Datasets:

---

DanielPFlorian

/

Transformers-Github-Issues

like 0

[]

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

[ "Hey @moinnadeem,\r\n\r\nSorry to answer so late...yeah this will require some work I think! Will put it in the generate() project, but not sure when we manage to take a deeper look into this. Feel free to open a PR and tag me if you have some good ideas :-) ", "Just added this to `example/seq2seq/`: https://github.com/huggingface/transformers/blob/master/examples/seq2seq/README.md#multi-gpu-evalulation", "+1 on getting `generate()` to run when using nn.dataparallel for evaluation\r\n\r\nIs the current finetune.py script (which I think uses DDP) running generate on all GPUs or just on one device?", "it uses all devices.", "Regarding getting generate() to run using nn.dataparallel, it actually probably isn't worth building the functionality. It seems that dataparallel (e.g. with 2 GPUs) is rarely faster (and actually sometimes quite a bit slower) than running with a single GPU, even when the multi-gpu is on the same node.", "Hello,\r\n\r\nI want to use generate function with single GPU. Specifically, I fine tuned a GPT-2 model (on GPU) and subsequently, I want to generate text with it. \r\n\r\nWhen I run this \r\n\r\n```\r\ninput_ids.to(device)\r\nsample_output = model.generate(\r\n input_ids, \r\n do_sample=True, \r\n max_length=150,\r\n top_k=50,\r\n top_p=0.92\r\n)\r\n```\r\nI get this error\r\n\r\n`RuntimeError: Input, output and indices must be on the current device`\r\n\r\nWhen i move the model and the input `.to('cpu')`, it works.", "@contribcode this looks like another issue, do you mind opening a new issue with the issue template filled-in? Thank you.", "You are right @LysandreJik, I will open a new issue.", "This issue has been stale for 1 month.", "Gently pinging @stas00 here - is this possible now thanks to your recent PR in `generate()`? ", "1. The work I did in `generate`'s search functions is to make those work under deepspeed zero-3+ regime, where all gpus must work in sync to complete, even if some of them finished their sequence early - it uses all gpus because the params are sharded across all gpus and thus all gpus contribute their part to make it happen. \r\n\r\n2. But otherwise, the current design is that we already use all gpus. Is it not the case? We just don't do anything with the results from all but rank 0 process.\r\n\r\n3. In current examples we for some reason save the generated tokens only from rank0: e.g. in `translation_run.py`:\r\n\r\nhttps://github.com/huggingface/transformers/blob/50f4539b8201b26b18085260bf801cdeadfa6640/examples/pytorch/translation/run_translation.py#L564\r\n\r\nI guess it's easier than to try to somehow write them out in non-interleaved way - probably could to add `flock` and append them all to the same file if needed.\r\n\r\n4. metrics we calculate on all gpus, but only save metrics from rank0 process\r\n\r\nhttps://github.com/huggingface/transformers/blob/5c00918681d6b4027701eb46cea8f795da0d4064/src/transformers/trainer_pt_utils.py#L924\r\n\r\nIf I missed something please clarify what is not working in `generate` under multiple gpus or what is the desired functionality?", "This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.\n\nPlease note that issues that do not follow the [contributing guidelines](https://github.com/huggingface/transformers/blob/master/CONTRIBUTING.md) are likely to be ignored.", "Also curious about this", "We should definitely enable `generate()` on multiple GPUs - did anybody give it a try with DDP? ", "See: https://github.com/huggingface/transformers/issues/6821#issuecomment-825770020", "> Hi!\r\n> \r\n> How would I run generation on multiple GPUs at the same time? Running model.generate on a DataParallel layer isn't possible, and model.module.generate run on a single GPU.\r\n> \r\n> Any advice would be appreciated!\r\n\r\nHello,\r\nDid you resolve this issue? If you remember can you share me the solution for it?\r\n\r\nThanks in advance", "As I already replied earlier `generate` works on multiple gpus including `deepspeed`", "> \r\n@kkavyashankar0009 @JulesGM \r\n\r\nHi, it seems that model.generatre() does not support DP, but it can be used in DDP. Here is some example code:\r\n\r\n\r\n```\r\nimport torch.multiprocessing as mp\r\nimport torch.distributed as dist\r\nimport argparse\r\nimport torch\r\nfrom transformers import get_linear_schedule_with_warmup, BartConfig, BartTokenizer, BartForConditionalGeneration\r\n\r\nparser = argparse.ArgumentParser()\r\nparser.add_argument('--nodes', type=int, default=1) # how many nodes (machines) you have\r\nparser.add_argument('--gpus', type=int, default=-1, help='num gpus per node')\r\nparser.add_argument('--nr', type=int, default=0, help='ranking within the nodes')\r\nargs = parser.parse_args()\r\ntokenizer = BartTokenizer.from_pretrained(args.tokenizer_name)\r\n\r\ndef test_model_generation(local_gpu_rank, args):\r\n set_seed(args.seed)\r\n args.rank = args.nr * args.gpus + local_gpu_rank # compute the rank of the current GPU\r\n dist.init_process_group(backend=\"nccl\", init_method=\"env://\", world_size=args.world_size, rank=args.rank)\r\n\r\n test_data = \"../data/\" + args.dataset + \"/test.txt\"\r\n print(\"Processing data: \" + test_data, flush=True)\r\n config = BartConfig.from_pretrained(args.config_name)\r\n bart_ctx = BartForConditionalGeneration.from_pretrained(args.model_name_or_path, config=config)\r\n bart_rep = BartForConditionalGeneration.from_pretrained(args.model_name_or_path, config=config)\r\n bart_ctx.resize_token_embeddings(len(tokenizer))\r\n bart_rep.resize_token_embeddings(len(tokenizer))\r\n\r\n model = MyModel(bart, config)\r\n model_state_dict = torch.load(\"../output/model/gen.ddp.pt\") \r\n model.load_state_dict(model_state_dict, strict=True) # load model\r\n torch.cuda.set_device(local_gpu_rank) \r\n args.device = torch.device(\"cuda\", local_gpu_rank)\r\n model.to(args.device) # move the model to GPU\r\n # model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[local_gpu_rank], find_unused_parameters=True)\r\n model.eval()\r\n\r\n test_dataset = FileDataset(test_data, tokenizer, max_context_len=args.max_ctx_len, max_response_len=args.max_rep_len, dataset=args.dataset)\r\n test_sampler = torch.utils.data.distributed.DistributedSampler(test_dataset, shuffle=False)\r\n test_dataloader = DataLoader(test_dataset, batch_size=args.test_batch_size, num_workers=2, sampler=test_sampler) # the batch size on each GPU\r\n\r\n if args.rank == 0:\r\n count = 0\r\n fw = open(\"../output/test.responses.txt\", \"w\", encoding=\"utf-8\") # open file for writting result\r\n with torch.no_grad():\r\n test_sampler.set_epoch(0) # keep all data the same on all GPUs, it is usually used in training, I'm not sure if it is necessary in inference\r\n for test_data in test_dataloader:\r\n for key in test_data.keys():\r\n test_data[key] = test_data[key].to(args.device)\r\n outputs = model.bart_model.generate(\r\n input_ids=test_data[\"input_ids\"],\r\n attention_mask=test_data[\"attention_mask\"],\r\n max_length=args.max_rep_len,\r\n no_repeat_ngram_size=3,\r\n num_beams=10,\r\n ) # my model contains a BART model in self.bart_model，so I use model.module.bart_model to get it\r\n if outputs.size(1) < args.max_rep_len: # need padding because the lengths from different GPUs may be different\r\n batch_pred_padding = torch.ones((outputs.size(0), args.max_rep_len - outputs.size(1)), dtype=outputs.dtype).cuda() # use the padding token of BART, and its token id is 1. Be careful with the data type.\r\n outputs = torch.cat([outputs, batch_pred_padding], dim=1)\r\n batch_pred = [torch.zeros_like(outputs, dtype=outputs.dtype).cuda() for _ in range(args.world_size)] # initialized a list for collecting tensors from all GPUs. Be careful with the data type.