from_pretrained(pretrained_weights)\r\n> \r\n> text = \"After a morning of Thrift Store hunting, a friend and I were thinking of lunch, and he ... \" #this sentence is very long, Its length is more than 512. In order to save space, not all of them are shown here\r\n> # text = tokenizer.tokenize(text)\r\n> # if len(text) > 512:\r\n> # text = text[:512]\r\n> #text = \"After a morning of Thrift Store hunting, a friend and I were thinking of lunch\"\r\n> \r\n> text = tokenizer.encode(text, add_special_tokens=True, max_length=10)\r\n> print(text)\r\n> print(len(text))\r\n> ```\r\n> \r\n> It works. I previously set `max_length` to 512, just output the encoded list, so I didn't notice that the length has changed. But the warning still occurs:\r\n> \r\n\r\nHow to apply this method in csv file i have csv file \"data.csv\" in 2nd column it contains news that to be pass in bert of 512 length ", "I am trying to create an arbitrary length text summarizer using Huggingface; should I just partition the input text to the max model length, summarize each part to, say, half its original length, and repeat this procedure as long as necessary to reach the target length for the whole sequence? \r\n\r\nIt feels to me that this is quite a general problem. Shouldn't this be supported as part of the `pipeline` API itself? (I can do a PR if it's a good fit for the API.)", "Not sure if this is the best approach, but I did something like this and it solves the problem ^\r\n\r\n```python\r\nsummarizer = pipeline(\"summarization\", model=\"facebook/bart-large-cnn\")\r\ndef summarize_text(text: str, max_len: int) -> str:\r\n try:\r\n summary = summarizer(text, max_length=max_len, min_length=10, do_sample=False)\r\n return summary[0][\"summary_text\"]\r\n except IndexError as ex:\r\n logging.warning(\"Sequence length too large for model, cutting text in half and calling again\")\r\n return summarize_text(text=text[:(len(text) // 2)], max_len=max_len//2) + summarize_text(text=text[(len(text) // 2):], max_len=max_len//2)\r\n```" ]

[]

[]

[ "Hi, since `pytorch_pretrained_BERT`, many breaking changes have happened, two of which are causing confusion in your snippet:\r\n\r\n- The order of arguments in the forward call has been slightly changed [(v2.0.0)](https://github.com/huggingface/transformers/releases/tag/v2.0.0)\r\n- The models now always return tuples [(v1.0.0)](https://github.com/huggingface/transformers/releases/tag/v1.0.0).\r\n\r\nThe attention mask and token type ids order has been changed for the forward method to better respect the order of importance, which is important when compiling a model with torchscript.\r\n\r\nUpdate the forward call of your model as such to obtain identical results on both:\r\n\r\n```py\r\nprediction = model(tokens_tensor, token_type_ids=segments_tensors)\r\n```\r\n\r\nTo prevent further breaking changes from affecting your workflow, we recommend using named arguments when calling different methods, like it is done in the aforementioned snippet.", "@LysandreJik thanks for the information .Dose this apply for all the bert models or only for the next sentence prediction alone ?", "I applies to all models." ]

[]

[]

[ "The same error occurs to me too. I write a small check code for BertForMultipleChoice and it works as expected (taken from the [documentation](https://github.com/huggingface/transformers/blob/albert/transformers/modeling_bert.py) - rows 945-951). Here the code I wrote.\r\n```\r\ntokenizer = BertTokenizer.from_pretrained('bert-base-uncased')\r\nmodel = BertForMultipleChoice.from_pretrained('bert-base-uncased', output_hidden_states=True, output_attentions=True)\r\nchoices = [\"Hello, my dog is cute\", \"Hello, my cat is amazing\"]\r\ninput_ids = torch.tensor([tokenizer.encode(s) for s in choices]).unsqueeze(0) # Batch size 1, 2 choices\r\nlabels = torch.tensor(1).unsqueeze(0) # Batch size 1\r\noutputs = model(input_ids, labels=labels)\r\nloss, classification_scores, hidden_states, attentions = outputs\r\n```\r\n\r\n> ## Bug\r\n> Model I am using (BertForMultipleChoice):\r\n> \r\n> Language I am using the model on (English.):\r\n> \r\n> The problem arise when using:\r\n> \r\n> * [x] the official example scripts:\r\n> Arises when running through the last piece of example code found here:\r\n> https://github.com/huggingface/transformers#quick-tour\r\n> * [ ] my own modified scripts: (give details)\r\n> \r\n> The tasks I am working on is:\r\n> \r\n> * [ ] an official GLUE/SQUaD task: (give the name)\r\n> * [ ] my own task or dataset: (give details)\r\n> \r\n> ## To Reproduce\r\n> Steps to reproduce the behavior:\r\n> \r\n> 1. Run the sample code from the last loop of the Quick Tour\r\n> 2. Need to supply directory names, see my code (below) for how I did that...\r\n> 3. When BertForMultipleChoice runs the line `reshaped_logits = logits.view(-1, num_choices)` in modeling_bert.py we get a runtime error `RuntimeError: shape '[-1, 16]' is invalid for input of size 1`\r\n> \r\n> ```\r\n> # Each architecture is provided with several class for fine-tuning on down-stream tasks, e.g.\r\n> BERT_MODEL_CLASSES = [BertModel, BertForPreTraining, BertForMaskedLM, BertForNextSentencePrediction,\r\n> BertForSequenceClassification, BertForMultipleChoice, BertForTokenClassification,\r\n> BertForQuestionAnswering]\r\n> ```\r\n> \r\n> ```\r\n> # All the classes for an architecture can be initiated from pretrained weights for this architecture\r\n> # Note that additional weights added for fine-tuning are only initialized\r\n> # and need to be trained on the down-stream task\r\n> pretrained_weights = 'bert-base-uncased'\r\n> tokenizer = BertTokenizer.from_pretrained(pretrained_weights)\r\n> for model_class in BERT_MODEL_CLASSES:\r\n> \r\n> print(\"Processing\", model_class.__name__, \"...\")\r\n> \r\n> # Store class name as target directory\r\n> model_dir_name = model_class.__name__+\"/\"\r\n> \r\n> # Load pretrained model/tokenizer\r\n> model = model_class.from_pretrained(pretrained_weights)\r\n> \r\n> # Models can return full list of hidden-states & attentions weights at each layer\r\n> model = model_class.from_pretrained(pretrained_weights,\r\n> output_hidden_states=True,\r\n> output_attentions=True)\r\n> input_ids = torch.tensor([tokenizer.encode(\"Let's see all hidden-states and attentions on this text\")])\r\n> all_hidden_states, all_attentions = model(input_ids)[-2:]\r\n> \r\n> # Models are compatible with Torchscript\r\n> model = model_class.from_pretrained(pretrained_weights, torchscript=True)\r\n> traced_model = torch.jit.trace(model, (input_ids,))\r\n> \r\n> save_directory = 'BERT_test/'+ model_dir_name\r\n> if not os.path.isdir(save_directory):\r\n> os.mkdir(save_directory)\r\n> \r\n> # Simple serialization for models and tokenizers\r\n> model.save_pretrained(save_directory) # save\r\n> model = model_class.from_pretrained(save_directory) # re-load\r\n> tokenizer.save_pretrained(save_directory) # save\r\n> tokenizer = BertTokenizer.from_pretrained(save_directory) # re-load\r\n> \r\n> # SOTA examples for GLUE, SQUAD, text generation...\r\n> ```\r\n> \r\n> The error:\r\n> \r\n> ```\r\n> I1111 20:47:30.383128 21676 modeling_utils.py:383] loading weights file https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-pytorch_model.bin from cache at C:\\Users\\User\\.cache\\torch\\transformers\\aa1ef1aede4482d0dbcd4d52baad8ae300e60902e88fcb0bebdec09afd232066.36ca03ab34a1a5d5fa7bc3d03d55c4fa650fed07220e2eeebc06ce58d0e9a157\r\n> I1111 20:47:33.363116 21676 modeling_utils.py:453] Weights of BertForMultipleChoice not initialized from pretrained model: ['classifier.weight', 'classifier.bias']\r\n> I1111 20:47:33.365122 21676 modeling_utils.py:456] Weights from pretrained model not used in BertForMultipleChoice: ['cls.predictions.bias', 'cls.predictions.transform.dense.weight', 'cls.predictions.transform.dense.bias', 'cls.predictions.decoder.weight', 'cls.seq_relationship.weight', 'cls.seq_relationship.bias', 'cls.predictions.transform.LayerNorm.weight', 'cls.predictions.transform.LayerNorm.bias']\r\n> ---------------------------------------------------------------------------\r\n> RuntimeError Traceback (most recent call last)\r\n> <ipython-input-10-da6597bd484d> in <module>()\r\n> 19 output_attentions=True)\r\n> 20 input_ids = torch.tensor([tokenizer.encode(\"Let's see all hidden-states and attentions on this text\")])\r\n> ---> 21 all_hidden_states, all_attentions = model(input_ids)[-2:]\r\n> 22 \r\n> 23 # Models are compatible with Torchscript\r\n> \r\n> G:\\Anaconda3\\envs\\pytorch1\\lib\\site-packages\\torch\\nn\\modules\\module.py in __call__(self, *input, **kwargs)\r\n> 539 result = self._slow_forward(*input, **kwargs)\r\n> 540 else:\r\n> --> 541 result = self.forward(*input, **kwargs)\r\n> 542 for hook in self._forward_hooks.values():\r\n> 543 hook_result = hook(self, input, result)\r\n> \r\n> g:\\deeplearning\\huggingface\\transformers\\transformers\\modeling_bert.py in forward(self, input_ids, attention_mask, token_type_ids, position_ids, head_mask, inputs_embeds, labels)\r\n> 1096 pooled_output = self.dropout(pooled_output)\r\n> 1097 logits = self.classifier(pooled_output)\r\n> -> 1098 reshaped_logits = logits.view(-1, num_choices)\r\n> 1099 \r\n> 1100 outputs = (reshaped_logits,) + outputs[2:] # add hidden states and attention if they are here\r\n> \r\n> RuntimeError: shape '[-1, 16]' is invalid for input of size 1\r\n> ```\r\n> \r\n> ## Expected behavior\r\n> Should run without error!\r\n> \r\n> ## Environment\r\n> * OS: Windows 10\r\n> * Python version: 3.6.6\r\n> * PyTorch version: 1.3.0\r\n> * PyTorch Transformers version (or branch): 2.1.1\r\n> * Using GPU ? Yes\r\n> * Distributed of parallel setup ? No\r\n> * Any other relevant information:\r\n> \r\n> ## Additional context", "Thanks! This works for me too. But since the Quick Tour is an example on the github repo I believe it would be a good idea to fix it there... Is there anything you can see wrong with the code?", "Re-opened as I believe we should have the sample code fixed - *and* like [this issue](https://github.com/huggingface/transformers/issues/1787) it should work with Pytorch >= 1.0.0 or at least Pytorch >= 1.2.0 not just with >= 1.3.0", "Indeed thanks, it's fixed now on master", "Thanks - I note that it's been fixed by removing the use of BertForMultipleChoice in the evaluation section!" ]

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

[]

[ "As explained in #1790, you're passing the `token_type_ids` as the attention mask. Change the model forward pass as such:\r\n\r\n```py\r\nprediction = BertNSP(tokens_tensor, token_type_ids=segments_tensors)\r\n```\r\nYour results will be more accurate:\r\n```py\r\ntensor([[-2.3808, 5.4018]], grad_fn=<AddmmBackward>)\r\ntensor([[4.1673e-04, 9.9958e-01]], grad_fn=<SoftmaxBackward>)\r\n```", "@LysandreJik thanks for the information .Dose this apply for all the bert models or only for the next sentence prediction alone ?", "It would be better to use named arguments in all the models. They are bound to change with breaking changes as new versions come up.\r\n\r\nI recommend specifying the arguments' names when possible.", "> As explained in #1790, you're passing the `token_type_ids` as the attention mask. Change the model forward pass as such:\r\n> \r\n> ```python\r\n> prediction = BertNSP(tokens_tensor, token_type_ids=segments_tensors)\r\n> ```\r\n> \r\n> Your results will be more accurate:\r\n> \r\n> ```python\r\n> tensor([[-2.3808, 5.4018]], grad_fn=<AddmmBackward>)\r\n> tensor([[4.1673e-04, 9.9958e-01]], grad_fn=<SoftmaxBackward>)\r\n> ```\r\n\r\nHi,\r\n@LysandreJik Can you explain me what are the scores for? \r\nWhat is the 4.1673e-04 used for and what is the 9.9958e-01 used for?\r\nWhich one of them says that: X% sure that A sentence is followed by B sentence ? And what is the other for?\r\n\r\nThanks!", "I'm having the same problem, and I think I've followed all the directions mentioned above.\r\n\r\ndef line_continues(model, tokenizer, line1, line2):\r\n line1 = [tokenizer.cls_token] + tokenizer.tokenize(line1) + [tokenizer.sep_token]\r\n line2 = tokenizer.tokenize(line2) + [tokenizer.sep_token]\r\n input_idx = tokenizer.convert_tokens_to_ids(line1 + line2)\r\n segment_idx = [0]*len(line1) + [1]*len(line2)\r\n tokens_tensor = torch.tensor([input_idx])\r\n segment_tensor = torch.tensor([segment_idx])\r\n predictions = model(tokens_tensor, token_type_ids=segment_tensor)\r\n probs = F.softmax(predictions[0], dim=1)\r\n return probs[0][0]\r\n\r\nmodel = BertForNextSentencePrediction.from_pretrained('bert-base-cased')\r\nmodel.eval()\r\n\r\n\r\nrandom sentences:\r\n\r\nline1='these articles tell us about where leadership communication is going and where it'\r\nline2='issues gave us the chance to engage with many well-established and emerging experts'\r\nprob = line_continues(model, tokenizer, line1, line2)\r\n0.9993\r\n\r\n\r\ncontiguous sentences:\r\n\r\nline1='these articles tell us about where leadership communication is going and where it'\r\nline2='needs to go in addition to using the model.'\r\nprob = line_continues(model, tokenizer, line1, line2)\r\n0.9991\r\n\r\n\r\nThanks!", "I am also experiencing this kind of issue. After experimenting with some sequence pairs, I think the relation between two sequence should be zero (and also non-sensical). For example:\r\n```\r\nSent1: Paris is the capital of France.\r\nSent2: Cow is a domestic animal. \r\n\r\n```\r\nThis level of stupid sequence.\r\nif there is any slightest connection on word level, it outputs that it is 90% sure that two sequence is coherent. \r\nIn your example, may be `leadership` and `experts` are two near words in semantic space. \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", "This is still a problem and should be re-opened. From the documentation: https://huggingface.co/transformers/model_doc/bert.html#bertfornextsentenceprediction\r\n\r\n```\r\n# documentation example - good\r\nIn Italy, pizza served in formal settings, such as at a restaurant, is presented unsliced.\r\nThe sky is blue due to the shorter wavelength of blue light.\r\nlogits[0, 0]: -3.072946548461914, logits[0, 1]: 5.905644416809082, is random: True\r\n\r\n# my own example - ???\r\nI took my money to the bank on 23rd street\r\nMy monkey was cake and cockroaches have radiation\r\nlogits[0, 0]: 3.0128183364868164, logits[0, 1]: -1.984398365020752, is random: False\r\n```\r\n\r\nI'm not sure how to interpret this.", "I think that it would also depend on train data on which the BERT model was trained on (from the paper \" For the pre-training corpus we use the BooksCorpus (800M words) (Zhu et al., 2015) and English Wikipedia (2,500M words).\"[1]. \r\n\r\n[1] - https://arxiv.org/pdf/1810.04805.pdf " ]

[]

[]

[ "> ## Bug\r\n> Model I am using (CTRLModel):\r\n> \r\n> Language I am using the model on (English):\r\n> \r\n> The problem arise when using:\r\n> \r\n> * [X ] the official example scripts: (give details)\r\n> \r\n> The tasks I am working on is:\r\n> \r\n> * [X ] the Quick Tour for transformers\r\n> \r\n> ## To Reproduce\r\n> Steps to reproduce the behavior:\r\n> \r\n> 1. run the quick tour as found here https://github.com/huggingface/transformers#quick-tour\r\n> 2. in the loop `for model_class, tokenizer_class, pretrained_weights in MODELS` it will crash out when processing the CTRLModel\r\n> \r\n> ```\r\n> ---------------------------------------------------------------------------\r\n> OSError Traceback (most recent call last)\r\n> <ipython-input-5-25349833af91> in <module>\r\n> 5 \r\n> 6 tokenizer = tokenizer_class.from_pretrained(pretrained_weights)\r\n> ----> 7 model = model_class.from_pretrained(pretrained_weights)\r\n> 8 \r\n> 9 # Encode text\r\n> \r\n> g:\\deeplearning\\huggingface\\transformers\\transformers\\modeling_utils.py in from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)\r\n> 389 \r\n> 390 if state_dict is None and not from_tf:\r\n> --> 391 state_dict = torch.load(resolved_archive_file, map_location='cpu')\r\n> 392 \r\n> 393 missing_keys = []\r\n> \r\n> G:\\Anaconda3\\envs\\fastai\\lib\\site-packages\\torch\\serialization.py in load(f, map_location, pickle_module, **pickle_load_args)\r\n> 384 f = f.open('rb')\r\n> 385 try:\r\n> --> 386 return _load(f, map_location, pickle_module, **pickle_load_args)\r\n> 387 finally:\r\n> 388 if new_fd:\r\n> \r\n> G:\\Anaconda3\\envs\\fastai\\lib\\site-packages\\torch\\serialization.py in _load(f, map_location, pickle_module, **pickle_load_args)\r\n> 578 for key in deserialized_storage_keys:\r\n> 579 assert key in deserialized_objects\r\n> --> 580 deserialized_objects[key]._set_from_file(f, offset, f_should_read_directly)\r\n> 581 if offset is not None:\r\n> 582 offset = f.tell()\r\n> \r\n> OSError: [Errno 22] Invalid argument\r\n> ```\r\n> \r\n> ## What should have happened\r\n> Should have run to completion...\r\n> \r\n> ## Environment\r\n> * OS:\r\n> * Python version: 3.3.9\r\n> * PyTorch version: 1.2.0+cu92\r\n> * PyTorch Transformers version (or branch): 2.1.1\r\n> * Using GPU ? Yes\r\n> * Distributed of parallel setup ? no\r\n> * Any other relevant information:\r\n> \r\n> ## Additional context\r\n\r\nCan you test the same code with Python version >= 3.5 and with PyTorch version 1.3.0?", "Sure, l'll get back to you... but I realize that I had a typo with my Python, its 3.6.9, which I've corrected. \r\nAdditionally, I had compiled this with `pip install -e .` from a fresh clone of the repo yesterday", "OK, under Pytorch 1.3.0 it seems good - I got the following output (which I wasn't getting from 1.2.0):\r\n```\r\nI1111 20:26:06.461704 21676 tokenization_utils.py:374] loading file https://raw.githubusercontent.com/salesforce/ctrl/master/ctrl-vocab.json from cache at C:\\Users\\User\\.cache\\torch\\transformers\\a858ad854d3847b02da3aac63555142de6a05f2a26d928bb49e881970514e186.285c96a541cf6719677cfb634929022b56b76a0c9a540186ba3d8bbdf02bca42\r\nI1111 20:26:06.464703 21676 tokenization_utils.py:374] loading file https://raw.githubusercontent.com/salesforce/ctrl/master/ctrl-merges.txt from cache at C:\\Users\\User\\.cache\\torch\\transformers\\aa2c569e6648690484ade28535a8157aa415f15202e84a62e82cc36ea0c20fa9.26153bf569b71aaf15ae54be4c1b9254dbeff58ca6fc3e29468c4eed078ac142\r\nI1111 20:26:07.624670 21676 configuration_utils.py:152] loading configuration file https://storage.googleapis.com/sf-ctrl/pytorch/ctrl-config.json from cache at C:\\Users\\User\\.cache\\torch\\transformers\\d6492ca334c2a4e079f43df30956acf935134081b2b3844dc97457be69b623d0.1ebc47eb44e70492e0c20494a084f108332d20fea7fe5ad408ef5e7a8f2baef4\r\nI1111 20:26:07.628657 21676 configuration_utils.py:169] Model config {\r\n \"attn_pdrop\": 0.1,\r\n \"dff\": 8192,\r\n \"embd_pdrop\": 0.1,\r\n \"finetuning_task\": null,\r\n \"from_tf\": false,\r\n \"initializer_range\": 0.02,\r\n \"is_decoder\": false,\r\n \"layer_norm_epsilon\": 1e-06,\r\n \"n_ctx\": 512,\r\n \"n_embd\": 1280,\r\n \"n_head\": 16,\r\n \"n_layer\": 48,\r\n \"n_positions\": 50000,\r\n \"num_labels\": 1,\r\n \"output_attentions\": false,\r\n \"output_hidden_states\": false,\r\n \"output_past\": true,\r\n \"pruned_heads\": {},\r\n \"resid_pdrop\": 0.1,\r\n \"summary_activation\": null,\r\n \"summary_first_dropout\": 0.1,\r\n \"summary_proj_to_labels\": true,\r\n \"summary_type\": \"cls_index\",\r\n \"summary_use_proj\": true,\r\n \"torchscript\": false,\r\n \"use_bfloat16\": false,\r\n \"vocab_size\": 246534\r\n}\r\n\r\nI1111 20:26:08.047282 21676 modeling_utils.py:383] loading weights file https://storage.googleapis.com/sf-ctrl/pytorch/seqlen256_v1.bin from cache at C:\\Users\\User\\.cache\\torch\\transformers\\c146cc96724f27295a0c3ada1fbb3632074adf87e9aef8269e44c9208787f8c8.b986347cbab65fa276683efbb9c2f7ee22552277bcf6e1f1166557ed0852fdf0\r\nW1111 20:31:14.310047 21676 tokenization_utils.py:936] This tokenizer does not make use of special tokens. Input is returned with no modification.\r\nW1111 20:31:14.650039 21676 tokenization_utils.py:936] This tokenizer does not make use of special tokens. Input is returned with no modification.\r\nW1111 20:31:14.653039 21676 tokenization_utils.py:923] This tokenizer does not make use of special tokens.\r\n```", "[Here](https://colab.research.google.com/drive/1d23JwkEkS_s6XUY2_Me_MHnOrTHmNrqR#scrollTo=K4-yYInLt6YP) you can view a Google Colab written by me and **it works as expected**. I think you can close this issue.\r\n\r\n> OK, under Pytorch 1.3.0 it seems good - I got the following output (which I wasn't getting from 1.2.0):\r\n> \r\n> ```\r\n> I1111 20:26:06.461704 21676 tokenization_utils.py:374] loading file https://raw.githubusercontent.com/salesforce/ctrl/master/ctrl-vocab.json from cache at C:\\Users\\User\\.cache\\torch\\transformers\\a858ad854d3847b02da3aac63555142de6a05f2a26d928bb49e881970514e186.285c96a541cf6719677cfb634929022b56b76a0c9a540186ba3d8bbdf02bca42\r\n> I1111 20:26:06.464703 21676 tokenization_utils.py:374] loading file https://raw.githubusercontent.com/salesforce/ctrl/master/ctrl-merges.txt from cache at C:\\Users\\User\\.cache\\torch\\transformers\\aa2c569e6648690484ade28535a8157aa415f15202e84a62e82cc36ea0c20fa9.26153bf569b71aaf15ae54be4c1b9254dbeff58ca6fc3e29468c4eed078ac142\r\n> I1111 20:26:07.624670 21676 configuration_utils.py:152] loading configuration file https://storage.googleapis.com/sf-ctrl/pytorch/ctrl-config.json from cache at C:\\Users\\User\\.cache\\torch\\transformers\\d6492ca334c2a4e079f43df30956acf935134081b2b3844dc97457be69b623d0.1ebc47eb44e70492e0c20494a084f108332d20fea7fe5ad408ef5e7a8f2baef4\r\n> I1111 20:26:07.628657 21676 configuration_utils.py:169] Model config {\r\n> \"attn_pdrop\": 0.1,\r\n> \"dff\": 8192,\r\n> \"embd_pdrop\": 0.1,\r\n> \"finetuning_task\": null,\r\n> \"from_tf\": false,\r\n> \"initializer_range\": 0.02,\r\n> \"is_decoder\": false,\r\n> \"layer_norm_epsilon\": 1e-06,\r\n> \"n_ctx\": 512,\r\n> \"n_embd\": 1280,\r\n> \"n_head\": 16,\r\n> \"n_layer\": 48,\r\n> \"n_positions\": 50000,\r\n> \"num_labels\": 1,\r\n> \"output_attentions\": false,\r\n> \"output_hidden_states\": false,\r\n> \"output_past\": true,\r\n> \"pruned_heads\": {},\r\n> \"resid_pdrop\": 0.1,\r\n> \"summary_activation\": null,\r\n> \"summary_first_dropout\": 0.1,\r\n> \"summary_proj_to_labels\": true,\r\n> \"summary_type\": \"cls_index\",\r\n> \"summary_use_proj\": true,\r\n> \"torchscript\": false,\r\n> \"use_bfloat16\": false,\r\n> \"vocab_size\": 246534\r\n> }\r\n> \r\n> I1111 20:26:08.047282 21676 modeling_utils.py:383] loading weights file https://storage.googleapis.com/sf-ctrl/pytorch/seqlen256_v1.bin from cache at C:\\Users\\User\\.cache\\torch\\transformers\\c146cc96724f27295a0c3ada1fbb3632074adf87e9aef8269e44c9208787f8c8.b986347cbab65fa276683efbb9c2f7ee22552277bcf6e1f1166557ed0852fdf0\r\n> W1111 20:31:14.310047 21676 tokenization_utils.py:936] This tokenizer does not make use of special tokens. Input is returned with no modification.\r\n> W1111 20:31:14.650039 21676 tokenization_utils.py:936] This tokenizer does not make use of special tokens. Input is returned with no modification.\r\n> W1111 20:31:14.653039 21676 tokenization_utils.py:923] This tokenizer does not make use of special tokens.\r\n> ```", "Is there any way to get this working correctly under Pytorch 1.2? I have some work I need to continue with in that version. It doesn't matter too much as I wasn't intending to use that model, but I guess it might matter to someone else...", "Thanks very much for the Colab notebook - yes it works as expected there. I am now having an issue with running the last part of the Quick Tour, but I'll post that separately, and close this one!", "I'm very honest with you: I don't have the exact answer to your notable question. I'm aware of the fact that PyTorch v1.3.0 has got many changes and improvements from PyTorch v1.2, as stated [here](url). Some changes have been made on the `torch.load()` method too.\r\n\r\n> Is there any way to get this working correctly under Pytorch 1.2? I have some work I need to continue with in that version. It doesn't matter too much as I wasn't intending to use that model, but I guess it might matter to someone else...", "@ChrisPalmerNZ Would you care to re-open this issue?\r\nThere really should not be a bug with the intro sample code.", "OK, re-opened - interesting to see what fixes this for Pytorch 1.2 - but I felt that @TheEdoardo93 was indicating that this will only work under Pytorch >= 1.3.0 ", "Hi, you seem to be having problems when loading the CTRLModel in memory. Usually problems like this can occur if behind a firewall or if the file was corrupted. Could you try to load the model outside of the script, for example in a python console? The following lines should work:\r\n\r\n```py\r\nfrom transformers import CTRLModel\r\n\r\nmodel = CTRLModel.from_pretrained(\"ctrl\", force_download=True)\r\n```\r\n\r\nThe `force_download` option will re-download the file, this will confirm that the file being corrupted is not an issue.\r\n\r\n", "Here the results:\r\n```\r\nPython 3.6.9 |Anaconda, Inc.| (default, Jul 30 2019, 19:07:31) \r\n[GCC 7.3.0] on linux\r\nType \"help\", \"copyright\", \"credits\" or \"license\" for more information.\r\n>>> import platform\r\n>>> platform.platform()\r\n'Linux-4.15.0-66-generic-x86_64-with-debian-buster-sid'\r\n>>> platform.python_version()\r\n'3.6.9'\r\n>>> import transformers\r\n>>> transformers.__version__\r\n'2.1.1'\r\n>>> import torch\r\n>>> torch.__version__\r\n'1.2.0'\r\n>>> from transformers import CTRLModel\r\n>>> model = CTRLModel.from_pretrained('ctrl', force_download=True)\r\n100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 611/611 [00:00<00:00, 259358.34B/s]\r\n100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 6552025106/6552025106 [09:45<00:00, 11197418.01B/s]\r\n>>> CTRLModel\r\n<class 'transformers.modeling_ctrl.CTRLModel'>\r\n>>> \r\n```\r\n\r\n> Hi, you seem to be having problems when loading the CTRLModel in memory. Usually problems like this can occur if behind a firewall or if the file was corrupted. Could you try to load the model outside of the script, for example in a python console? The following lines should work:\r\n> \r\n> ```python\r\n> from transformers import CTRLModel\r\n> \r\n> model = CTRLModel.from_pretrained(\"ctrl\", force_download=True)\r\n> ```\r\n> \r\n> The `force_download` option will re-download the file, this will confirm that the file being corrupted is not an issue." ]

[]

[]

[ "Masked language modeling is an example of autoencoding language modeling , typically mask one or more of words in a sentence and have the model predict those masked words given the other words in sentence. When you changed the vocab size, the number of word to predic from the model also changed.\r\nThis unofficial methods leaved the output embedding and prediction bias unchanged while changed the vocab size, and caused these errors . \r\nI think the output embeddings should be copied from input embeddings, but the bias needs you to tune-up manually.\r\n\r\n------ \r\nFor now, these two lines may make you avoid the error\r\n```\r\nmodel_dict['cls.predictions.bias'] = model_dict['cls.predictions.bias'][:2]\r\nmodel_dict['cls.predictions.decoder.weight'] = model_dict['cls.predictions.decoder.weight'][:2]\r\n\r\n\r\n```", "thank!" ]

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

[]