\r\n dist.all_gather(batch_pred, outputs) # collect data\r\n batch_pred = torch.stack(batch_pred, dim=1) # use stack, take care of the dimension\r\n batch_pred = batch_pred.reshape(-1, args.max_rep_len)\r\n if args.rank == 0:\r\n batch_out_sentences = tokenizer.batch_decode(batch_pred, skip_special_tokens=True, clean_up_tokenization_spaces=False) # decode the token id to token\r\n for r in batch_out_sentences:\r\n fw.write(r + \"\\n\")\r\n fw.flush()\r\n count += len(batch_out_sentences)\r\n print(count)\r\n if args.rank == 0:\r\n fw.close()\r\n \r\nif __name__ == \"__main__\":\r\n if args.gpus < 0:\r\n args.gpus = torch.cuda.device_count()\r\n args.world_size = args.nodes * args.gpus\r\n os.environ['MASTER_ADDR']='localhost'\r\n os.environ['MASTER_PORT']='8888'\r\n mp.spawn(test_model_generation, nprocs=args.gpus, args=(args, ))\r\n\r\n```\r\n\r\n\r\n\r\n\r\nBe careful with the data order. It is better to add IDs and write them to the file, which can be used for checking the order.", "@DaoD \r\n\r\nThanks a lot for this example! If I understand correctly, I think you don't need to use `torch.nn.parallel.DistributedDataParallel` wrapper on the model if only running inference though, since we don't care about gradients. I found that only using the distributed data sampler and leaving the model as is (without the DDP wrapper) led to better memory usage. Haven't looked into why, but let's me use larger batch sizes just FYI.", "@bkleiner2 Thanks for your information! I just want to use multiple GPUs for inference, so I don't how it works if DDP is not used. ", "Yeah I'm just saying for inference only I don't think you need this line:\r\n\r\n`model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[local_gpu_rank], find_unused_parameters=True)`\r\n\r\nbecause we don't need to aggregate gradients. Once you've done this:\r\n\r\n` model.to(args.device) # move the model to GPU` you've already copied the model to each device right? Because there is a separate process running for each device. So you can get the data parallelism across devices by simply using `test_sampler = torch.utils.data.distributed.DistributedSampler(test_dataset, shuffle=False)`, as far as I can tell.", "@bkleiner2 Oh, I got it! Thanks for your explanation!", "I think the real problem we are facing is this error when we try to called `model.generate`\r\n``` bash\r\nAttributeError: 'DistributedDataParallel' object has no attribute 'generate'\r\n```\r\n\r\nIndeed, it can be solved by distributed sampler or using the huggingface trainer class. \r\n\r\nBut it could be quite tricky if we don't use them and write our own trainer. Even we have to be careful in using the distributed sampler.", "remove the outside wrapping to get to the `transformers`'s model and it should work, \r\n\r\n`model.module.generate()`\r\n\r\nYou can use the helper that deals with arbitrary number of wrappers.\r\nhttps://github.com/huggingface/transformers/blob/d90a36d192e2981a41122c30a765c63158dd0557/src/transformers/modeling_utils.py#L3027-L3038\r\n\r\n", "> As I already replied earlier `generate` works on multiple gpus including `deepspeed`\r\n\r\n\r\n\r\n> remove the outside wrapping to get to the `transformers`'s model and it should work,\r\n> \r\n> `model.module.generate()`\r\n> \r\n> You can use the helper that deals with arbitrary number of wrappers.\r\n> \r\n> https://github.com/huggingface/transformers/blob/d90a36d192e2981a41122c30a765c63158dd0557/src/transformers/modeling_utils.py#L3027-L3038\r\n\r\nI tried unwraping the model as \"model.module.generate\" but I guess there's an issue in the generators it's all special tokens' index as decoding returns empty strings. \r\n\r\ncould you please clarify what model.module is exactly doing?", "@HebaGamalElDin I think your problem is not related to this issue. Or we might need more context here, in your case.\r\n", "> could you please clarify what model.module is exactly do?\r\n\r\nWrappers like DDP, Deepspeed, and others hide the original model inside their objects - usually under `model.module`, so you can get from the wrapped object to the original model using `unwrap_model` which will handle multiple wrappers. If it's just one you can just access the original PreTrainedModel HF subclass model with `model.module`.\r\n\r\nThe wrappers have no `generate` method, only the PreTrainedModel subclasses have it. That's why if you want to call `generate` you must call it on the HF model and not its wrappers.\r\n\r\nIf you have issues I suggest you first remove DDP and debug your issue on a single GPU, once working it'd most likely work under DDP..\r\n\r\nAs @allanj suggested your issue most likely has nothing to do with unwrapping, so anything is possible if you write your own code. Hence I suggest to sort it out on 1-gpu first, then try 1+.", "@stas00 thank you, what I'm wondering is which model version exactly does model.module access or which model version in which GPU exactly? \r\n\r\n**It was working already in 1 GPU, since I switched to ml.p3.16xlarge instance It makes this behavior.**\r\n\r\nHere's the full training process:\r\n```\r\n\r\nimport os\r\nimport torch\r\nimport pandas as pd\r\nimport random\r\nimport math\r\nfrom copy import deepcopy\r\nfrom tqdm import tqdm\r\nimport os\r\nimport re\r\nimport shutil\r\nimport tarfile, zipfile\r\nimport pickle,json\r\nimport numpy as np\r\nimport itertools\r\nfrom PIL import Image\r\nimport PIL.ImageOps\r\nimport cv2\r\nfrom torch.utils.data import DataLoader\r\nfrom transformers import AdamW, TrOCRProcessor, VisionEncoderDecoderModel, get_scheduler\r\nfrom Data_pipeline import Context, HCRDataset, OCRDataLoad\r\nfrom Validation_Metrics import getWordLevelError, getCharacterLevelError\r\n\r\n\r\nfrom datasets import load_metric\r\ncer_metric = load_metric(\"cer\")\r\n\r\n# SageMaker data parallel: Import the library PyTorch API\r\n#round(random.uniform(),2)\r\nimport smdistributed.dataparallel.torch.torch_smddp\r\n\r\n# SageMaker data parallel: Import PyTorch's distributed API\r\nimport torch.distributed as dist\r\nfrom torch.nn.parallel import DistributedDataParallel as DDP\r\n\r\n# SageMaker data parallel: Initialize the process group\r\ndist.init_process_group(backend='smddp')\r\n\r\n# LOAD MODEL\r\ndef load_model() -> VisionEncoderDecoderModel:\r\n model: VisionEncoderDecoderModel = VisionEncoderDecoderModel.from_pretrained('gagan3012/ArOCRv4')\r\n return model.cuda()\r\n\r\n# SETUP MODEL CONFIGUATIONS\r\ndef init_model_for_training(model: VisionEncoderDecoderModel, processor: TrOCRProcessor):\r\n model.config.decoder_start_token_id = processor.tokenizer.cls_token_id\r\n model.config.pad_token_id = processor.tokenizer.pad_token_id\r\n model.config.vocab_size = model.config.decoder.vocab_size\r\n model.config.bos_token_id = processor.tokenizer.bos_token_id\r\n model.config.max_length = 162\r\n model.config.decoder.is_decoder = True\r\n model.config.decoder.add_cross_attention = True\r\n torch.cuda.manual_seed_all(42)\r\n\r\n \r\n\r\ndef compute_cer(processor, pred_ids, label_ids):\r\n pred_str = processor.batch_decode(pred_ids, skip_special_tokens=True)\r\n label_ids[label_ids == -100] = processor.tokenizer.pad_token_id\r\n label_str = processor.batch_decode(label_ids, skip_special_tokens=True)\r\n\r\n cer = cer_metric.compute(predictions=pred_str, references=label_str)\r\n return cer\r\n\r\n\r\n# LOAD PRE_PROCESSOR\r\ndef load_processor() -> TrOCRProcessor:\r\n return TrOCRProcessor.from_pretrained('gagan3012/ArOCRv4')\r\n\r\ndef train(context: Context, train_epochs, learning_rate):\r\n model = context.model\r\n optimizer = AdamW(model.parameters(), lr=learning_rate)\r\n \r\n num_training_steps = train_epochs * len(context.train_dataloader)\r\n lr_scheduler = get_scheduler(\"linear\", optimizer=optimizer, num_warmup_steps=0, num_training_steps=num_training_steps)\r\n \r\n train_loss = 0.0\r\n min_cer = 1.0\r\n min_train_loss = 1.0\r\n for epoch in range(train_epochs):\r\n model.train() \r\n for j, batch in enumerate(context.train_dataloader):\r\n inputs: torch.Tensor = batch[\"input\"].cuda(non_blocking=True)\r\n labels: torch.Tensor = batch[\"label\"].cuda(non_blocking=True)\r\n #print(inputs)\r\n #print(labels)\r\n \r\n outputs = model(pixel_values=inputs, labels=labels)\r\n #print(outputs)\r\n loss = outputs.loss\r\n loss.backward()\r\n optimizer.step()\r\n lr_scheduler.step()\r\n optimizer.zero_grad(set_to_none=True)\r\n \r\n train_loss+=loss\r\n if (loss < min_train_loss) or (min_train_loss==1.0):\r\n min_train_loss = loss\r\n print(f\"Epoch {epoch}-----Loss---{train_loss/len(context.train_dataloader)}--------- min-cer: {min_train_loss}\")\r\n \r\n # evaluate\r\n #model.eval()\r\n valid_cer = 0.0\r\n with torch.no_grad():\r\n for batch in tqdm(context.val_dataloader):\r\n #print(f\"INPUT: {batch['input']} ------ LABEL: {batch['label']}\")\r\n outputs = model.module.generate(batch[\"input\"].cuda(non_blocking=True))\r\n #print(f\"OUTPUTS: {outputs}\")\r\n #print(f\"OUTPUTS on CPU: {outputs.cpu().numpy()}\")\r\n cer = compute_cer(context.processor, pred_ids=outputs.detach(), label_ids=batch[\"label\"])\r\n valid_cer += cer \r\n\r\n print(\"Validation CER:\", valid_cer / len(context.val_dataloader))\r\n\r\ndef main():\r\n batch_size = 64\r\n train_epochs = 100\r\n learning_rate = 0.0001\r\n checkpoints_path = \"checkpoints\"\r\n \r\n # SageMaker data parallel: Scale batch size by world size\r\n batch_size //= dist.get_world_size()\r\n batch_size = max(batch_size, 1)\r\n\r\n # Prepare dataset\r\n #train_dataset = torchvision.datasets.MNIST(...)\r\n \r\n processor = load_processor()\r\n (x_train,y_train),(x_valid,y_valid),(x_test,y_test) = OCRDataLoad()\r\n train_dataset = HCRDataset(x_train, y_train, processor)\r\n\r\n # SageMaker data parallel: Set num_replicas and rank in DistributedSampler\r\n train_sampler = torch.utils.data.distributed.DistributedSampler(\r\n train_dataset,\r\n num_replicas=dist.get_world_size(),\r\n rank=dist.get_rank())\r\n \r\n train_dataloader = DataLoader(train_dataset, batch_size, shuffle=False, sampler=train_sampler)\r\n \r\n val_dataset = HCRDataset(x_valid, y_valid, processor)\r\n val_sampler = torch.utils.data.distributed.DistributedSampler(\r\n val_dataset,\r\n num_replicas=dist.get_world_size(),\r\n rank=dist.get_rank())\r\n val_dataloader = DataLoader(val_dataset, batch_size, shuffle=False, sampler=val_sampler)\r\n \r\n # SageMaker data parallel: Wrap the PyTorch model with the library's DDP\r\n model = load_model()\r\n init_model_for_training(model, processor)\r\n \r\n model = DDP(model, find_unused_parameters=True)\r\n context = Context(model, processor, train_dataset, train_dataloader, val_dataset, val_dataloader)\r\n\r\n \r\n # SageMaker data parallel: Pin each GPU to a single library process.\r\n local_rank = os.environ[\"LOCAL_RANK\"] \r\n torch.cuda.set_device(int(local_rank))\r\n model.cuda(int(local_rank))\r\n train(context, train_epochs, learning_rate)\r\n\r\n # SageMaker data parallel: Save model on master node.\r\n if dist.get_rank() == 0:\r\n model.module.save_pretrained(checkpoints_path)\r\n\r\nif __name__ == '__main__':\r\n main()\r\n \r\n```\r\n\r\n@allanj @stas00 ", "In the simple case of DDP you basically have multiple GPUs and each gpu just runs its own `generate` on the unwrapped model.\r\n\r\nI see that you use SageMaker here, I don't have experience with this environment, perhaps there is something non-standard about it?\r\n\r\nI think you want to start testing with a very simple base case reducing your test to just creating DDP and doing `generate` with some hardcoded string - you can remove all the other code including training and dataloaders, etc.\r\n\r\n1. create smddp on 2 gpus\r\n2. run `generate` on that model\r\n\r\nnothing else. \r\n\r\nperhaps smddp does something different from other wrappers? Once you have a simple repro code then we can tag someone who knows SMDDP better.", "> > \r\n> \r\n> @kkavyashankar0009 @JulesGM\r\n> \r\n> Hi, it seems that model.generatre() does not support DP, but it can be used in DDP. Here is some example code:\r\n> \r\n> ```\r\n> import torch.multiprocessing as mp\r\n> import torch.distributed as dist\r\n> import argparse\r\n> import torch\r\n> from transformers import get_linear_schedule_with_warmup, BartConfig, BartTokenizer, BartForConditionalGeneration\r\n> \r\n> parser = argparse.ArgumentParser()\r\n> parser.add_argument('--nodes', type=int, default=1) # how many nodes (machines) you have\r\n> parser.add_argument('--gpus', type=int, default=-1, help='num gpus per node')\r\n> parser.add_argument('--nr', type=int, default=0, help='ranking within the nodes')\r\n> args = parser.parse_args()\r\n> tokenizer = BartTokenizer.from_pretrained(args.tokenizer_name)\r\n> \r\n> def test_model_generation(local_gpu_rank, args):\r\n> set_seed(args.seed)\r\n> args.rank = args.nr * args.gpus + local_gpu_rank # compute the rank of the current GPU\r\n> dist.init_process_group(backend=\"nccl\", init_method=\"env://\", world_size=args.world_size, rank=args.rank)\r\n> \r\n> test_data = \"../data/\" + args.dataset + \"/test.txt\"\r\n> print(\"Processing data: \" + test_data, flush=True)\r\n> config = BartConfig.from_pretrained(args.config_name)\r\n> bart_ctx = BartForConditionalGeneration.from_pretrained(args.model_name_or_path, config=config)\r\n> bart_rep = BartForConditionalGeneration.from_pretrained(args.model_name_or_path, config=config)\r\n> bart_ctx.resize_token_embeddings(len(tokenizer))\r\n> bart_rep.resize_token_embeddings(len(tokenizer))\r\n> \r\n> model = MyModel(bart, config)\r\n> model_state_dict = torch.load(\"../output/model/gen.ddp.pt\") \r\n> model.load_state_dict(model_state_dict, strict=True) # load model\r\n> torch.cuda.set_device(local_gpu_rank) \r\n> args.device = torch.device(\"cuda\", local_gpu_rank)\r\n> model.to(args.device) # move the model to GPU\r\n> # model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[local_gpu_rank], find_unused_parameters=True)\r\n> model.eval()\r\n> \r\n> test_dataset = FileDataset(test_data, tokenizer, max_context_len=args.max_ctx_len, max_response_len=args.max_rep_len, dataset=args.dataset)\r\n> test_sampler = torch.utils.data.distributed.DistributedSampler(test_dataset, shuffle=False)\r\n> test_dataloader = DataLoader(test_dataset, batch_size=args.test_batch_size, num_workers=2, sampler=test_sampler) # the batch size on each GPU\r\n> \r\n> if args.rank == 0:\r\n> count = 0\r\n> fw = open(\"../output/test.responses.txt\", \"w\", encoding=\"utf-8\") # open file for writting result\r\n> with torch.no_grad():\r\n> test_sampler.set_epoch(0) # keep all data the same on all GPUs, it is usually used in training, I'm not sure if it is necessary in inference\r\n> for test_data in test_dataloader:\r\n> for key in test_data.keys():\r\n> test_data[key] = test_data[key].to(args.device)\r\n> outputs = model.bart_model.generate(\r\n> input_ids=test_data[\"input_ids\"],\r\n> attention_mask=test_data[\"attention_mask\"],\r\n> max_length=args.max_rep_len,\r\n> no_repeat_ngram_size=3,\r\n> num_beams=10,\r\n> ) # my model contains a BART model in self.bart_model，so I use model.module.bart_model to get it\r\n> if outputs.size(1) < args.max_rep_len: # need padding because the lengths from different GPUs may be different\r\n> batch_pred_padding = torch.ones((outputs.size(0), args.max_rep_len - outputs.size(1)), dtype=outputs.dtype).cuda() # use the padding token of BART, and its token id is 1. Be careful with the data type.\r\n> outputs = torch.cat([outputs, batch_pred_padding], dim=1)\r\n> batch_pred = [torch.zeros_like(outputs, dtype=outputs.dtype).cuda() for _ in range(args.world_size)] # initialized a list for collecting tensors from all GPUs. Be careful with the data type.\r\n> dist.all_gather(batch_pred, outputs) # collect data\r\n> batch_pred = torch.stack(batch_pred, dim=1) # use stack, take care of the dimension\r\n> batch_pred = batch_pred.reshape(-1, args.max_rep_len)\r\n> if args.rank == 0:\r\n> batch_out_sentences = tokenizer.batch_decode(batch_pred, skip_special_tokens=True, clean_up_tokenization_spaces=False) # decode the token id to token\r\n> for r in batch_out_sentences:\r\n> fw.write(r + \"\\n\")\r\n> fw.flush()\r\n> count += len(batch_out_sentences)\r\n> print(count)\r\n> if args.rank == 0:\r\n> fw.close()\r\n> \r\n> if __name__ == \"__main__\":\r\n> if args.gpus < 0:\r\n> args.gpus = torch.cuda.device_count()\r\n> args.world_size = args.nodes * args.gpus\r\n> os.environ['MASTER_ADDR']='localhost'\r\n> os.environ['MASTER_PORT']='8888'\r\n> mp.spawn(test_model_generation, nprocs=args.gpus, args=(args, ))\r\n> ```\r\n> \r\n> Be careful with the data order. It is better to add IDs and write them to the file, which can be used for checking the order.\r\n\r\nMay have two errors:\r\n\r\n1. gather\r\n```python\r\ndist.all_gather(batch_pred, outputs) # collect data\r\nbatch_pred = torch.stack(batch_pred, dim=1) # use stack, take care of the dimension\r\nbatch_pred = batch_pred.reshape(-1, args.max_rep_len)\r\n```\r\nchange to\r\n```python\r\ndist.all_gather(batch_pred, outputs) # collect data\r\nbatch_pred = torch.cat(batch_pred, dim=0)\r\n```\r\n\r\n2. DistributedSampler causes more samples\r\n`batch_out_sentences` need to slice:`batch_out_sentences[:total_examples]`" ]