[ "Hello,\r\n\r\nSorry, the web app is not open source right now.", "Thanks @julien-c . Can you comment on how elastic inference with pytorch models is done?", "Is there any chance this could be reconsidered? I'd love to use it for experimenting with custom fine-tuned models.", "Would also love for this for experimenting with custom models.\r\n", "It'd be very interesting to open source this.\r\n\r\nIs there any intention of doing so in the future?", "Is there any possibility of this becoming reconsidered? I've been trying to look for this as well... :/", "Any chance of this being public? @thomwolf ", "> Would also love for this for experimenting with custom models.\r\n\r\nSame!", "the microsoft/fastformers repo says \"Write With Transformer, built by the Hugging Face team at transformer.huggingface.co, is the official demo of this repo’s text generation capabilities\" according to this [readme](https://github.com/microsoft/fastformers/blob/main/README_transformers.md#online-demo). Last commit October 2020\r\n\r\n" ]

[]

[]

[ "Indeed, this is a documentation error! Thank you for letting us know!", "@LysandreJik This was reintroduced in https://github.com/huggingface/transformers/pull/2989 apparently", "Thanks @Evpok for letting me know, this flew under my radar.", "My pleasure 😄" ]

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

[]

[ "You should check out BLEU and ROUGE scores. These are often referred to as the precision and recall of NLP.\r\n\r\nYou may be able to craft a pseudo-f1 score out of these as well \r\n`f1 = 2*(bleu*rouge)/(bleu + rouge)`\r\n\r\nI haven't tried this myself yet, but hopefully this helps!\r\n\r\n[https://en.wikipedia.org/wiki/BLEU](https://en.wikipedia.org/wiki/BLEU)\r\n[https://en.wikipedia.org/wiki/ROUGE_(metric)](https://en.wikipedia.org/wiki/ROUGE_(metric))", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "What string was passed under `args.model_path`? Is it a directory or a file name? (It should be a directory)", "model_path is a model trained with xxx.pth format.\r\n\r\nIs that a directory? contains what？", "The `model_path` should link to a directory holding the pytorch model and the configuration file associated with it, as if it was saved with `model.save_pretrained(directory)`.\r\n\r\nWhat is your `xxx.pth` format, how did you obtain such a file?", "Oh, no, I just got others trained xx.pth model, no config file.\r\n\r\nWhy a pytorch model need additional config.json? And people really don't know what's the format inside that file at all..... no convinient", "The Pytorch models require configuration files because the architectures in the library can have very different sizes. For example GPT-2 small, medium, large and XL all share the same model but have different sizes. Configuration files are essential for this purpose.\r\n\r\nThe format is defined by the `save_pretrained` method on the models. Models acquired from other sources than our library may not be loadable, or may need to be converted using one of our conversion scripts.\r\n\r\nPlease read the [quickstart portion of the documentation](https://huggingface.co/transformers/quickstart.html) to get a deeper understanding of the way the library works." ]

[]

[]

[ "Hi, BERT uses the NSP task as well as the MLM task during its pre-training phase. The `run_lm_finetuning.py` script only does MLM so you would need to modify it to pre-train BERT the same way it was done in the paper." ]

[ { "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 provide information about your setup? Which TensorFlow version are you using, which python version, which Transformers version? Thank you.", "> Hi, could you provide information about your setup? Which TensorFlow version are you using, which python version, which Transformers version? Thank you.\r\n\r\nHi, the information about my setup:\r\nPython version: 3.6.9\r\nTensorflow Version: 2.0.0b1\r\nPyTorch Transformers version (or branch):2.1.1\r\n\r\nThank you", "I can't reproduce this on my side while on the same versions. Would you happen to have a folder called \"bert-base-uncased\" in the same directory from which you're calling the python script?", "Hi,\r\n\r\nI have exactly the same error while doing the same task...\r\n\r\nPython version: Python 3.5.3\r\nTensorflow Version: 2.0.0-beta1\r\nPyTorch Transformers version : transformers (2.1.1)\r\n", "In my environment, **it works as expected**!\r\n@YumingLiu1996 @LeonardoGracioS \r\n\r\nN.B: It works without `force_download=True` parameter too.\r\n\r\n```\r\n>>> from transformers import TFBertForSequenceClassification\r\n2019-11-12 14:38:24.462663: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA\r\n2019-11-12 14:38:24.466451: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 3408000000 Hz\r\n2019-11-12 14:38:24.466942: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x56248b3c66f0 executing computations on platform Host. Devices:\r\n2019-11-12 14:38:24.466956: I tensorflow/compiler/xla/service/service.cc:175] StreamExecutor device (0): Host, Default Version\r\n>>> model = TFBertForSequenceClassification.from_pretrained('bert-base-uncased', force_download=True)\r\n100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 313/313 [00:00<00:00, 154303.85B/s]\r\n100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 536063208/536063208 [00:48<00:00, 11140340.62B/s]\r\n>>> import platform\r\n>>> platform.platform()\r\n'Linux-4.15.0-66-generic-x86_64-with-debian-buster-sid'\r\n>>> platform.python_version()\r\n'3.6.9'\r\n>>> import torch\r\n>>> torch.__version__\r\n'1.0.1'\r\n>>> import tensorflow as tf\r\n>>> tf.__version__\r\n'2.0.0'\r\n>>> import transformers\r\n>>> transformers.__version__\r\n'2.1.1'\r\n```", "> I can't reproduce this on my side while on the same versions. Would you happen to have a folder called \"bert-base-uncased\" in the same directory from which you're calling the python script?\r\n\r\n\r\n\r\n> I can't reproduce this on my side while on the same versions. Would you happen to have a folder called \"bert-base-uncased\" in the same directory from which you're calling the python script?\r\n\r\nNo, I don't have such as folder.. ", "> I can't reproduce this on my side while on the same versions. Would you happen to have a folder called \"bert-base-uncased\" in the same directory from which you're calling the python script?\r\n\r\nI redownload tensorflow 2.0.0 and transformers. And now I encounter another problem:\r\n\r\n```\r\nh5py/_objects.pyx in h5py._objects.with_phil.wrapper()\r\n\r\nh5py/_objects.pyx in h5py._objects.with_phil.wrapper()\r\n\r\nh5py/h5f.pyx in h5py.h5f.open()\r\n\r\nOSError: Unable to open file (truncated file: eof = 485070229, sblock->base_addr = 0, stored_eof = 497933648)\r\n```\r\nDo you know the solution for this problem?\r\n", "@YumingLiu1996 the error `OSError: Unable to open file` tells you that the file was not downloaded in its entirety. Please add the `force_download=True` parameter when downloading the pre-trained model:\r\n\r\n```py\r\nmodel = TFBertForSequenceClassification.from_pretrained('bert-base-uncased', force_download=True)\r\n```", "I solved the issue by using a Google Cloud Platform virtual machine with a TF 2.0 environment.", "> I solved the issue by using a Google Cloud Platform virtual machine with a TF 2.0 environment.\r\n\r\nI am using a Google Cloud Platform with an K80 and I have the TF2 installed but it doesn't works...\r\nthe same problem here guys. \r\n\r\nI don't know why, but it only works on Google Colab until this comment. ", "> > I solved the issue by using a Google Cloud Platform virtual machine with a TF 2.0 environment.\r\n> \r\n> I am using a Google Cloud Platform with an K80 and I have the TF2 installed but it doesn't works...\r\n> the same problem here guys.\r\n> \r\n> I don't know why, but it only works on Google Colab until this comment.\r\n\r\nDo you have exactly the same error ? Did you start your virtual machine with a TF 2.0 environment and not just a \"pip install tensorflow=2.0.0\" ?", "I did through the pip install, because I have this machine running some\nother things. So, the problem is the same but I think could be the\nTensorflow version.\n\nI will try to build TF 2 from source.\n\n\n\nEm sex, 15 de nov de 2019 06:07, Samuel Leonardo Gracio <\[email protected]> escreveu:\n\n> I solved the issue by using a Google Cloud Platform virtual machine with a\n> TF 2.0 environment.\n>\n> I am using a Google Cloud Platform with an K80 and I have the TF2\n> installed but it doesn't works...\n> the same problem here guys.\n>\n> I don't know why, but it only works on Google Colab until this comment.\n>\n> Do you have exactly the same error ? Did you start your virtual machine\n> with a TF 2.0 environment and not just a \"pip install tensorflow=2.0.0\" ?\n>\n> —\n> You are receiving this because you commented.\n> Reply to this email directly, view it on GitHub\n> <https://github.com/huggingface/transformers/issues/1780?email_source=notifications&email_token=AB4RK3ZWJWXUQDPECFSVTPLQTZRGXA5CNFSM4JLK64N2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEEEZC4Q#issuecomment-554275186>,\n> or unsubscribe\n> <https://github.com/notifications/unsubscribe-auth/AB4RK32QNPOWOL3OVKAP2FTQTZRGXANCNFSM4JLK64NQ>\n> .\n>\n", "same problem here:\r\n\r\n OS:\r\n Python version: 3.6.8\r\n PyTorch version: 1.2.0\r\n TF version: 2.0.0b1\r\n PyTorch Transformers version (or branch): master (Nov 25, 2019)\r\n\r\nmodel = TFBertForSequenceClassification.from_pretrained('bert-base-cased')\r\n\r\nValueError: Expected floating point type, got <dtype: 'int32'>.\r\n\r\nlooks like a tensorflow problem\r\n", "I've found your problem! It's a TensorFlow bug in version 2.0.0beta1. If you uninstall this version with `pip uninstall tensorflow` and after that you install the TensorFlow with version 2.0.0a0 with `pip install tensorflow==2.0.0,` **the code works as expected**!\r\n\r\nI've tried different settings and it works as expected!\r\n- PyTorch 1.2.0, TensorFlow(CPU-version) 2.0.0\r\n- PyTorch 1.2.0, TensorFlow(GPU-version) 2.0.0\r\n- PyTorch 1.3.1, TensorFlow(CPU-version) 2.0.0\r\n- PyTorch 1.3.1, TensorFlow(GPU-version) 2.0.0\r\n\r\n> same problem here:\r\n> \r\n> ```\r\n> OS:\r\n> Python version: 3.6.8\r\n> PyTorch version: 1.2.0\r\n> TF version: 2.0.0b1\r\n> PyTorch Transformers version (or branch): master (Nov 25, 2019)\r\n> ```\r\n> \r\n> model = TFBertForSequenceClassification.from_pretrained('bert-base-cased')\r\n> \r\n> ValueError: Expected floating point type, got <dtype: 'int32'>.\r\n> \r\n> looks like a tensorflow problem", "thanks. I tryied with a0 but it has even more problem, like \r\n\r\n'tensorflow.python.keras.api._v2.keras.layers' has no attribute 'LayerNormalization'\r\n\r\nwhich are said to be bugs in a0 and patched in b1. \r\n\r\nHowever b1 does not work because\r\nValueError: Expected floating point type, got <dtype: 'int32'>.\r\n\r\nLooks like this is a deadlock.\r\n\r\n> I've found your problem! It's a TensorFlow bug in version 2.0.0beta1. If you uninstall this version with `pip uninstall tensorflow` and after that you install the TensorFlow with version 2.0.0a0 with `pip install tensorflow==2.0.0,` **the code works as expected**!\r\n> \r\n> I've tried different settings and it works as expected!\r\n> \r\n> * PyTorch 1.2.0, TensorFlow(CPU-version) 2.0.0\r\n> \r\n> * PyTorch 1.2.0, TensorFlow(GPU-version) 2.0.0\r\n> \r\n> * PyTorch 1.3.1, TensorFlow(CPU-version) 2.0.0\r\n> \r\n> * PyTorch 1.3.1, TensorFlow(GPU-version) 2.0.0\r\n> \r\n> \r\n> > same problem here:\r\n> > ```\r\n> > OS:\r\n> > Python version: 3.6.8\r\n> > PyTorch version: 1.2.0\r\n> > TF version: 2.0.0b1\r\n> > PyTorch Transformers version (or branch): master (Nov 25, 2019)\r\n> > ```\r\n> > \r\n> > \r\n> > model = TFBertForSequenceClassification.from_pretrained('bert-base-cased')\r\n> > ValueError: Expected floating point type, got <dtype: 'int32'>.\r\n> > looks like a tensorflow problem\r\n\r\n", "I don't know what to say. I've just tested with Transformers v-2.2.0 and it works as expected. Sorry, but in my environment it works as I've said some days ago. How can i help you? How can i replicate your problem? Which OS are you using?\r\n\r\n> thanks. I tryied with a0 but it has even more problem, like\r\n> \r\n> 'tensorflow.python.keras.api._v2.keras.layers' has no attribute 'LayerNormalization'\r\n> \r\n> which are said to be bugs in a0 and patched in b1.\r\n> \r\n> However b1 does not work because\r\n> ValueError: Expected floating point type, got <dtype: 'int32'>.\r\n> \r\n> Looks like this is a deadlock.\r\n> \r\n> > I've found your problem! It's a TensorFlow bug in version 2.0.0beta1. If you uninstall this version with `pip uninstall tensorflow` and after that you install the TensorFlow with version 2.0.0a0 with `pip install tensorflow==2.0.0,` **the code works as expected**!\r\n> > I've tried different settings and it works as expected!\r\n> > ```\r\n> > * PyTorch 1.2.0, TensorFlow(CPU-version) 2.0.0\r\n> > \r\n> > * PyTorch 1.2.0, TensorFlow(GPU-version) 2.0.0\r\n> > \r\n> > * PyTorch 1.3.1, TensorFlow(CPU-version) 2.0.0\r\n> > \r\n> > * PyTorch 1.3.1, TensorFlow(GPU-version) 2.0.0\r\n> > ```\r\n> > \r\n> > \r\n> > > same problem here:\r\n> > > ```\r\n> > > OS:\r\n> > > Python version: 3.6.8\r\n> > > PyTorch version: 1.2.0\r\n> > > TF version: 2.0.0b1\r\n> > > PyTorch Transformers version (or branch): master (Nov 25, 2019)\r\n> > > ```\r\n> > > \r\n> > > \r\n> > > model = TFBertForSequenceClassification.from_pretrained('bert-base-cased')\r\n> > > ValueError: Expected floating point type, got <dtype: 'int32'>.\r\n> > > looks like a tensorflow problem", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "Even I'm facing this issue @LysandreJik @thomwolf can you throw some input on this?\r\nall the input is of the same length still this issue occurs.\r\n@devroy73 meanwhile in data loader you can set drop_last = True", "Hey @anandhperumal thanks for that it solved my crashing issue. ", "I tried setting drop_last=True, but it did not fix the issue for me.", "maybe we should add parameter to Trainer. \r\nor it should added to the doc for others " ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1778?src=pr&el=h1) Report\n> Merging [#1778](https://codecov.io/gh/huggingface/transformers/pull/1778?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/7a9aae1044aa4699310a8004f631fc0a4bdf1b65?src=pr&el=desc) will **increase** coverage by `0.06%`.\n> The diff coverage is `100%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1778?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1778 +/- ##\n==========================================\n+ Coverage 84.03% 84.09% +0.06% \n==========================================\n Files 94 94 \n Lines 14032 14036 +4 \n==========================================\n+ Hits 11792 11804 +12 \n+ Misses 2240 2232 -8\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1778?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_tf\\_pytorch\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1778/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3B5dG9yY2hfdXRpbHMucHk=) | `95.13% <100%> (+2.18%)` | :arrow_up: |\n| [transformers/modeling\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1778/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3V0aWxzLnB5) | `91.05% <100%> (+1.41%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1778?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1778?src=pr&el=footer). Last update [7a9aae1...90f6e73](https://codecov.io/gh/huggingface/transformers/pull/1778?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Looks good to me, thanks @eukaryote31 !", "also cc @sshleifer ", "LGTM!", "Ok great, merging!", "This was incorrect: we have multiple models that contain a valid `lm_head.decoder` layer (RoBERTa and its variants, CamemBERT, XLM-R...) and this PR broke `from_pretrained` for those models in the `ForMaskedLM` case.\r\n\r\nThe way to go in the particular `DialoGPT` case would be to either:\r\n- create a conversion script and upload the GPT2 compatible weights somewhere.\r\n- or, add a special test in from_pretrained that the current instance is an instance of `GPT2...Head` (this is not pretty in terms of OOP encapsulation 💩)\r\n- or, expose some kind of overrideable list of key mapping (that can be implemented by the GPT2 subclass)\r\n\r\n(1. seems like the easier solution) \r\n\r\ncc @LysandreJik @sshleifer @thomwolf ", "Also cc @patrickvonplaten 🤯" ]