[ { "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" ]

[]

[]

[ "The issue was with the file structur: using the downloaded glue data in the filesystem.", "@aalok-sathe - Could you please explain how you resolved it ?. I am having the same problem with XLNET for glue(STS-B)", "I had the data placed in the wrong location, and I was giving the incorrect path." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6818?src=pr&el=h1) Report\n> Merging [#6818](https://codecov.io/gh/huggingface/transformers/pull/6818?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/5ab21b072fa2a122da930386381d23f95de06e28?el=desc) will **increase** coverage by `0.51%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6818?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6818 +/- ##\n==========================================\n+ Coverage 79.58% 80.10% +0.51% \n==========================================\n Files 157 157 \n Lines 28588 28588 \n==========================================\n+ Hits 22752 22900 +148 \n+ Misses 5836 5688 -148 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6818?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/tokenization\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/6818/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fcm9iZXJ0YS5weQ==) | `87.67% <0.00%> (-10.96%)` | :arrow_down: |\n| [src/transformers/tokenization\\_utils\\_base.py](https://codecov.io/gh/huggingface/transformers/pull/6818/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHNfYmFzZS5weQ==) | `86.31% <0.00%> (-7.19%)` | :arrow_down: |\n| [src/transformers/tokenization\\_utils\\_fast.py](https://codecov.io/gh/huggingface/transformers/pull/6818/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHNfZmFzdC5weQ==) | `92.85% <0.00%> (-1.43%)` | :arrow_down: |\n| [src/transformers/tokenization\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/6818/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fYmVydC5weQ==) | `91.07% <0.00%> (-0.45%)` | :arrow_down: |\n| [src/transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6818/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHMucHk=) | `89.45% <0.00%> (-0.40%)` | :arrow_down: |\n| [src/transformers/tokenization\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6818/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fb3BlbmFpLnB5) | `82.57% <0.00%> (+11.36%)` | :arrow_up: |\n| [src/transformers/tokenization\\_reformer.py](https://codecov.io/gh/huggingface/transformers/pull/6818/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fcmVmb3JtZXIucHk=) | `95.00% <0.00%> (+13.33%)` | :arrow_up: |\n| [src/transformers/tokenization\\_bart.py](https://codecov.io/gh/huggingface/transformers/pull/6818/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fYmFydC5weQ==) | `100.00% <0.00%> (+57.89%)` | :arrow_up: |\n| [src/transformers/tokenization\\_xlm.py](https://codecov.io/gh/huggingface/transformers/pull/6818/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25feGxtLnB5) | `82.93% <0.00%> (+66.66%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6818?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/6818?src=pr&el=footer). Last update [5ab21b0...8b59678](https://codecov.io/gh/huggingface/transformers/pull/6818?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

[]

[]

[ "@patrickvonplaten , @jplu any insights into what could be the problem?", "You should create your model into a strategy.", "> You should create your model into a strategy.\r\n\r\nAs in tf distributed strategies? but i am using a single gpu at the moment. ", "This one https://www.tensorflow.org/api_docs/python/tf/distribute/OneDeviceStrategy", "Device placement strategy works and the error is no longer there. i should point out this is not the usual way to train a model in TF. We normally do not need to place the model explicitly on a device while creating a model. ", "Is this method correct?\r\n```\r\nwith mirrored_strategy.scope():\r\n ...\r\n model.compile(...)\r\nmodel.fit(...)\r\n```\r\nThis still gives me the same error on GPT2LMHeadModel.", "@ksjae Please open a new issue with more detail of your issue." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6816?src=pr&el=h1) Report\n> Merging [#6816](https://codecov.io/gh/huggingface/transformers/pull/6816?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/367235ee52537ff7cada5e1c5c41cdd78731f092?el=desc) will **increase** coverage by `2.48%`.\n> The diff coverage is `67.85%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6816?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6816 +/- ##\n==========================================\n+ Coverage 76.27% 78.76% +2.48% \n==========================================\n Files 157 157 \n Lines 28795 28823 +28 \n==========================================\n+ Hits 21963 22701 +738 \n+ Misses 6832 6122 -710 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6816?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/testing\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6816/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90ZXN0aW5nX3V0aWxzLnB5) | `66.24% <67.85%> (+0.35%)` | :arrow_up: |\n| [src/transformers/configuration\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6816/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX29wZW5haS5weQ==) | `34.28% <0.00%> (-62.86%)` | :arrow_down: |\n| [src/transformers/tokenization\\_albert.py](https://codecov.io/gh/huggingface/transformers/pull/6816/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fYWxiZXJ0LnB5) | `28.84% <0.00%> (-58.66%)` | :arrow_down: |\n| [src/transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6816/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19vcGVuYWkucHk=) | `23.87% <0.00%> (-57.10%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_distilbert.py](https://codecov.io/gh/huggingface/transformers/pull/6816/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9kaXN0aWxiZXJ0LnB5) | `64.47% <0.00%> (-34.36%)` | :arrow_down: |\n| [src/transformers/modeling\\_t5.py](https://codecov.io/gh/huggingface/transformers/pull/6816/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190NS5weQ==) | `77.63% <0.00%> (-6.21%)` | :arrow_down: |\n| [src/transformers/tokenization\\_dpr.py](https://codecov.io/gh/huggingface/transformers/pull/6816/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fZHByLnB5) | `53.15% <0.00%> (-4.51%)` | :arrow_down: |\n| [src/transformers/configuration\\_bart.py](https://codecov.io/gh/huggingface/transformers/pull/6816/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX2JhcnQucHk=) | `90.00% <0.00%> (-4.00%)` | :arrow_down: |\n| [src/transformers/generation\\_tf\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6816/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9nZW5lcmF0aW9uX3RmX3V0aWxzLnB5) | `83.95% <0.00%> (-1.51%)` | :arrow_down: |\n| [src/transformers/modeling\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6816/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ191dGlscy5weQ==) | `87.50% <0.00%> (-0.56%)` | :arrow_down: |\n| ... and [13 more](https://codecov.io/gh/huggingface/transformers/pull/6816/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6816?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/6816?src=pr&el=footer). Last update [367235e...05ec0d0](https://codecov.io/gh/huggingface/transformers/pull/6816?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "[moved to the normal comment from the code review comment, as it might be removed]\r\n\r\n> What is the difference between this method and the previous set_verbosity?\r\n> I think we should select just one way to set the verbosity of the library.\r\n\r\n@thomwolf, I agree. Now that I added a test I can see that `set_verbosity` is an equivalent of `set_global_logging_level(prefices=[\"transformers\"])` (the proposed function).\r\n\r\nSo the main question then is this: do we want to provide a util that allows to do the setting not just for `transformers.`? or leave that to the user - sort of contrib library somewhere?\r\n\r\nThe main reason for setting a global log level not just for `transfromers`, but also for `torch`, `wandb`, etc. is to be able to quickly turn off the noise when it's interfering. And currently each of these external libraries `transformers` uses add their noise to the output. When debugging tests it's very helpful to control the noise-levels. So having a quick switch --logger-be-quiet saves a lot of time.\r\n\r\n\r\n", "I also added: \r\n- a logger setting integration test\r\n- a helper `CaptureLogger` ctx manager", "Could someone please explain why CI gets `logging.ERROR` as the default logging level, when it should be `logging.WARNING` https://github.com/stas00/transformers/blob/loglevels/src/transformers/utils/logging.py#L58 (I rebased this branch to catch that very recent change)\r\n\r\nWhen I run it on my machine, I get `logging.WARNING`. \r\n\r\nOn CI the failure is:\r\n\r\n```\r\n[gw4] linux -- Python 3.7.9 /usr/local/bin/python\r\n\r\nself = <tests.test_logging.HfArgumentParserTest testMethod=test_set_level>\r\n\r\n def test_set_level(self):\r\n logger = logging.get_logger()\r\n \r\n level_origin = logging.get_verbosity()\r\n> self.assertEqual(level_origin, logging.WARNING)\r\nE AssertionError: 40 != 30\r\n```\r\n(`logging.ERROR == 40`, `logging.WARNING == 30`)\r\n\r\n**edit**: found the culprit - it was another test not cleaning up after itself. fixed in this PR.", "Thank you all for your excellent feedback. I made changes and updated the first post to reflect the PR's current state of things.", "I'm not sure we really need to control the logging level of all libraries. Since the logging level was changed back to its initial level `WARNING`, do you feel like there are too much logs during tests?", "For Bart tests there is a repetitive warning, which I raised here: https://github.com/huggingface/transformers/issues/6652\r\n\r\nIf you run others, you will see a bunch still, e.g.:\r\n\r\n```RUN_SLOW=1 pytest -sv --disable-warnings tests/test_modeling_t5.py ```\r\n\r\n```\r\ntests/test_modeling_t5.py::T5ModelTest::test_generate_with_past_key_value_states You might want to consider setting `use_cache=True` to speed up decoding\r\nYou might want to consider setting `use_cache=True` to speed up decoding\r\nYou might want to consider setting `use_cache=True` to speed up decoding\r\nYou might want to consider setting `use_cache=True` to speed up decoding\r\n[...]\r\nDownloading: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 242M/242M [00:06<00:00, 40.1MB/s]\r\nSome weights of T5Model were not initialized from the model checkpoint at t5-small and are newly initialized: ['encoder.embed_tokens.weight', 'decoder.embed_tokens.weight']\r\nYou should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\r\nPASSED\r\n```\r\n\r\nAnd this is just one test.\r\n\r\nOf course, the other approach is to go and fix all those warnings, so that the tests that are fully under our control can be written according to the requirements the library sets and then warnings won't be there :) But see the next comment with a large dump of loggers that aren't `transformers`.