[]

[]

[ "Duplicate of #1649 #1420 #1522 #1564 🤣🤣" ]

[]

[]

[ "Please do not open duplicate issues (#1774)" ]

[]

[]

[ "We haven't updated the pip version yet, we'll do so in the following weeks. Please install it from source in the meantime:\r\n\r\n```\r\npip install git+https://github.com/huggingface/transformers\r\n```", "Traceback (most recent call last):\r\n File \"run_pplm.py\", line 936, in <module>\r\n run_pplm_example(**vars(args))\r\n File \"run_pplm.py\", line 738, in run_pplm_example\r\n tokenizer = GPT2Tokenizer.from_pretrained(pretrained_model)\r\n File \"/home/xps/anaconda3/envs/nlg_entity/lib/python3.7/site-packages/transformers/tokenization_utils.py\", line 911, in from_pretrained\r\n return cls._from_pr\r\netrained(*inputs, **kwargs)\r\n File \"/home/xps/anaconda3/envs/nlg_entity/lib/python3.7/site-packages/transformers/tokenization_utils.py\", line 1014, in _from_pretraine\r\n list(cls.vocab_files_names.values()),\r\nOSError: Model name 'gpt2-medium' was not found in tokenizers model name list (gpt2, gpt2-medium, gpt2-large, gpt2-xl, distilgpt2). We asswas a path, a model identifier, or url to a directory containing vocabulary files named ['vocab.json', 'merges.txt'] but couldn't find sucat this path or url.\r\n\r\n\r\n\r\n\r\nI also reported the same error....@LysandreJik\r\n", "> 回溯（最近一次调用）：\r\n> 文件“run_pplm.py”，第 936 行，在\r\n> run_pplm_example(**vars(args))\r\n> 文件“run_pplm.py”，第 738 行，在 run_pplm_example\r\n> tokenizer = GPT2Tokenizer.from_pretrained(pretrained_model)\r\n> 文件“ /home/xps/anaconda3/envs/nlg_entity/lib/python3.7/site-packages/transformers/tokenization_utils.py\", line 911, in from_pretrained\r\n> return\r\n> cls._from_pretrained(*inputs, **kwargs)\r\n> File \"/home /xps/anaconda3/envs/nlg_entity/lib/python3.7/site-packages/transformers/tokenization_utils.py”，第 1014 行，在 _from_pretraine\r\n> 列表（cls.vocab_files_names.values()）中，\r\n> OSError：在标记器模型名称列表（gpt2、gpt2-medium、gpt2-large、gpt2-xl、distlgpt2）中找不到模型名称“gpt2-medium”。我们为包含名为 ['vocab.json', 'merges.txt'] 的词汇文件的目录添加了路径、模型标识符或 url，但找不到此路径或 url。\r\n> \r\n> 我也报了同样的错误.... @LysandreJik\r\n\r\nhi,you can run this model?" ]

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

[]

[ "Hi! You have access to the models' internals so you could definitely rewrite the main loop function to handle such a use-case!\r\n\r\nYou can access the hidden layers as follows:\r\n\r\n```py\r\nblocks = model.transformer.h # this is a list of \"block\"s\r\nblock[0] # contains the MLP/LayerNorm/attention layers\r\n```\r\n\r\nI'd recommend taking a look at the `forward` method of the [`GPT2DoubleLMHeadModel`](https://github.com/huggingface/transformers/blob/master/transformers/modeling_gpt2.py#L642) to \r\nunderstand how the `transformer` is managed, and to take a look at the [loop inside the `GPT2Model`](https://github.com/huggingface/transformers/blob/master/transformers/modeling_gpt2.py#L451) which you will probably have to replicate a little to achieve what you want to do.", "Hello, \r\nThank you very much for your reply.\r\nSo if I am understanding this correctly, the part of the loop inside GPT2 model below is what I am supposed to use to generate output based on a given hidden state, am I correct?\r\n```\r\noutputs = block(hidden_states,\r\n layer_past=layer_past,\r\n attention_mask=attention_mask,\r\n head_mask=head_mask[i])\r\n```\r\nAlso, I would like to clarify what ‘block’ actually refers to in this code. Is ‘block’ same as GPT-2 ‘layer’? It’s not very clear to me what `block[1]`, `block[2]`, etc. represents.\r\n\r\nThank you again,", "Yes.", "Hello,\r\n\r\nWhat exactly is the `block`? Does `block[0]` refer to the first gtp-2 layer, and `block[1]` refers to the second gtp-2 layer and so on? How can I utilize `block` to have access to the uppermost output layer of a gtp-2?\r\n\r\nThank you,", "Yes, `block` refers to the GPT-2 layers. `block[0]` is the first layer, ... `block[-1]` is the last layer.", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1773?src=pr&el=h1) Report\n> Merging [#1773](https://codecov.io/gh/huggingface/transformers/pull/1773?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/0cb163865a4c761c226b151283309eedb2b1ca4d?src=pr&el=desc) will **increase** coverage by `1.06%`.\n> The diff coverage is `100%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1773?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1773 +/- ##\n==========================================\n+ Coverage 82.67% 83.74% +1.06% \n==========================================\n Files 111 108 -3 \n Lines 16162 15749 -413 \n==========================================\n- Hits 13362 13189 -173 \n+ Misses 2800 2560 -240\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1773?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/1773/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2JlcnQucHk=) | `87.7% <100%> (+0.02%)` | :arrow_up: |\n| [transformers/hf\\_api.py](https://codecov.io/gh/huggingface/transformers/pull/1773/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2hmX2FwaS5weQ==) | `96.87% <0%> (-0.63%)` | :arrow_down: |\n| [transformers/tests/modeling\\_bert\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1773/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX2JlcnRfdGVzdC5weQ==) | `96.5% <0%> (-0.58%)` | :arrow_down: |\n| [transformers/tests/modeling\\_common\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1773/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX2NvbW1vbl90ZXN0LnB5) | `74.23% <0%> (-0.31%)` | :arrow_down: |\n| [transformers/tokenization\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/1773/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl9vcGVuYWkucHk=) | `81.81% <0%> (-0.3%)` | :arrow_down: |\n| [transformers/tests/modeling\\_openai\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1773/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX29wZW5haV90ZXN0LnB5) | `93.2% <0%> (-0.2%)` | :arrow_down: |\n| [transformers/tests/modeling\\_xlnet\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1773/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3hsbmV0X3Rlc3QucHk=) | `94.68% <0%> (-0.2%)` | :arrow_down: |\n| [transformers/tests/modeling\\_albert\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1773/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX2FsYmVydF90ZXN0LnB5) | `95.08% <0%> (-0.16%)` | :arrow_down: |\n| [transformers/tests/modeling\\_gpt2\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1773/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX2dwdDJfdGVzdC5weQ==) | `94.01% <0%> (-0.15%)` | :arrow_down: |\n| [transformers/tokenization\\_xlm.py](https://codecov.io/gh/huggingface/transformers/pull/1773/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl94bG0ucHk=) | `83.2% <0%> (-0.14%)` | :arrow_down: |\n| ... and [40 more](https://codecov.io/gh/huggingface/transformers/pull/1773/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1773?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1773?src=pr&el=footer). Last update [0cb1638...6fb9900](https://codecov.io/gh/huggingface/transformers/pull/1773?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "@rlouf Thanks for the the PR!\r\nI'm trying to run run_summarization.py with the following command, but got an error. Any idea why?\r\nI have unzipped the cnn and dailymail folders in my data directory. \r\n\r\n```\r\npython examples/run_summarization.py \\\r\n --data_dir /data/home/hlu/notebooks/summarization/data \\\r\n --output_dir ./summarization_output \\\r\n --do_train True \\\r\n --model_name_or_path bert-base-cased \\\r\n --num_train_epochs 1 \\\r\n```\r\n\r\nW1108 21:29:27.643532 139837444753152 run_summarization.py:488] Process rank: 0, device: cuda, n_gpu: 1, distributed training: False, 16-bits training: False\r\nI1108 21:29:27.643716 139837444753152 run_summarization.py:491] Training/evaluation parameters Namespace(data_dir='/data/home/hlu/notebooks/summarization/data', device=device(type='cuda'), do_evaluate=False, do_overwrite_output_dir=False, do_train=True, gradient_accumulation_steps=1, max_grad_norm=1.0, max_steps=-1, model_name_or_path='bert-base-cased', model_type='bert', n_gpu=1, num_train_epochs=1, output_dir='./summarization_output', per_gpu_eval_batch_size=4, per_gpu_train_batch_size=4, seed=42, to_cpu=False)\r\nI1108 21:30:40.805522 139837444753152 run_summarization.py:198] ***** Running training *****\r\nI1108 21:30:40.805693 139837444753152 run_summarization.py:199] Num examples = 312085\r\nI1108 21:30:40.806245 139837444753152 run_summarization.py:200] Num Epochs = 1\r\nI1108 21:30:40.806311 139837444753152 run_summarization.py:201] Instantaneous batch size per GPU = 4\r\nI1108 21:30:40.806385 139837444753152 run_summarization.py:204] Total train batch size (w. parallel, distributed & accumulation) = 4\r\nI1108 21:30:40.806446 139837444753152 run_summarization.py:207] Gradient Accumulation steps = 1\r\nI1108 21:30:40.806501 139837444753152 run_summarization.py:208] Total optimization steps = 78022\r\nTraceback (most recent call last):\r\n File \"examples/run_summarization.py\", line 545, in <module>\r\n main()\r\n File \"examples/run_summarization.py\", line 497, in main\r\n global_step, tr_loss = train(args, model, tokenizer)\r\n File \"examples/run_summarization.py\", line 236, in train\r\n decoder_lm_labels=lm_labels,\r\n File \"/data/anaconda/envs/nlp_gpu/lib/python3.6/site-packages/torch/nn/modules/module.py\", line 547, in __call__\r\n result = self.forward(*input, **kwargs)\r\n File \"/data/home/hlu/notebooks/huggingface/transformers/transformers/modeling_encoder_decoder.py\", line 240, in forward\r\n decoder_outputs = self.decoder(decoder_input_ids, **kwargs_decoder)\r\n File \"/data/anaconda/envs/nlp_gpu/lib/python3.6/site-packages/torch/nn/modules/module.py\", line 547, in __call__\r\n result = self.forward(*input, **kwargs)\r\n File \"/data/home/hlu/notebooks/huggingface/transformers/transformers/modeling_bert.py\", line 865, in forward\r\n encoder_attention_mask=encoder_attention_mask)\r\n File \"/data/anaconda/envs/nlp_gpu/lib/python3.6/site-packages/torch/nn/modules/module.py\", line 547, in __call__\r\n result = self.forward(*input, **kwargs)\r\n File \"/data/home/hlu/notebooks/huggingface/transformers/transformers/modeling_bert.py\", line 677, in forward\r\n extended_attention_mask = causal_mask[:, None, :, :] * attention_mask[:, None, None, :]\r\nRuntimeError: expected device cuda:0 and dtype Long but got device cuda:0 and dtype Bool\r\n", "Hi! Thank you for raising the issue. I think it comes from the following lines in `modeling_bert.py`:\r\n\r\n```python\r\ncausal_mask = seq_ids[None, None, :].repeat(batch_size, seq_length, 1) <= seq_ids[None, :, None]\r\nextended_attention_mask = causal_mask[:, None, :, :] * attention_mask[:, None, None, :]\r\n```\r\n\r\n`causal_mask` as defined is a boolean tensor, while the `attention_mask` being passed is of type `long`. I am very confused because I never got the error when I tried locally. I will fix the issue asap. In the meantime, would you mind executing the following code in your environment and pasting the result here?\r\n\r\n```python\r\nimport platform; print(\"Platform\", platform.platform())\r\nimport sys; print(\"Python\", sys.version)\r\nimport torch; print(\"PyTorch\", torch.__version__)\r\n```", "@rlouf Thanks! Please see the environment info below.\r\n>>> import platform; print(\"Platform\", platform.platform())\r\nPlatform Linux-4.15.0-1061-azure-x86_64-with-debian-stretch-sid\r\n>>> import sys; print(\"Python\", sys.version)\r\nPython 3.6.8 |Anaconda, Inc.| (default, Dec 30 2018, 01:22:34) \r\n[GCC 7.3.0]\r\n>>> import torch; print(\"Pytorch\", torch.__version__)\r\nPytorch 1.2.0", "Thank you! I’m working with PyTorch 1.3.0, and they introduced type promotion in this release (https://github.com/pytorch/pytorch/releases) which surely explains why it worked for me but not for you. I will work on a fix this week, in the meantime you can simply convert causal mask manually to long, or upgrade to pytorch 1.3 if that is possible.\r\n\r\nEdit: I pushed the changes. Let me know if there still is an issue.", "Here is the latest on summarization:\r\n\r\n- I added utilities to load the CNN/DailyMail dataset, to separate stories from summaries. The utilities are tested;\r\n- I wrapped the BertAbs model along with its configuration to be able to use it with the library's usual API `BertAbs.from_pretrained`. I uploaded the finetuned weights and configuration on S3 and tested that model loads properly;\r\n- I also wrapped the sequence generation so we can generate summaries from any text using Beam Search;\r\n- I removed my previous attempt at Beam Search and saved it for later;\r\n- I added a small fix in `modeling_bert.py` for users that use PyTorch v < 1.30. We convert the causal mask explicitly from `bool` to `long` since type promotion was only introduce in PyTorch 1.3\r\n- I added ROUGE evaluation and am able to reproduce the authors' results on CNN/DailyMail\r\n\r\n- [ ] revert changes in `modeling_encoderdecoder.py`\r\n\r\n\r\n\r\nI squashed my commits and rebased on the repo's master branch. The docs are updated. Good to go on my side." ]

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

[]