\r\n\r\n----\r\n\r\nYet another alternative solution is instead of flag we add an env var, `LOG_LEVEL_GLOBAL`\r\n", "Here is some more samples of noise coming from outside `transformers` - a lot of it:\r\n\r\n```\r\ntests/test_modeling_tf_albert.py::TFAlbertModelTest::test_albert_model 2020-09-02 10:32:56.462871: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcuda.so.1\r\n2020-09-02 10:32:56.467570: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.469326: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1716] Found device 0 with properties: \r\npciBusID: 0000:01:00.0 name: GeForce GTX TITAN X computeCapability: 5.2\r\ncoreClock: 1.2155GHz coreCount: 24 deviceMemorySize: 11.93GiB deviceMemoryBandwidth: 313.37GiB/s\r\n2020-09-02 10:32:56.469400: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.470032: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1716] Found device 1 with properties: \r\npciBusID: 0000:02:00.0 name: GeForce GTX TITAN X computeCapability: 5.2\r\ncoreClock: 1.2155GHz coreCount: 24 deviceMemorySize: 11.93GiB deviceMemoryBandwidth: 313.37GiB/s\r\n2020-09-02 10:32:56.470303: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudart.so.10.1\r\n2020-09-02 10:32:56.470670: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcublas.so.10\r\n2020-09-02 10:32:56.470719: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcufft.so.10\r\n2020-09-02 10:32:56.470752: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcurand.so.10\r\n2020-09-02 10:32:56.495979: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcusolver.so.10\r\n2020-09-02 10:32:56.496076: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcusparse.so.10\r\n2020-09-02 10:32:56.594292: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudnn.so.7\r\n2020-09-02 10:32:56.594768: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.597007: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.599207: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.601306: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.603994: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1858] Adding visible gpu devices: 0, 1\r\n2020-09-02 10:32:56.612943: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN)to use the following CPU instructions in performance-critical operations: AVX2 FMA\r\nTo enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\r\n2020-09-02 10:32:56.672605: I tensorflow/core/platform/profile_utils/cpu_utils.cc:104] CPU Frequency: 3199980000 Hz\r\n2020-09-02 10:32:56.675701: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x556ed093a910 initialized for platform Host (this does not guarantee that XLA will be used). Devices:\r\n2020-09-02 10:32:56.675767: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): Host, Default Version\r\n2020-09-02 10:32:56.678402: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.680525: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1716] Found device 0 with properties: \r\npciBusID: 0000:01:00.0 name: GeForce GTX TITAN X computeCapability: 5.2\r\ncoreClock: 1.2155GHz coreCount: 24 deviceMemorySize: 11.93GiB deviceMemoryBandwidth: 313.37GiB/s\r\n2020-09-02 10:32:56.680889: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.683022: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1716] Found device 1 with properties: \r\npciBusID: 0000:02:00.0 name: GeForce GTX TITAN X computeCapability: 5.2\r\ncoreClock: 1.2155GHz coreCount: 24 deviceMemorySize: 11.93GiB deviceMemoryBandwidth: 313.37GiB/s\r\n2020-09-02 10:32:56.683197: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudart.so.10.1\r\n2020-09-02 10:32:56.683257: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcublas.so.10\r\n2020-09-02 10:32:56.683304: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcufft.so.10\r\n2020-09-02 10:32:56.683346: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcurand.so.10\r\n2020-09-02 10:32:56.683504: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcusolver.so.10\r\n2020-09-02 10:32:56.683566: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcusparse.so.10\r\n2020-09-02 10:32:56.683693: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudnn.so.7\r\n2020-09-02 10:32:56.684014: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.686245: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.688465: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.690589: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.692497: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1858] Adding visible gpu devices: 0, 1\r\n2020-09-02 10:32:56.692670: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1257] Device interconnect StreamExecutor with strength 1 edge matrix:\r\n2020-09-02 10:32:56.692706: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1263] 0 1 \r\n2020-09-02 10:32:56.693071: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1276] 0: N Y \r\n2020-09-02 10:32:56.693135: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1276] 1: Y N \r\n2020-09-02 10:32:56.694784: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.696986: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.699094: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.701214: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.703406: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1402] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 10865 MB memory) -> physical GPU (device: 0, name: GeForce GTX TITAN X, pci bus id: 0000:01:00.0, compute capability: 5.2)\r\n2020-09-02 10:32:56.706854: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.709029: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\r\n2020-09-02 10:32:56.710969: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1402] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:1 with 10856 MB memory) -> physical GPU (device: 1, name: GeForce GTX TITAN X, pci bus id: 0000:02:00.0, compute capability: 5.2)\r\n2020-09-02 10:32:56.718970: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x556e0ea3c200 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:\r\n2020-09-02 10:32:56.719031: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): GeForce GTX TITAN X, Compute Capability 5.2\r\n2020-09-02 10:32:56.719056: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (1): GeForce GTX TITAN X, Compute Capability 5.2\r\n2020-09-02 10:32:57.410269: W tensorflow/python/util/util.cc:348] Sets are not currently considered sequences, but this may change in the future, so consider avoiding using them.\r\nPASSED\r\n[...]\r\ntests/test_modeling_tf_albert.py::TFAlbertModelTest::test_for_sequence_classification PASSED\r\ntests/test_modeling_tf_albert.py::TFAlbertModelTest::test_graph_mode WARNING:tensorflow:5 out of the last 5 calls to <function TFModelTesterMixin.test_graph_mode.<locals>.run_in_graph_mode at 0x7f273870ba70> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\r\nWARNING:tensorflow:6 out of the last 6 calls to <function TFModelTesterMixin.test_graph_mode.<locals>.run_in_graph_mode at 0x7f2738779ef0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\r\nWARNING:tensorflow:7 out of the last 7 calls to <function TFModelTesterMixin.test_graph_mode.<locals>.run_in_graph_mode at 0x7f2742cb89e0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\r\nPASSED\r\n[...]\r\ntests/test_modeling_tf_albert.py::TFAlbertModelTest::test_model_common_attributes PASSED\r\nDownloading: 100%|███████████████████████████████████████████████████████████████████████████████████████████████| 63.0M/63.0M [00:02<00:00, 28.8MB/s]\r\nPASSED\r\ntests/test_modeling_tf_albert.py::TFAlbertModelTest::test_model_outputs_equivalence PASSED\r\ntests/test_modeling_tf_albert.py::TFAlbertModelTest::test_pt_tf_model_equivalence PASSED\r\ntests/test_modeling_tf_albert.py::TFAlbertModelTest::test_resize_token_embeddings PASSED\r\ntests/test_modeling_tf_albert.py::TFAlbertModelTest::test_save_load PASSED\r\ntests/test_modeling_tf_albert.py::TFAlbertModelTest::test_saved_model_with_attentions_output WARNING:tensorflow:Skipping full serialization of Keras layer <tensorflow.python.keras.layers.core.Dropout object at 0x7f273831d450>, because it is not built.\r\nWARNING:tensorflow:From /home/stas/anaconda3/envs/main/lib/python3.7/site-packages/tensorflow/python/training/tracking/tracking.py:111: Model.state_updates (from tensorflow.python.keras.engine.training) is deprecated and will be removed in a future version.\r\nInstructions for updating:\r\nThis property should not be used in TensorFlow 2.0, as updates are applied automatically.\r\nWARNING:tensorflow:From /home/stas/anaconda3/envs/main/lib/python3.7/site-packages/tensorflow/python/training/tracking/tracking.py:111: Layer.updates (from tensorflow.python.keras.engine.base_layer) is deprecated and will be removed in a future version.\r\nInstructions for updating:\r\nThis property should not be used in TensorFlow 2.0, as updates are applied automatically.\r\nWARNING:tensorflow:No training configuration found in save file, so the model was *not* compiled. Compile it manually.\r\nFAILED\r\ntests/test_modeling_tf_albert.py::TFAlbertModelTest::test_saved_model_with_hidden_states_output WARNING:tensorflow:Skipping full serialization of Keras layer <tensorflow.python.keras.layers.core.Dropout object at 0x7f26cb69b7d0>, because it is not built.\r\nWARNING:tensorflow:No training configuration found in save file, so the model was *not* compiled. Compile it manually.\r\nWARNING:tensorflow:Skipping full serialization of Keras layer <tensorflow.python.keras.layers.core.Dropout object at 0x7f26c625ccd0>, because it is not built.\r\nWARNING:tensorflow:No training configuration found in save file, so the model was *not* compiled. Compile it manually.\r\nFAILED\r\ntests/test_modeling_tf_auto.py::TFAutoModelTest::test_from_identifier_from_model_type PASSED\r\ntests/test_modeling_tf_auto.py::TFAutoModelTest::test_from_pretrained_identifier PASSED\r\nDownloading: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 536M/536M [00:18<00:00, 28.9MB/s]\r\n2020-09-02 10:34:15.259729: W tensorflow/core/framework/cpu_allocator_impl.cc:81] Allocation of 93763584 exceeds 10% of free system memory.\r\n2020-09-02 10:34:15.394172: W tensorflow/core/framework/cpu_allocator_impl.cc:81] Allocation of 93763584 exceeds 10% of free system memory.\r\nPASSED\r\nDownloading: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 498M/498M [00:12<00:00, 40.0MB/s]\r\n2020-09-02 10:34:29.779196: W tensorflow/core/framework/cpu_allocator_impl.cc:81] Allocation of 154389504 exceeds 10% of free system memory.\r\n2020-09-02 10:34:30.859094: W tensorflow/core/framework/cpu_allocator_impl.cc:81] Allocation of 154389504 exceeds 10% of free system memory.\r\nPASSED\r\ntests/test_modeling_tf_auto.py::TFAutoModelTest::test_model_for_encoder_decoder_lm 2020-09-02 10:34:32.437951: W tensorflow/core/framework/cpu_allocator_impl.cc:81] Allocation of 65798144 exceeds 10% of free system memory.\r\nPASSED\r\n[...]\r\ntests/test_modeling_tf_bert.py::TFBertModelTest::test_for_token_classification PASSED\r\ntests/test_modeling_tf_bert.py::TFBertModelTest::test_graph_mode WARNING:tensorflow:8 out of the last 8 calls to <function TFModelTesterMixin.test_graph_mode.<locals>.run_in_graph_mode at 0x7f26c5ccbef0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\r\nWARNING:tensorflow:9 out of the last 9 calls to <function TFModelTesterMixin.test_graph_mode.<locals>.run_in_graph_mode at 0x7f26c5db8950> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\r\nWARNING:tensorflow:10 out of the last 10 calls to <function TFModelTesterMixin.test_graph_mode.<locals>.run_in_graph_mode at 0x7f26c5ccbcb0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\r\nWARNING:tensorflow:11 out of the last 11 calls to <function TFModelTesterMixin.test_graph_mode.<locals>.run_in_graph_mode at 0x7f277c0eb5f0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\r\nWARNING:tensorflow:11 out of the last 11 calls to <function TFModelTesterMixin.test_graph_mode.<locals>.run_in_graph_mode at 0x7f26c8298dd0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\r\nWARNING:tensorflow:11 out of the last 11 calls to <function TFModelTesterMixin.test_graph_mode.<locals>.run_in_graph_mode at 0x7f273816d5f0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\r\nWARNING:tensorflow:11 out of the last 11 calls to <function TFModelTesterMixin.test_graph_mode.<locals>.run_in_graph_mode at 0x7f26c8235560> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\r\nWARNING:tensorflow:11 out of the last 11 calls to <function TFModelTesterMixin.test_graph_mode.<locals>.run_in_graph_mode at 0x7f26caab45f0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\r\nWARNING:tensorflow:11 out of the last 11 calls to <function TFModelTesterMixin.test_graph_mode.<locals>.run_in_graph_mode at 0x7f26c9da34d0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\r\nPASSED\r\n```\r\n", "Thinking about it more while working with other tools, it'd be of a great help to have an env var that can set the default logging level for `transformers`. e.g. I wanted to change the logging level for `run_eval.py` and I couldn't do that w/o modifying it. If we had an env var that would have been trivial and much faster to use.\r\n\r\nThis is regardless of the outcome of this discussion of whether we should have a way to turn non-transformers-related loggers off.", "I understand the issue, and while I agree that some frameworks are extremely log intensive (TensorFlow ...), I wonder if it's such a bad thing to have too many logs during testing. If a test fails, the logs may help to understand the issue quicker when the stack trace isn't helping much. Removing these logs would mean needing to restart the CI with a different logging level to see what's happening in the logs around this error.\r\n\r\nRegarding your second point, yes, I think it would be nice to control the default logging level with an environment variable! Would welcome such a PR.", "I would find some more control over logging very useful! A lot of our users are on colab, and warnings waste a ton of screen space there. Same with my debugging workflow -- there are so many logger statements that can't see my full traceback on the screen.", "> I would find some more control over logging very useful! A lot of our users are on colab, and warnings waste a ton of screen space there. Same with my debugging workflow -- there are so many logger statements that can't see my full traceback on the screen.\r\n\r\nI wonder whether we should just have an env var `DISABLE_LOGGING=info` that will just do:\r\n```\r\nimport logging\r\nlogging.disable(logging.INFO) # disable INFO and DEBUG logger everywhere\r\n```\r\n`DISABLE_LOGGING=warning` for WARNING, INFO and DEBUG...\r\n\r\nIn addition to the transformers-specific one `TRANSFORMERS_VERBOSITY=info...` which I will add.\r\n", "> I understand the issue, and while I agree that some frameworks are extremely log intensive (TensorFlow ...), I wonder if it's such a bad thing to have too many logs during testing. If a test fails, the logs may help to understand the issue quicker when the stack trace isn't helping much. Removing these logs would mean needing to restart the CI with a different logging level to see what's happening in the logs around this error.\r\n\r\nIn no way I am proposing to impact CI in any way - on the contrary - on CI the more debug info the merrier. I'm only proposing a way for a developer to turn the logging off on their own setup. i.e. we won't be enabling any such features on CI.\r\n\r\nDifferent developers have different needs and for me, for example, noise is very counterproductive for development. When debugging something I only want to see outputs that are relevant to what I'm debugging and nothing else - and seconding @sshleifer's comment - I too want them to fit into the current screen so I don't need to scroll. Especially in complicated situations when I need to look at output numbers. I understand how this can be a total non-issue for others.