[ "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "> ## Bug\r\n> Model I am using (XLM):\r\n> \r\n> Language I am using the model on (English and Chinese):\r\n> \r\n> The problem arise when using:\r\n> \r\n> [CUDA_VISIBLE_DEVICES=2 python run_squad.py --model_type xlm --model_name_or_path xlm-mlm-tlm-xnli15-1024 --do_train --do_eval --train_file $SQUAD_DIR/Crosslingual_QZH_train.json --predict_file $SQUAD_DIR/Crosslingual_QZH_valid.json --per_gpu_train_batch_size 12 --learning_rate 3e-5 --num_train_epochs 100 --max_seq_length 384 --doc_stride 128 --eval_all_checkpoints --save_steps 50 --evaluate_during_training --output_dir /home/weihua/Sqad/transformers/xlm_out/ ]\r\n> \r\n> error:\r\n> Traceback (most recent call last):\r\n> File \"run_squad.py\", line 569, in \r\n> main()\r\n> File \"run_squad.py\", line 515, in main\r\n> global_step, tr_loss = train(args, train_dataset, model, tokenizer)\r\n> File \"run_squad.py\", line 179, in train\r\n> results = evaluate(args, model, tokenizer)\r\n> File \"run_squad.py\", line 275, in evaluate\r\n> args.version_2_with_negative, tokenizer, args.verbose_logging)\r\n> File \"/home/weihua/Sqad/transformers/examples/utils_squad.py\", line 814, in write_predictions_extended\r\n> final_text = get_final_text(tok_text, orig_text, tokenizer.do_lower_case,\r\n> AttributeError: 'XLMTokenizer' object has no attribute 'do_lower_case'\r\n> \r\n> \r\n> \r\n> \r\n> How can I deal with this problem?\r\n\r\nLook [here](https://github.com/huggingface/transformers/blob/master/transformers/tokenization_xlm.py). This is the source code of the XLM tokenizer. In the ___init__()_ method, you can see that the _do_lower_case_ parameter doesn't exist, but it exists the **do_lowercase_and_remove_accent** parameter.\r\n\r\nSaid this, I tell you a solution to your problem (a wordaround).\r\nIf you see the source code of [run_squad.py](https://github.com/huggingface/transformers/blob/master/examples/run_squad.py) at rows 487-489, you see that here the tokenizer (different according to the model chosen by you - in your case XLM) is loaded, passing the _do_lower_case_ parameter. But as I said before, the tokenizer for XLM model accepts a different parameter. So, you can insert an if/else statement (in pseudo-code):\r\n```\r\nif args.model_type == 'XLM':\r\n tokenizer = tokenizer_class.from_pretrained(..., do_lower_case_and_remove_accent=args.do_lower_case, ...)\r\nelse:\r\n tokenizer = tokenizer_class.from_pretrained(..., do_lower_case=args.do_lower_case, ...)\r\n```\r\nN.B. if you want to be more comfortable of my solution to your problem, please see the source code of the `tokenization_xlm.py` in your HuggingFace version whether it contains the _do_lowercase_and_remove_accent_ parameter and not the _do_lowercase_ parameter. \r\n\r\nKeep us updated on your problem! Good luck", "Actually, I just replace the line 814 in utils_squad.py with 'final_text = get_final_text(tok_text, orig_text,verbose_logging)'. It means I no longer pass ‘tokenizer.do_lower_case’ to this function. Because I find that in the tokenization_xlm.py, the parameter 'do_lower_case_and_remove_accent' is already set. Although this allows my program to run, it does not give good results. So I suspect that my change is wrong.\r\nI also tried to follow your suggestion above. But the results did not improve. \r\nThis is the result of the 700th step. \r\n\r\nThe exact and f1 is very low. I used bert-base-multilingual-cased to do SQuAD with the same data set before. The exact and f1 increased very fast. At around 200th step, the f1 can reach 40+, and exact is about 30.\r\nIs it wrong with my parameter settings that caused these two scores to rise slowly when I use XLM?\r\nOr is it necessary to take longer to finetune the SQuAD task in XLM than in BERT?\r\nThis is my first time to do the SQuAD. I sincerely hope that you can give me some advice.\r\nThank you.", "I also tried to train and test on the official dataset,\r\nthat is \r\n\r\nin the examples document.\r\n\r\nbut the situation was the same as what I got on my own dataset.\r\nAt the same time, I find that when I use the XLM model with learning-rate= 3e-5, the loss decreases very slowly. \r\nIs this due to the inappropriateness of my parameter settings? Or is it because the loss function of the model needs to be changed?", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1770?src=pr&el=h1) Report\n> Merging [#1770](https://codecov.io/gh/huggingface/transformers/pull/1770?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/1c542df7e554a2014051dd09becf60f157fed524?src=pr&el=desc) will **increase** coverage by `<.01%`.\n> The diff coverage is `90%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1770?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1770 +/- ##\n==========================================\n+ Coverage 84.03% 84.03% +<.01% \n==========================================\n Files 94 94 \n Lines 14032 14034 +2 \n==========================================\n+ Hits 11792 11794 +2 \n Misses 2240 2240\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1770?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/1770/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2JlcnQucHk=) | `87.82% <90%> (+0.04%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1770?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1770?src=pr&el=footer). Last update [1c542df...cd286c2](https://codecov.io/gh/huggingface/transformers/pull/1770?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Ok great @rlouf!", "I wonder why encoder_attention_mask use decoder inputs shape when it's not specified. \r\n`encoder_attention_mask = torch.ones(input_shape, device=device)`", "Indeed, wanna have a look @rlouf?", "That's a very good point, thanks @efeiefei ! See #2107" ]

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

[]

[ "cc @aconneau 😬", "is there any update in the XLM-R model? ", "Let me know if you need some-help in porting the xlm-r models to HF.", "I think that's maybe not the correct way, but I adjusted the `convert_roberta_original_pytorch_checkpoint_to_pytorch.py` script to convert the `fairseq` model into a `transformers` compatible model file. I used the `sentencepiece` BPE loader and adjusted the vocab size.\r\n\r\nThen I used the `CamemBERT` model class to perform some evaluations on NER. But the result is not really good (I tried to replicate the CoNLL-2003 for English).\r\n\r\nSo I guess it is not as simple as this first attempt 😅\r\n\r\n---\r\n\r\nGist for the conversion script is [here](https://gist.github.com/stefan-it/1eaf537a853aee6ab8fe5a425bf09ed7).\r\n\r\nThe `CamemBERT` model configuration looks pretty much the same as XLM-R large?!", "> I think that's maybe not the correct way, but I adjusted the `convert_roberta_original_pytorch_checkpoint_to_pytorch.py` script to convert the `fairseq` model into a `transformers` compatible model file. I used the `sentencepiece` BPE loader and adjusted the vocab size.\r\n> \r\n> Then I used the `CamemBERT` model class to perform some evaluations on NER. But the result is not really good (I tried to replicate the CoNLL-2003 for English).\r\n> \r\n> So I guess it is not as simple as this first attempt 😅\r\n> \r\n> Gist for the conversion script is [here](https://gist.github.com/stefan-it/1eaf537a853aee6ab8fe5a425bf09ed7).\r\n> \r\n> The `CamemBERT` model configuration looks pretty much the same as XLM-R large?!\r\n\r\nHi @stefan-it, do you have any update for your attempt?", "The final models have been released today 😍\r\n\r\nhttps://github.com/pytorch/fairseq/tree/master/examples/xlmr\r\n\r\nSo I'm going to try the conversion with these models tomorrow/in the next days :)", "I think the model conversion is done correctly. But: the `CamembertTokenizer` implementation can't be used, because it adds some special tokens. I had to modify the tokenizer to match the output of the `fairseq` tokenization/`.encode()` method :) I'll report back some results on NER later.\r\n\r\n**update**: I could achieve 90.41% on CoNLL-2003 (English), paper reports 92.74 (using Flair).\r\n**update 2**: Using the `run_ner.py` example (incl. some hours of tokenization debugging...): 96.22 (dev) and 91.91 (test).", "Btw I was using the XLM-R v0 checkpoints in a project I'm working on and the v0 checkpoints worked slightly better than the checkpoints added today. Is it possible to also add the older checkpoints?", "I think it's the best solution to offer both checkpoint versions! In my opinion, the _ideal case_ is that, as like to other models in Transformers, you can select which version of XLM-R checkpoints to use, e.g.\r\n```\r\n> from transformers import XLMRModel\r\n> base_model = XLMRModel.from_pretrained('xlmr-base') # 250M parameters\r\n> large_model = XLMRModel.from_pretrained('xlmr-large') # 560M parameters\r\n```\r\n\r\n> Btw I was using the XLM-R v0 checkpoints in a project I'm working on and the v0 checkpoints worked slightly better than the checkpoints added today. Is it possible to also add the older checkpoints?", "Btw using XLM-R I encounter this issue: \r\nBatch size affecting output. #2401\r\n\r\nThis is really annoying and makes it hard to use the model.", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n", "@ricardorei Did you happen to successfully use the XLM-R model ?\r\n\r\nI'm trying to see how this model can be used as pretrained step for NMT tasks, I tried raw version from XLM facebook repo and ran into multiple OOM issues. \r\n\r\nThe best suggestion so far I got is to try smaller version of Fairseq xlmr (base) on p3dn.24xlarge instance or the Google TPU Pytorch way. \r\n\r\nThanks ! ", "@mohammedayub44\r\n\r\nI am using the base model which runs well in a 12GB GPU with batch size of 8. Depending on your implementation and task you can run even bigger batches (16, 24 for example).\r\n\r\nAnd I am also using the version directly from Fairseq, because you can load the v0 checkpoint.\r\n\r\nThe variability in my prediction with different batch sizes I could never figure out. Probably some floating-point precision issues going on under the hood. It doesn't change overall performance but it is annoying...\r\n\r\n ", "BTW, I am using the TF variant from https://huggingface.co/jplu/tf-xlm-roberta-base and https://huggingface.co/jplu/tf-xlm-roberta-large . I have successfully finetuned even the large model on a 16GB GPU and it was performing substantially better than the base model (on Czech Q&A).", "@ricardorei \r\nThanks for the confirmation. I'm okay with v0 checkpoints, I just need to check if the model can be fine-tuned for NMT. I'm guessing you're fine tuning for Classification tasks. \r\n\r\nIf you could share the prepare and train commands you are using. It would be easier than digging deep into every fairseq hyperparamter. \r\n\r\nThanks ! ", "@foxik Is TF variant more suitable for fine-tuning. Any particular preprocessing steps you carried out for fine-tuning. If you can share them, I can map the same for NMT task. \r\n\r\nThanks ! ", "@mohammedayub44 Yes I was using it for classification/regression. In your case, you need the encoder and decoder part which would take a lot more space. I would suggest that you share parameters between you encoder and decoder. \r\n\r\nI know that, with the right hyperparameter, you can achieve good results by sharing the parameters between your encoder and decoder -> [A Simple and Effective Approach to Automatic Post-Editing\r\nwith Transfer Learning](https://arxiv.org/pdf/1906.06253.pdf)\r\n\r\nIn terms of hyperparameters that I am using, they are very simple. I freeze the encoder for 1 epoch while fine-tuning the classification head and then I fine-tune the entire model. My classification-head has a learning rate of 0.00003 while XLM-R has 0.00001. The optimizer is a standard Adam. This combination of gradual unfreezing with discriminative learning rates works well in my task.", "@ricardorei Thanks for sharing the paper. Some interesting results there. \r\nAny hints on how I can setup both encoder and decoder of XLM-R and share the parameters using HuggingFace library. I could only find [LM fine-tuning](https://github.com/huggingface/blog/blob/master/notebooks/01_how_to_train.ipynb) examples and notebook file. Nothing on NMT based fine-tuning. " ]

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

[]

[ "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" ]

[]

[]

[ "> ## Bug\r\n> Model I am using (Bert, XLNet....):\r\n> \r\n> Language I am using the model on (English, Chinese....):\r\n> \r\n> The problem arise when using:\r\n> \r\n> * [X ] the official example scripts: (give details)\r\n> * [ ] my own modified scripts: (give details)\r\n> \r\n> The tasks I am working on is:\r\n> \r\n> * [ ] an official GLUE/SQUaD task: (give the name)\r\n> * [x] my own task or dataset: (give details)\r\n> \r\n> ## To Reproduce\r\n> Steps to reproduce the behavior:\r\n> \r\n> 1. In the documentation, it's given how to load GPT2-XL. But I'm getting an error when I try to load is GPT2 XL available as of now or it would take some time?\r\n> \r\n> Error:\r\n> \r\n> ```\r\n> >>> model = GPT2DoubleHeadsModel.from_pretrained('gpt2-xl')\r\n> Model name 'gpt2-xl' was not found in model name list (gpt2-large, gpt2, gpt2-medium). We assumed 'gpt2-xl' was a path or url but couldn't find any file associated to this path or url.\r\n> Traceback (most recent call last):\r\n> File \"<stdin>\", line 1, in <module>\r\n> File \"/cshome/anandhpe/.local/lib/python3.5/site-packages/pytorch_transformers/modeling_utils.py\", line 492, in from_pretrained\r\n> **kwargs\r\n> File \"/cshome/anandhpe/.local/lib/python3.5/site-packages/pytorch_transformers/modeling_utils.py\", line 194, in from_pretrained\r\n> raise e\r\n> File \"/cshome/anandhpe/.local/lib/python3.5/site-packages/pytorch_transformers/modeling_utils.py\", line 180, in from_pretrained\r\n> resolved_config_file = cached_path(config_file, cache_dir=cache_dir, force_download=force_download, proxies=proxies) File \"/cshome/anandhpe/.local/lib/python3.5/site-packages/pytorch_transformers/file_utils.py\", line 124, in cached_path\r\n> raise EnvironmentError(\"file {} not found\".format(url_or_filename))\r\n> OSError: file gpt2-xl not found\r\n> ```\r\n> \r\n> ## Expected behavior\r\n> ## Environment\r\n> * OS:Ubuntu\r\n> * Python version:3.7\r\n> * PyTorch version: 1.3.1\r\n> * PyTorch Transformers version (or branch):\r\n> * Using GPU ? yes\r\n> * Distributed of parallel setup ? no\r\n> * Any other relevant information:\r\n> \r\n> ## Additional context\r\n\r\nThis \"issue\" you have opened is the same as the #1747 . It's not a bug, but a **mismatch between version on Pypi and source GitHub**.\r\n\r\nAs stated by @LysandreJik 2 days ago , they haven't released the version with OpenAI GPT-2-xl yet on Pypi. Please, install Huggingface from source with the following command:\r\n`pip install git+https://github.com/huggingface/transformers`", "@TheEdoardo93 sure thanks" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1766?src=pr&el=h1) Report\n> Merging [#1766](https://codecov.io/gh/huggingface/transformers/pull/1766?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/1c542df7e554a2014051dd09becf60f157fed524?src=pr&el=desc) will **decrease** coverage by `1.38%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1766?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1766 +/- ##\n==========================================\n- Coverage 84.03% 82.65% -1.39% \n==========================================\n Files 94 94 \n Lines 14032 14032 \n==========================================\n- Hits 11792 11598 -194 \n- Misses 2240 2434 +194\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1766?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_tf\\_pytorch\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1766/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3B5dG9yY2hfdXRpbHMucHk=) | `9.85% <0%> (-83.1%)` | :arrow_down: |\n| [transformers/tests/modeling\\_tf\\_common\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1766/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2NvbW1vbl90ZXN0LnB5) | `79.78% <0%> (-17.03%)` | :arrow_down: |\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/1766/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `59.41% <0%> (-12.36%)` | :arrow_down: |\n| [transformers/modeling\\_xlnet.py](https://codecov.io/gh/huggingface/transformers/pull/1766/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3hsbmV0LnB5) | `71.18% <0%> (-2.44%)` | :arrow_down: |\n| [transformers/modeling\\_ctrl.py](https://codecov.io/gh/huggingface/transformers/pull/1766/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2N0cmwucHk=) | `94.24% <0%> (-2.22%)` | :arrow_down: |\n| [transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/1766/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX29wZW5haS5weQ==) | `80.66% <0%> (-1.34%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1766?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1766?src=pr&el=footer). Last update [1c542df...5f1eca9](https://codecov.io/gh/huggingface/transformers/pull/1766?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Ok with that. Merging.\r\ncc @LysandreJik and his refactoring of the `run_squad` example.", "Actually seems like I can't push a merge commit on your branch to fix conflict.\r\nWill close the PR later unless you can fix it." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1765?src=pr&el=h1) Report\n> Merging [#1765](https://codecov.io/gh/huggingface/transformers/pull/1765?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/1c542df7e554a2014051dd09becf60f157fed524?src=pr&el=desc) will **not change** coverage.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1765?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1765 +/- ##\n=======================================\n Coverage 84.03% 84.03% \n=======================================\n Files 94 94 \n Lines 14032 14032 \n=======================================\n Hits 11792 11792 \n Misses 2240 2240\n```\n\n\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1765?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1765?src=pr&el=footer). Last update [1c542df...c199d96](https://codecov.io/gh/huggingface/transformers/pull/1765?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "That looks great, thanks @adrianbg !" ]

[]

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

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1764?src=pr&el=h1) Report\n> Merging [#1764](https://codecov.io/gh/huggingface/transformers/pull/1764?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/a80778f40e4738071b5d01420a0328bb00cdb356?src=pr&el=desc) will **increase** coverage by `0.02%`.\n> The diff coverage is `100%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1764?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1764 +/- ##\n==========================================\n+ Coverage 79.82% 79.84% +0.02% \n==========================================\n Files 131 131 \n Lines 19496 19519 +23 \n==========================================\n+ Hits 15562 15585 +23 \n Misses 3934 3934\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1764?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/1764/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3JvYmVydGEucHk=) | `72.57% <100%> (+0.8%)` | :arrow_up: |\n| [transformers/tests/modeling\\_roberta\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1764/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3JvYmVydGFfdGVzdC5weQ==) | `78.14% <100%> (+2.95%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1764?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1764?src=pr&el=footer). Last update [a80778f...3e52915](https://codecov.io/gh/huggingface/transformers/pull/1764?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Thanks! Any way you could do this PR with just the relevant changes (no whitespace changes, no extraneous commits)?", "Hi,\r\n\r\nI have pushed just the relevant changes with a single commit.\r\n\r\nMany thanks,\r\nDom", "Thanks! We need to update the TF model as well. (this might be the reason behind the failing unit test, btw)\r\n\r\nWill do it unless you have the bandwidth (and the inclination) to do it yourself @DomHudson ", "@DomHudson Fixed a bug introduced by this PR's commit + rebased on top of current master.\r\n\r\nPlease note that for most models, we expect users to define `attention_mask`s themselves. \r\n\r\nIf I'm not mistaken, if a user defined an attention_mask that correctly excluded padding, he would not need the padding-aware position_ids from this PR.", "LGTM merging" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1763?src=pr&el=h1) Report\n> Merging [#1763](https://codecov.io/gh/huggingface/transformers/pull/1763?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/d49c43ff789d309e688fb7b252511e9e618e46db?src=pr&el=desc) will **decrease** coverage by `0.08%`.\n> The diff coverage is `100%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1763?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1763 +/- ##\n==========================================\n- Coverage 84.11% 84.03% -0.09% \n==========================================\n Files 105 94 -11 \n Lines 15545 14032 -1513 \n==========================================\n- Hits 13076 11792 -1284 \n+ Misses 2469 2240 -229\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1763?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/modeling\\_tf\\_xlnet.py](https://codecov.io/gh/huggingface/transformers/pull/1763/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3hsbmV0LnB5) | `87.82% <100%> (-0.37%)` | :arrow_down: |\n| [transformers/modeling\\_tf\\_pytorch\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1763/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3B5dG9yY2hfdXRpbHMucHk=) | `92.95% <0%> (-2.19%)` | :arrow_down: |\n| [transformers/modeling\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1763/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3V0aWxzLnB5) | `89.64% <0%> (-1.22%)` | :arrow_down: |\n| [transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1763/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rva2VuaXphdGlvbl91dGlscy5weQ==) | `91.2% <0%> (-0.71%)` | :arrow_down: |\n| [transformers/configuration\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/1763/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL2NvbmZpZ3VyYXRpb25fYXV0by5weQ==) | `58.82% <0%> (-0.64%)` | :arrow_down: |\n| [transformers/modeling\\_tf\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/1763/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3JvYmVydGEucHk=) | `89.9% <0%> (-0.53%)` | :arrow_down: |\n| [transformers/modeling\\_tf\\_transfo\\_xl.py](https://codecov.io/gh/huggingface/transformers/pull/1763/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3RyYW5zZm9feGwucHk=) | `92.21% <0%> (-0.33%)` | :arrow_down: |\n| [transformers/modeling\\_tf\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/1763/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX3RmX3V0aWxzLnB5) | `92.4% <0%> (-0.33%)` | :arrow_down: |\n| [transformers/tests/modeling\\_tf\\_common\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1763/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX3RmX2NvbW1vbl90ZXN0LnB5) | `96.8% <0%> (-0.28%)` | :arrow_down: |\n| ... and [37 more](https://codecov.io/gh/huggingface/transformers/pull/1763/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1763?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1763?src=pr&el=footer). Last update [d49c43f...41e0859](https://codecov.io/gh/huggingface/transformers/pull/1763?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "This looks good to me, thank you @tlkh !", "I see that the changes were already incorporated, so closing this PR." ]

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

[]

[ "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

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

[ "I think you are right. Awesome if you can fix this in a PR.", "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", "Closed in #1764" ]

[]

[]

[ "> ## Bug\r\n> Model I am using (Bert, XLNet....): I am trying to use Bert encoder\r\n> \r\n> Language I am using the model on (English, Chinese....): English\r\n> \r\n> There are two problems when I use BERT for encoding text to vectors\r\n> Here is Error log\r\n> \r\n> First,\r\n> \r\n> ```\r\n> File \"/home/jcw/VAIRLtext/Text2Vec.py\", line 192, in encode\r\n> embeddings = self.encoder(padded_tokens, attention_mask=masks)[0]\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/transformers/modeling_bert.py\", line 627, in forward\r\n> head_mask=head_mask)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/transformers/modeling_bert.py\", line 348, in forward\r\n> layer_outputs = layer_module(hidden_states, attention_mask, head_mask[i])\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/transformers/modeling_bert.py\", line 326, in forward\r\n> attention_outputs = self.attention(hidden_states, attention_mask, head_mask)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/transformers/modeling_bert.py\", line 283, in forward\r\n> self_outputs = self.self(input_tensor, attention_mask, head_mask)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/transformers/modeling_bert.py\", line 202, in forward\r\n> mixed_query_layer = self.query(hidden_states)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/linear.py\", line 87, in forward\r\n> return F.linear(input, self.weight, self.bias)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/functional.py\", line 1374, in linear\r\n> output += bias\r\n> RuntimeError: CUDA error: device-side assert triggered\r\n> ```\r\n> \r\n> Second,\r\n> \r\n> ```\r\n> File \"/home/jcw/VAIRLtext/Text2Vec.py\", line 192, in encode\r\n> embeddings = self.encoder(padded_tokens, attention_mask=masks)[0]\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/transformers/modeling_bert.py\", line 627, in forward\r\n> head_mask=head_mask)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/transformers/modeling_bert.py\", line 348, in forward\r\n> layer_outputs = layer_module(hidden_states, attention_mask, head_mask[i])\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/transformers/modeling_bert.py\", line 326, in forward\r\n> attention_outputs = self.attention(hidden_states, attention_mask, head_mask)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/transformers/modeling_bert.py\", line 283, in forward\r\n> self_outputs = self.self(input_tensor, attention_mask, head_mask)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/transformers/modeling_bert.py\", line 202, in forward\r\n> mixed_query_layer = self.query(hidden_states)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n> result = self.forward(*input, **kwargs)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/modules/linear.py\", line 87, in forward\r\n> return F.linear(input, self.weight, self.bias)\r\n> File \"/home/jcw/tmp/anaconda3/envs/VAIRLtext/lib/python3.7/site-packages/torch/nn/functional.py\", line 1372, in linear\r\n> output = input.matmul(weight.t())\r\n> RuntimeError: CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling `cublasSgemm( handle, opa, opb, m, n, k, &alpha, a, lda, b, ldb, &beta, c, ldc)`\r\n> ```\r\n> \r\n> These two errors are coming from BERT encode :(\r\n> here is my code\r\n> \r\n> ```\r\n> class Text2Vec(object):\r\n> def __init__(self, args):\r\n> self.max_lens = args.max_lens\r\n> self.tokenizer = AutoTokenizer.from_pretrained('bert-base-multilingual-cased')\r\n> self.encoder = self.get_encoder('bert-base-multilingual-cased', args.num_layers)\r\n> self.encoder.to('cuda')\r\n> self.pad_token = self.tokenizer.convert_tokens_to_ids(self.tokenizer.pad_token)\r\n> \r\n> def get_encoder(self, encoder_type, num_layers):\r\n> model = AutoModel.from_pretrained(encoder_type)\r\n> model.eval()\r\n> model.encoder.layer = torch.nn.ModuleList([layer for layer in model.encoder.layer[:num_layers]])\r\n> return model\r\n> \r\n> def padding(self, arr, pad_token):\r\n> lens = torch.LongTensor([len(a) for a in arr])\r\n> padded = torch.ones(len(arr), self.max_lens) * pad_token\r\n> mask = torch.zeros(len(arr), self.max_lens)\r\n> for i, a in enumerate(arr):\r\n> padded[i, :lens[i]] = torch.tensor(a)\r\n> mask[i, :lens[i]] = 1\r\n> return padded, mask\r\n> \r\n> def encode(self, text):\r\n> text = sent_tokenize(text)\r\n> tokens = [self.tokenizer.encode(a, add_special_tokens=True) for a in text]\r\n> padded_tokens, masks = self.padding(tokens, self.pad_token)\r\n> \r\n> padded_tokens = padded_tokens.to('cuda')\r\n> masks = masks.to('cuda')\r\n> \r\n> self.encoder.eval()\r\n> with torch.no_grad():\r\n> embeddings = self.encoder(padded_tokens, attention_mask=masks)[0]\r\n> \r\n> embed = []\r\n> masks = list(torch.split(masks, 1, dim=0))\r\n> embeddings = list(torch.split(embeddings, 1, dim=0))\r\n> for embedding, mask in zip(embeddings, masks):\r\n> masked_embed = embedding * mask.squeeze()[:, None]\r\n> embed.append(masked_embed)\r\n> \r\n> return torch.cat(embed, dim=0)\r\n> ```\r\n> \r\n> What am I missing? please advise to me!!\r\n> \r\n> ## Environment\r\n> * OS: ubuntu 16.04\r\n> * Python version: Python 3.7.4\r\n> * PyTorch version: 1.3.0\r\n> * PyTorch Transformers version (or branch): how do I check PyTorch Transformers version?\r\n> * Using GPU ? titan-xp\r\n> * Distributed of parallel setup ? not yet\r\n> * Any other relevant information:\r\n\r\nIn order to check the Transformers version you can insert this line in your code:\r\n\r\n```\r\nimport transformers\r\nprint(transformers.__version__)\r\n```", "Hi! There was a few things to change on my side to get your code to run.\r\n\r\n1 - when doing `torch.ones(...)`, it initializes a tensor of dtype `float` while it should be of dtype `long`, i had to change the following line in Text2Vec.padding():\r\n\r\n```py\r\npadded = torch.ones(len(arr), self.max_lens, dtype=torch.int64) * pad_token\r\n```\r\n\r\n2 - it crashed on my side at first because I was on the pypi release which has a bug related to the initialization of `token_type_ids` when they're not given to the model (they're not put on GPU). I had to install from master with the following command:\r\n\r\n```\r\npip install git+https://github.com/huggingface/transformers\r\n```\r\n\r\nOnce you've done that it should work fine!" ]

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

[]

[ "> ## Can someone help me with resolving this error?\r\n> \r\n\r\nWithout any other information, it's very difficult to understand which is the problem with your Jupyter Notebook! **Which is the content of your directory?** Moreover, **provide the steps to reproduce this bug!**\r\nPlease, link your Jupyter Notebook and **provide more information** (e.g. OS, PyTorch version, TensorFlow version, Transformers version, etc.)", "@TheEdoardo93 Below are the package version details which I'm using on Windows 10 64 bit machine:\r\njupyter 1.0.0 \r\ntensorflow 2.0.0\r\ntorch 1.3.0+cpu \r\ntorchvision 0.4.1+cpu\r\ntransformers 2.1.1\r\n\r\nI have trained the model on colab GPU version, now was trying to load the pre-trained model on my local machine.\r\nLink of the notebook: https://colab.research.google.com/drive/1URz5MkjZwH621zGG_89S2KahCCBK0Se8\r\nIt's working fine on colab, but when I'm trying to resume on my local machine from loading the trained model, I'm getting this error.\r\nLet me know if any more info needed.", "Same problem here, using transformers in Tensorflow 2.0.0b0. Training is ok:\r\n\r\n```\r\nimport tensorflow as tf\r\nimport tensorflow_datasets\r\nfrom transformers import *\r\n\r\ntf.compat.v1.enable_eager_execution()\r\n\r\ntokenizer = BertTokenizer.from_pretrained('bert-base-cased')\r\nmodel = TFBertForSequenceClassification.from_pretrained('bert-base-cased')\r\ndata = tensorflow_datasets.load('glue/mrpc')\r\n\r\ntrain_dataset = glue_convert_examples_to_features(data['train'], tokenizer, max_length=128, task='mrpc')\r\nvalid_dataset = glue_convert_examples_to_features(data['validation'], tokenizer, max_length=128, task='mrpc')\r\ntrain_dataset = train_dataset.shuffle(100).batch(32).repeat(2)\r\nvalid_dataset = valid_dataset.batch(64)\r\n\r\noptimizer = tf.keras.optimizers.Adam(learning_rate=3e-5, epsilon=1e-08, clipnorm=1.0)\r\nloss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)\r\nmetric = tf.keras.metrics.SparseCategoricalAccuracy('accuracy')\r\nmodel.compile(optimizer=optimizer, loss=loss, metrics=[metric])\r\n\r\nhistory = model.fit(train_dataset, epochs=2, steps_per_epoch=115,\r\n validation_data=valid_dataset, validation_steps=7)\r\n\r\nmodel.save_pretrained('/home/rubens_gmail_com/tf')\r\n\r\nTrain for 115 steps, validate for 7 steps\r\nEpoch 1/2\r\n115/115 [==============================] - 57s 494ms/step - loss: 0.5496 - accuracy: 0.7331 - val_loss: 0.4226 - val_accuracy: 0.8235\r\nEpoch 2/2\r\n115/115 [==============================] - 34s 291ms/step - loss: 0.2939 - accuracy: 0.8817 - val_loss: 0.4039 - val_accuracy: 0.8456\r\n````\r\nBut this line of code give the following error:\r\n\r\n```\r\npytorch_model = BertForSequenceClassification.from_pretrained('/home/rubens_gmail_com/tf', from_tf=True,num_labels = 8)\r\n\r\nNameError Traceback (most recent call last)\r\n<ipython-input-19-ebb1b098c156> in <module>\r\n----> 1 pytorch_model = BertForSequenceClassification.from_pretrained('/home/rubensvectomobile_gmail_com/tf', from_tf=True,num_labels = 8)\r\n\r\n~/.local/lib/python3.7/site-packages/transformers/modeling_utils.py in from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)\r\n 357 try:\r\n 358 from transformers import load_tf2_checkpoint_in_pytorch_model\r\n--> 359 model = load_tf2_checkpoint_in_pytorch_model(model, resolved_archive_file, allow_missing_keys=True)\r\n 360 except ImportError as e:\r\n 361 logger.error(\"Loading a TensorFlow model in PyTorch, requires both PyTorch and TensorFlow to be installed. Please see \"\r\n\r\n~/.local/lib/python3.7/site-packages/transformers/modeling_tf_pytorch_utils.py in load_tf2_checkpoint_in_pytorch_model(pt_model, tf_checkpoint_path, tf_inputs, allow_missing_keys)\r\n 199 \r\n 200 if tf_inputs is None:\r\n--> 201 tf_inputs = tf.constant(DUMMY_INPUTS)\r\n 202 \r\n 203 if tf_inputs is not None:\r\n\r\nNameError: name 'DUMMY_INPUTS' is not defined\r\n```\r\nPyTorch, Tensorflow, transformers are up to date, as well as nvidia-ml-py3 and fast.ai. I'm on GCP using Debian, Anaconda environment w/ 8 x V100.\r\n\r\n", "Have same issue following the examples. Everything most recent version. I am on a MacBook Pro. Virtualenv environment. Python3. ", "I had the same issue. I think it was fixed with this commit - https://github.com/huggingface/transformers/pull/1509/commits/099358675899f759110ad8ccecc22c2fab9b1888.\r\n\r\nPerhaps try to re-install from source as @LysandreJik has mentioned in https://github.com/huggingface/transformers/issues/1532, or just follow the file path in the error and manually make the changes yourself. Works for me now. :)\r\n\r\n```\r\n201 - tf_inputs = tf.constant(DUMMY_INPUTS)\r\n + tf_inputs = tf_model.dummy_inputs\r\n```", "As I mentioned here [issue](https://github.com/huggingface/transformers/issues/1810#issuecomment-553108898) I was able to make a workaround, as I'm using GCP:\r\n\r\nI installed `transformers` using:\r\n\r\n```\r\nconda install -c conda-forge transformers\r\n```\r\n\r\nIn Python 3.7.4, then I added `DUMMY_INPUTS = [[7, 6, 0, 0, 1], [1, 2, 3, 0, 0], [0, 0, 0, 4, 5]]` after variable `logger` in `/opt/anaconda3/lib/python3.7/site-packages/transformers/modeling_tf_utils.py`, because it was missing.\r\n\r\nIt was interesting, because in GCP, transformers were not showing in `python3` , only in `sudo python3`", "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 } ]

[]

[ "By reading the code, I find the BertModel have the class \"BertForTokenClassification\", but XLNetModel\r\ndon't have the class \"XLNetForTokenClassification\". Why? Can someone explain this?", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "I have tried to fine-tune xlnet on squad1.0 and squad2.0, although it's a few points lower than original results, but it works fine. This #1803 is for fine-tuning on squad2.0, I'm not familiar with other datasets.", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[ "### Hi, \r\nAs far as I remember from my experiments with GPT2, the GPT2 blocks will work on its own. What I mean with that is that **the model will produce reasonable outputs, even if you leave out some layers from the original model.** So your idea should work. \r\nWhat I'm not quite sure about is the performance of an ONLY 1-2 layers model. I would guess the model will produce valid sentences, however the quality might be not optimal. \r\n\r\n### Another Tip: \r\nIf you want to build this model, load the original GPT2 model from the lib and delete the layers you don't want manually. This is a relatively ugly solution but its very time efficient ). Also you dont have to define you own forward function etc.", "Did you take a look at `distilgpt2`? It has 6-layer, 768-hidden, 12-heads, 82M parameters.\r\n\r\ncc @VictorSanh @LysandreJik ", "Hi! Indeed, if you're looking for a smaller GPT-2 model then DistilGPT-2 has half of the small GPT-2 layers. However, if you want to create your own GPT-2 model with an arbitrary number of layers, you can do so very easily with the configuration file:\r\n\r\n```py\r\nfrom transformers import GPT2Model, GPT2Config\r\n\r\nconfig = GPT2Config(n_layer=2)\r\nmodel = GPT2Model(config)\r\n```\r\n\r\nThis will define a model with only two layers. Please be aware that the weights will be randomly initialized if you go down this route, whereas the weights will already be trained if you use DistilGPT-2.", "Thank you guys! Really helped me a lot. I changed the GPT2Config as @LysandreJik told me and loaded pretrained weight to those layers :)\r\n" ]

[]

[]

[ "Hi, no specific advice other than computing the loss your-self outside of the model's forward method." ]

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

[]

[ "Have you tried the following lines after the call to the `evaluate` function (in the loop over checkpoints)?\r\n\r\n```python\r\ndel model\r\ntorch.cuda.empty_cache()\r\n```\r\n\r\nLet me know if this works.", "@rlouf, I changed\r\n\r\n```\r\n# Evaluation\r\n results = {}\r\n if args.do_eval and args.local_rank in [-1, 0]:\r\n checkpoints = [args.output_dir]\r\n if args.eval_all_checkpoints:\r\n checkpoints = list(os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + '/**/' + WEIGHTS_NAME, recursive=True)))\r\n logging.getLogger(\"transformers.modeling_utils\").setLevel(logging.WARN) # Reduce logging\r\n logger.info(\"Evaluate the following checkpoints: %s\", checkpoints)\r\n for checkpoint in checkpoints:\r\n global_step = checkpoint.split('-')[-1] if len(checkpoints) > 1 else \"\"\r\n prefix = checkpoint.split('/')[-1] if checkpoint.find('checkpoint') != -1 else \"\"\r\n \r\n model = model_class.from_pretrained(checkpoint)\r\n model.to(args.device)\r\n result = evaluate(args, model, tokenizer, prefix=prefix)\r\n result = dict((k + '_{}'.format(global_step), v) for k, v in result.items())\r\n results.update(result)\r\n\r\n return results\r\n```\r\nto\r\n```\r\n# Evaluation\r\n results = {}\r\n if args.do_eval and args.local_rank in [-1, 0]:\r\n checkpoints = [args.output_dir]\r\n if args.eval_all_checkpoints:\r\n checkpoints = list(os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + '/**/' + WEIGHTS_NAME, recursive=True)))\r\n logging.getLogger(\"transformers.modeling_utils\").setLevel(logging.WARN) # Reduce logging\r\n logger.info(\"Evaluate the following checkpoints: %s\", checkpoints)\r\n for checkpoint in checkpoints:\r\n global_step = checkpoint.split('-')[-1] if len(checkpoints) > 1 else \"\"\r\n prefix = checkpoint.split('/')[-1] if checkpoint.find('checkpoint') != -1 else \"\"\r\n \r\n model = model_class.from_pretrained(checkpoint)\r\n model.to(args.device)\r\n result = evaluate(args, model, tokenizer, prefix=prefix)\r\n del model\r\n torch.cuda.empty_cache()\r\n result = dict((k + '_{}'.format(global_step), v) for k, v in result.items())\r\n results.update(result)\r\n\r\n return results\r\n\r\n```\r\nwhich didn't help.\r\n\r\nSo, I looked through https://github.com/huggingface/transformers/issues/1742 and added\r\n\r\n```\r\ndel model\r\ngc.collect()\r\ntorch.cuda.empty_cache()\r\n```\r\n\r\nThis didn't work either; so, I change the scheduler to the following while leaving\r\n\r\n```\r\ndel model\r\ngc.collect()\r\ntorch.cuda.empty_cache()\r\n```\r\n\r\n```\r\ndef warmup_linear_schedule(optimizer, warmup_steps, t_total, last_epoch=-1):\r\n def lr_lambda(step):\r\n if step < warmup_steps:\r\n return float(step) / float(max(1, warmup_steps))\r\n return max(0.0, float(t_total - step) / float(max(1.0, t_total - warmup_steps)))\r\n\r\n return LambdaLR(optimizer, lr_lambda, last_epoch=-1)\r\n\r\n```\r\nThis hasn't worked either.\r\n\r\nAny ideas? It doesn't seem like I can `del model`", "Hi! Changing the scheduler should not do anything since you are in evaluation mode. This is likely not due to the model either, since `del` + `gc.collect()` should remove it from memory. Could you look at by how much the memory increases after each step?", "So, when using `roberta-large` the memory increases by ~2.25 GB after each checkpoint starting at ~7GB.", "Can you have a look at which tensors are still being tracked, using [this code](https://github.com/huggingface/transformers/issues/1742#issuecomment-550327172) after your del statement?", "My apologies for the delay in my answer. I had hoped that https://github.com/huggingface/transformers/commit/9629e2c676efab31d1f27a3f27d811e91fd05a92 would solve my issue; however, evaluation doesn't appear to be distributed. So, it is still using one card and is still ~2.25 GB after each checkpoint. Should I open up a separate issue for the distributed evaluation issue? After adding \r\n```\r\ndef print_gpu_obj():\r\n import gc\r\n GPU_count = 0\r\n Pinned_count = 0\r\n for tracked_object in gc.get_objects():\r\n if torch.is_tensor(tracked_object):\r\n if tracked_object.is_cuda:\r\n GPU_count+=1\r\n if tracked_object.is_pinned():\r\n Pinned_count+=1\r\n \r\n\r\n print(\"There are {} cuda objects\".format(GPU_count))\r\n print(\"There are {} pinned objects\".format(Pinned_count))\r\n```\r\nto the code, I get\r\n\r\nbefore `del model`\r\ncheckpoint 0\r\n```\r\nThere are 1087 cuda objects\r\nThere are 0 pinned objects\r\n```\r\ncheckpoint 1\r\n```\r\nThere are 1673 cuda objects\r\nThere are 0 pinned objects\r\n```\r\ncheckpoint 2\r\n```\r\nThere are 2259 cuda objects\r\nThere are 0 pinned objects\r\n```\r\ncheckpoint 3\r\n```\r\nThere are 2845 cuda objects\r\nThere are 0 pinned objects\r\n```\r\n\r\nafter `del model` and before `gc.collect`\r\ncheckpoint 0\r\n```\r\nThere are 984 cuda objects\r\nThere are 0 pinned objects\r\n```\r\ncheckpoint 1\r\n```\r\nThere are 1570 cuda objects\r\nThere are 0 pinned objects\r\n```\r\ncheckpoint 2\r\n```\r\nThere are 2156 cuda objects\r\nThere are 0 pinned objects\r\n```\r\ncheckpoint 3\r\n```\r\nThere are 2742 cuda objects\r\nThere are 0 pinned objects\r\n```\r\n\r\nafter `gc.collect` and before `torch.cuda.empty_cache()` and after `torch.cuda.empty_cache()` are both the same as after `del model` and before `gc.collect`.\r\n\r\nIt appears that 586 objects are added at each checkpoint, which cannot be deleted and unallocated. Any suggestions?", "This looks like some data is kept in memory after evaluation. Any way you could try to track this down?", "My apologies for the delay in getting back to this, as well as my ignorance on this subject, but how would I track this down. As in what commands or tools should I use? Again sorry for the late reply, I got stuck on another project for a bit.", "I have a similar issue. During training with run_lm_finetuning.py, the memory usage of a single GPU is 30G/32G. And after the training stage, I mean at the beginning of eval, the memory doesn't drop down and the evaluation stage is always getting OOM. If I add torch.cuda.empty_cache() before evaluation, the memory drops down to 22G, which means there are still 22GB tensors exist but I don't know where they are and how to free them. If I do the evaluation with run_lm_finetuning.py without training, everything is ok and the memory usage is about 8G.", "> I have a similar issue. During training with run_lm_finetuning.py, the memory usage of a single GPU is 30G/32G. And after the training stage, I mean at the beginning of eval, the memory doesn't drop down and the evaluation stage is always getting OOM. If I add torch.cuda.empty_cache() before evaluation, the memory drops down to 22G, which means there are still 22GB tensors exist but I don't know where they are and how to free them. If I do the evaluation with run_lm_finetuning.py without training, everything is ok and the memory usage is about 8G.\r\n\r\nPS: If I set --evaluate_during_training, there is no OOM issue.", "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" ]

[]

[]

[ "Ok, why not." ]

[]

[]

[ "BPE is a data compression algorithm, the result depends on the statistical properties of the corpus of text on which it has been trained.\r\n\r\nYou can start with this article https://leimao.github.io/blog/Byte-Pair-Encoding/ to get more of an intuition of how it works.", "> BPE is a data compression algorithm, the result depends on the statistical properties of the corpus of text on which it has been trained.\r\n> \r\n> You can start with this article https://leimao.github.io/blog/Byte-Pair-Encoding/ to get more of an intuition of how it works.\r\n\r\nThank you! That helps!" ]

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

[]

[ "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "Current commercial GPU VRAM sizes are not large enough to process the XL model. You'll need probably around 32-64 GB of RAM in order to finetune successfully. So for now, you'll have to train on a CPU with a lot of memory attached.", "How did you come to that number though? \r\n\r\nIn Thomas Wolf's medium post on training large models he mentioned gradient checkpointing for models that couldn't fit a batch onto a single gpu. Is it possible to implement something like for GPT-2 and friends?", "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" ]

[]

[]

[ "@joelb-git Can you check the lowest ```length``` at which this occurs? \r\n\r\nAlso, is this an issue that only occurs when ```past``` is used? \r\n\r\nCan you check if same thing happens when running\r\n\r\n```python examples/run_generation.py --prompt \"Who was Jim Henson ? Jim Henson was a\" --model_type gpt2 --model_name_or_path gpt2 --length 40```\r\n\r\non the unchanged ```run_generation.py``` file", "@enzoampil\r\n\r\n> @joelb-git Can you check the lowest length at which this occurs?\r\n\r\n`--length 36` is the smallest length that produces the failure.\r\n\r\n> Also, is this an issue that only occurs when past is used?\r\n\r\nYes. Also, the results when using `past` when it does not crash are\r\ndifferent from without using `past`.\r\n\r\nWithout `past`:\r\n```\r\n$ python examples/run_generation.py --prompt \"Who was Jim Henson ? Jim Henson was a\" --model_type gpt2 --model_name_or_path gpt2 --length 20\r\n man of God. Jim Henson was a man who had built his family into this nation. His\r\n```\r\n\r\nWith `past`:\r\n```\r\n$ python examples/run_generation.py --prompt \"Who was Jim Henson ? Jim Henson was a\" --model_type gpt2 --model_name_or_path gpt2 --length 20\r\n man of almostWho was Jim Henson? Jim Henson was a man of almostWho was Jim\r\n```\r\n\r\n> Can you check if same thing happens when running\r\n> \r\n> python examples/run_generation.py --prompt \"Who was Jim Henson ? Jim Henson was a\" --model_type gpt2 --model_name_or_path gpt2 --length 40\r\n> \r\n> on the unchanged run_generation.py file\r\n\r\nOn the unchanged file, this does not crash.\r\n\r\nI also note a documentation issue:\r\n\r\nhttps://huggingface.co/transformers/model_doc/gpt2.html\r\n```\r\nOutputs:\r\npast:\r\n\r\nlist of torch.FloatTensor (one for each layer) of shape\r\n(batch_size, num_heads, sequence_length, sequence_length) ...\r\n```\r\n\r\nThe shape that I observe has an extra prefixed dimension:\r\n\r\n```\r\n(2, batch_size, num_heads, sequence_length, sequence_length)\r\n```\r\n", "I have the exact same issue (although I am using distilgpt2).\r\n\r\nAfter some experimenting I found out that the generation (starting from BOS, i.e. without a prompt) always crashes when the generated sequence goes from length 44 to 45. At this point, I suspect that the hidden states in past are concatenated. For the sum of all numbers from 1 to 44 is 990. So if you add 45 again, you exceed the maximum number of positions in GPT2 (1024).\r\n\r\nThis would also explain why your results with and without past are different @joelb-git . Also it seems that the prompt (\"Who was Jim Henson? ...\") is repeated in the output when past is used. That might also be an artifact of concatenation?", "I also get a similar issue. I suspect the error has something to do with the past attention weights being added to the present - since the error is a matrix multiplication shape mismatch. \r\n\r\nThe issue may lay in the modelling_gpt2.py starting on line 181.\r\n``` \r\nx = self.c_attn(x)\r\nquery, key, value = x.split(self.split_size, dim=2)\r\nquery = self.split_heads(query)\r\nkey = self.split_heads(key, k=True)\r\nvalue = self.split_heads(value)\r\nif layer_past is not None:\r\n past_key, past_value = layer_past[0].transpose(-2, -1), layer_past[1] # transpose back cf below\r\n key = torch.cat((past_key, key), dim=-1)\r\n value = torch.cat((past_value, value), dim=-2)\r\npresent = torch.stack((key.transpose(-2, -1), value))\r\n**attn_outputs = self._attn(query, key, value, attention_mask, head_mask)**\r\n``` \r\n\r\nWhere the error occurs in the method I have wrapped with **.\r\nGoing into the method, the error occurs in the line alsop marked with **:\r\n\r\n```\r\ndef _attn(self, q, k, v, attention_mask=None, head_mask=None):\r\n w = torch.matmul(q, k)\r\n if self.scale:\r\n w = w / math.sqrt(v.size(-1))\r\n nd, ns = w.size(-2), w.size(-1)\r\n b = self.bias[:, :, ns-nd:ns, :ns]\r\n **w = w * b - 1e4 * (1 - b)**\r\n```\r\nThe error occurs because this code tried to multiply **'w'** of shape ([1, 12, 1024, 2048]) by **'b'** of shape ([1, 1, 0, 1024]). However, I am unsure what are the appropiate shapes thus cannot provide the solution.\r\n\r\nHope this helps.\r\n\r\n", "@adigoryl This is definitely related.\r\nOne can run this very simple script to reproduce the error:\r\n```\r\nimport torch\r\nfrom transformers import GPT2Tokenizer, GPT2LMHeadModel\r\n\r\ntokenizer = GPT2Tokenizer.from_pretrained('distilgpt2')\r\nmodel = GPT2LMHeadModel.from_pretrained('distilgpt2')\r\ndevice = 'cuda:0'\r\nmodel.to(device)\r\n\r\ninput_ids = torch.tensor(tokenizer.encode(\r\n \"<|endoftext|>\")).unsqueeze(0).to(device)\r\ninputs = {'input_ids': input_ids}\r\n\r\nwith torch.no_grad():\r\n past = None\r\n for i in range(45):\r\n print(i)\r\n logits, past = model(**inputs, past=past)\r\n logits = logits[0, -1]\r\n\r\n next_token = logits.argmax()\r\n input_ids = torch.cat([input_ids, next_token.view(1, 1)], dim=1)\r\n\r\n inputs = {'input_ids': input_ids}\r\nprint(tokenizer.decode(input_ids[0].tolist()))\r\n```\r\n\r\nIt works fine for `range(44)` but at `range(45)` it crashes with the error @adigoryl describes:\r\n```\r\nTraceback (most recent call last):\r\n File \"test.py\", line 17, in <module>\r\n logits, past = model(**inputs, past=past)\r\n File \"/mounts/Users/martin/.local/lib/python3.6/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n result = self.forward(*input, **kwargs)\r\n File \"/mounts/Users/martin/.local/lib/python3.6/site-packages/transformers/modeling_gpt2.py\", line 566, in forward\r\n head_mask=head_mask)\r\n File \"/mounts/Users/martin/.local/lib/python3.6/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n result = self.forward(*input, **kwargs)\r\n File \"/mounts/Users/martin/.local/lib/python3.6/site-packages/transformers/modeling_gpt2.py\", line 470, in forward\r\n head_mask=head_mask[i])\r\n File \"/mounts/Users/martin/.local/lib/python3.6/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n result = self.forward(*input, **kwargs)\r\n File \"/mounts/Users/martin/.local/lib/python3.6/site-packages/transformers/modeling_gpt2.py\", line 250, in forward\r\n head_mask=head_mask)\r\n File \"/mounts/Users/martin/.local/lib/python3.6/site-packages/torch/nn/modules/module.py\", line 541, in __call__\r\n result = self.forward(*input, **kwargs)\r\n File \"/mounts/Users/martin/.local/lib/python3.6/site-packages/transformers/modeling_gpt2.py\", line 211, in forward\r\n attn_outputs = self._attn(query, key, value, attention_mask, head_mask)\r\n File \"/mounts/Users/martin/.local/lib/python3.6/site-packages/transformers/modeling_gpt2.py\", line 154, in _attn\r\n w = w * b - 1e4 * (1 - b)\r\nRuntimeError: The size of tensor a (1035) must match the size of tensor b (1024) at non-singleton dimension 3\r\n```\r\n\r\n`w` having length 1035 (which is the sum of all numbers from 1 to 45) also made me think that there is some concatenation going on that shouldn't be happening.", "As I suspect a wrong idea about the interface and thus some concatenation going on that shouldn't be happening, I modified my example script in the following way:\r\n```\r\nimport torch\r\nfrom transformers import GPT2Tokenizer, GPT2LMHeadModel\r\n\r\ntokenizer = GPT2Tokenizer.from_pretrained('distilgpt2')\r\nmodel = GPT2LMHeadModel.from_pretrained('distilgpt2')\r\ndevice = 'cuda:0'\r\nmodel.to(device)\r\n\r\ninput_ids = torch.tensor(tokenizer.encode(\r\n \"<|endoftext|>\")).unsqueeze(0).to(device)\r\ninputs = {'input_ids': input_ids}\r\n\r\nwith torch.no_grad():\r\n past = None\r\n for i in range(45):\r\n print(i)\r\n logits, past = model(**inputs, past=past)\r\n logits = logits[0, -1]\r\n\r\n next_token = logits.argmax().view(1, 1)\r\n input_ids = torch.cat([input_ids, next_token], dim=1)\r\n\r\n inputs = {'input_ids': next_token}\r\nprint(tokenizer.decode(input_ids[0].tolist()))\r\n```\r\nSo only `next_token` is given as an input to the gpt2 model. This permits me to generate sequences of arbitrary length while using the `past` parameter.\r\nThe only thing I would still really like to know is where this \"bug\" came from in the first place: do we as users have a wrong idea about the interface and should we use it in a way where `past` and `input_ids` do not overlap? Or were we right about the interface from the start and there really *is* a bug?\r\n\r\nSurprisingly, both versions give me natural results when presented with the prompt `\"<|endoftext|>Who was Jim Henson? Jim Henson was a\"`:\r\nWithout modification: `Who was Jim Henson? Jim Henson was a great father and a great father.`\r\nWith my modification: `Who was Jim Henson? Jim Henson was a great actor, and he was a great actor. He was a great actor. He was a great`\r\n\r\nThe same when using `gpt2` instead of `distilgpt2`:\r\nWithout modification: `Who was Jim Henson? Jim Henson was a writer and director of the film \"The Last Man on Earth\" and the sequel to the film \"`\r\nWith modification: `Who was Jim Henson? Jim Henson was a writer, director, producer, and producer of the hit television series \"The Henson Report.\" He`\r\n\r\nSo how are we supposed to use it?", "Hi, when the past is computed for certain tokens these tokens should not be passed as input ids. @mnschmit your last example displays correct usage of the `past` argument. I've clarified the documentation in d409aca, please let me know if it should be clarified further.\r\n\r\nThank you for looking into it.\r\n", "@LysandreJik great, thank you!\r\nWith the extended documentation, it would be clear to me now but, additionally, a usage example might be useful for new users. Feel free to put my short example (or parts of it) at some appropriate place if you like!", "You're right, having a snippet showing usage would definitely improve user experience on this point. I'll update the documentation later today, thank you!", "I've added a section in the [`quickstart` section of the documentation](https://huggingface.co/transformers/quickstart.html#using-the-past). Feel free to re-open if it's still not clear enough.", "That's a great new part of the documentation. Thank you!", "Yes, indeed, thanks for that doc change! And thanks for all who helped diagnose.\r\n\r\nI applied the suggestions to my local version of `run_generation.py`.\r\nIt no longer crashes for large generation lengths, and generation is\r\nindeed now faster (about 1.7 times faster in my tests).", "Here is the relevant portion of the modified `run_generation.py` script, changed for gpt2 only:\r\n\r\n```\r\ndef sample_sequence(model, length, context, num_samples=1, temperature=1, top_k=0, top_p=0.0, repetition_penalty=1.0,\r\n device='cpu'):\r\n context = torch.tensor(context, dtype=torch.long, device=device)\r\n context = context.unsqueeze(0).repeat(num_samples, 1)\r\n generated = context.clone().detach()\r\n past = None\r\n with torch.no_grad():\r\n for _ in range(length):\r\n output, past = model(context, past=past)\r\n next_token_logits = output[:, -1, :]\r\n filtered_logits = top_k_top_p_filtering(next_token_logits, top_k=top_k, top_p=top_p)\r\n next_token = torch.multinomial(F.softmax(filtered_logits, dim=-1), num_samples=1)\r\n generated = torch.cat((generated, next_token), dim=1)\r\n # when using `past`, the context for the next call should be only\r\n # the previous token\r\n context = next_token\r\n return generated\r\n```", "I encouter similar issue, and find the problem is cased by input_ids and length, beacuse wrongly use of special token [CLS] when tokenize input_ids ", "I am trying a BERT-GPT2(not BART) architecture in which I am passing BERT's hidden states of 12 layers to past. After reading doc's I find that past should be of different shape as compared to shape of hidden states from BERT. Since I need BERT's encoding layer for encoding and GPT2 decoding function, tell me how should I manage it." ]

[]

[]

[ "Hi, GPT-2 XL was added yesterday with commit d7d36181fdefdabadc53adf51bed4a2680f5880a", "Yes, sorry for opening a new issue! I didn't see this commit. I'll close the issue! Thank you for your time!" ]

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

[]

[ "You're probably using the command:\r\n```\r\npython run_generation.py --model_type=gpt2 --model_name_or_path=gpt2\r\n```\r\n\r\nto use GPT2-XL for generation you would change the last argument to `gpt2-xl`:\r\n```\r\npython run_generation.py --model_type=gpt2 --model_name_or_path=gpt2-xl\r\n```", "ah. its because the latest version isnt available on pypi yet. and i pip installed it. 2.1.1 is the version I've got installed.\r\n\r\nIs there any way to force pip to install the latest version that was commited 21 hours ago.", "Ah yes we're yet to release a pypi version with GPT2-XL. To install from source you would have to do:\r\n\r\n```\r\npip install git+https://github.com/huggingface/transformers\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", " pip install transformers==3.4.0\r\nthis version of transforrmers solved my problem: OSError: Model name 'ckiplab/gpt2-base-chinese' was not found in model name list (gpt2, gpt2-medium, gpt2-large, gpt2-xl, distilgpt2). We assumed 'ckiplab/gpt2-base-chinese' was a path or url to a configuration file named config.json or a directory containing such a file but couldn't find any such file at this path or url.\r\n" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1746?src=pr&el=h1) Report\n> Merging [#1746](https://codecov.io/gh/huggingface/transformers/pull/1746?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/f88c104d8f79e78a98c8ce6c1f4a78db73142eab?src=pr&el=desc) will **increase** coverage by `<.01%`.\n> The diff coverage is `91.3%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/1746?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #1746 +/- ##\n==========================================\n+ Coverage 84.03% 84.03% +<.01% \n==========================================\n Files 94 94 \n Lines 14021 14032 +11 \n==========================================\n+ Hits 11782 11792 +10 \n- Misses 2239 2240 +1\n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/1746?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [transformers/tests/conftest.py](https://codecov.io/gh/huggingface/transformers/pull/1746/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL2NvbmZ0ZXN0LnB5) | `93.33% <100%> (+3.33%)` | :arrow_up: |\n| [transformers/modeling\\_distilbert.py](https://codecov.io/gh/huggingface/transformers/pull/1746/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2Rpc3RpbGJlcnQucHk=) | `95.84% <100%> (+0.01%)` | :arrow_up: |\n| [transformers/modeling\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/1746/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL21vZGVsaW5nX2JlcnQucHk=) | `87.77% <75%> (ø)` | :arrow_up: |\n| [transformers/tests/modeling\\_bert\\_test.py](https://codecov.io/gh/huggingface/transformers/pull/1746/diff?src=pr&el=tree#diff-dHJhbnNmb3JtZXJzL3Rlc3RzL21vZGVsaW5nX2JlcnRfdGVzdC5weQ==) | `96.5% <91.66%> (-0.43%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/1746?src=pr&el=continue).\n> **Legend** - [Click here to learn more](https://docs.codecov.io/docs/codecov-delta)\n> `Δ = absolute <relative> (impact)`, `ø = not affected`, `? = missing data`\n> Powered by [Codecov](https://codecov.io/gh/huggingface/transformers/pull/1746?src=pr&el=footer). Last update [f88c104...fd29a30](https://codecov.io/gh/huggingface/transformers/pull/1746?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

[]

[]

[ "No, that's how the model works.\r\n\r\nShort of creating your own model (with you own custom tokenizer) there's no way to do what you want to do.", "Okkkkkk, it is unreasonable ...." ]