\r\n\r\n> Regarding your second point, yes, I think it would be nice to control the default logging level with an environment variable! Would welcome such a PR.\r\n\r\nI will do so. Thank you!", "> I would find some more control over logging very useful! A lot of our users are on colab, and warnings waste a ton of screen space there. Same with my debugging workflow -- there are so many logger statements that can't see my full traceback on the screen.\r\n\r\n@sshleifer have you tried the new library-wide control for logging that Lysandre added in #6434?\r\nThe doc is here: https://huggingface.co/transformers/master/main_classes/logging.html", "Added the env var to control the transformers verbosity level: https://github.com/huggingface/transformers/pull/6961", "It feels that this proposal is a no go at the moment, so I'm closing it down.\r\n\r\nThe extended tests and added testing utils which were part of this PR have been merged in https://github.com/huggingface/transformers/pull/6961\r\n\r\nThank you all who contributed to this discussion." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6815?src=pr&el=h1) Report\n> Merging [#6815](https://codecov.io/gh/huggingface/transformers/pull/6815?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/5ab21b072fa2a122da930386381d23f95de06e28?el=desc) will **decrease** coverage by `1.46%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6815?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6815 +/- ##\n==========================================\n- Coverage 79.58% 78.11% -1.47% \n==========================================\n Files 157 157 \n Lines 28588 28588 \n==========================================\n- Hits 22752 22332 -420 \n- Misses 5836 6256 +420 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6815?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/modeling\\_tf\\_albert.py](https://codecov.io/gh/huggingface/transformers/pull/6815/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9hbGJlcnQucHk=) | `21.47% <0.00%> (-69.44%)` | :arrow_down: |\n| [src/transformers/pipelines.py](https://codecov.io/gh/huggingface/transformers/pull/6815/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9waXBlbGluZXMucHk=) | `26.26% <0.00%> (-53.69%)` | :arrow_down: |\n| [src/transformers/optimization.py](https://codecov.io/gh/huggingface/transformers/pull/6815/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9vcHRpbWl6YXRpb24ucHk=) | `34.28% <0.00%> (-48.00%)` | :arrow_down: |\n| [src/transformers/optimization\\_tf.py](https://codecov.io/gh/huggingface/transformers/pull/6815/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9vcHRpbWl6YXRpb25fdGYucHk=) | `33.33% <0.00%> (-24.33%)` | :arrow_down: |\n| [src/transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/6815/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19yb2JlcnRhLnB5) | `74.81% <0.00%> (-22.27%)` | :arrow_down: |\n| [src/transformers/modeling\\_longformer.py](https://codecov.io/gh/huggingface/transformers/pull/6815/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19sb25nZm9ybWVyLnB5) | `71.55% <0.00%> (-20.48%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/6815/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9hdXRvLnB5) | `48.79% <0.00%> (-18.08%)` | :arrow_down: |\n| [src/transformers/data/processors/squad.py](https://codecov.io/gh/huggingface/transformers/pull/6815/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9kYXRhL3Byb2Nlc3NvcnMvc3F1YWQucHk=) | `13.76% <0.00%> (-14.38%)` | :arrow_down: |\n| [src/transformers/modeling\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/6815/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19hdXRvLnB5) | `64.36% <0.00%> (-14.37%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_xlm.py](https://codecov.io/gh/huggingface/transformers/pull/6815/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl94bG0ucHk=) | `88.42% <0.00%> (-4.85%)` | :arrow_down: |\n| ... and [24 more](https://codecov.io/gh/huggingface/transformers/pull/6815/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6815?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/6815?src=pr&el=footer). Last update [5ab21b0...d984fd8](https://codecov.io/gh/huggingface/transformers/pull/6815?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "After sleeping on this, I'm not quite sure of 2 things.\r\n\r\n1. the main switch from mixin to normal subclassing - if it's done it should be done for all other common testing mixins - the benefit would be - having simpler access to `unittest.TestCase` and the extended `unittest.TestCasePlus` features. As I proposed in the alternative solution, it's not at all required, as a different solution can be used for temp dirs during debug.\r\n\r\n2. a totally unrelated issue of having debugging code in place. Do we want to gradually make the test suite easier to debug, by leaving `if DEBUG: ...` in strategic places (currently, consisting of just one thing - having a fixed tmp dir and not having it removed, but there are probably others). \r\n\r\n For example, I find myself adding a debug message for various asserts, so it's easier to see what's not matching, but those are usually a 2nd/3rd argument to the assert function (or `msg=`), so it's a smooth feature requiring no `if DEBUG`.\r\n\r\ni.e. I'd love to hear what others think - if you think this is a useful discussion - I can open 2 unrelated issues if it helps to make discussing these 2 unrelated issues focused.\r\n\r\nMy inclination right now is to just provide a quick way to make a fixed temp dir w/o it being deleted, i.e. the alt solution in OP, and leave the original PR for maybe some time in the future if we see other benefits to doing so.", "I agree with having a quicker fix for this specific problem and think a bit more about a general way to have a specific debug behavior for our use.", "If you're joining in now, please ignore the proposed code (as it also requires changing from Mixin to a subclass), and what this needs is your feedback on this question: **do we want to have a simple DEBUG flag in tests, that once enabled it would switch to not deleting temp dirs and would use a fixed temp dir path, so that it's easy to monitor?** So instead of needing to manually tweak the tests, we have the debug setup already in place. That's the question.\r\n\r\nLet me know if perhaps I should scratch that PR and start a new one discussing just that, so that the initial attempts at solving the issue won't be confusing to you, the readers.\r\n\r\nAnd to quickly give you context, we are talking about:\r\n```\r\n def setUp(self):\r\n self.tmpdirname = tempfile.mkdtemp()\r\n```\r\nand the modified version is:\r\n```\r\nDEBUG=0\r\n[...]\r\n def setUp(self):\r\n super().setUp()\r\n \r\n # if you need to debug the contents of the tmpdirname, set DEBUG to True, which will then use\r\n # a hardcoded path and won't delete it at the end of the test\r\n if not DEBUG:\r\n self.tmpdirname = self.get_auto_remove_tmp_dir()\r\n else:\r\n self.tmpdirname = self.get_auto_remove_tmp_dir(tmp_dir=\"./tmp/token-test\", after=False)\r\n```\r\nhttps://github.com/huggingface/transformers/blob/d984fd82bf940c62700919da5735e60f3f883348/tests/test_tokenization_common.py#L69\r\n\r\nexcept the code itself will be different as we can't make it work with mixins in that way.\r\n\r\nIf it helps, here is the last time a related issue of working with temp dirs has been worked on with a successful PR merge:\r\nhttps://github.com/huggingface/transformers/pull/6494 - i.e. this is a continuation of the same to other parts of the test suite.\r\n", "> do we want to have a simple DEBUG flag in tests, that once enabled it would switch to not deleting temp dirs and would use a fixed temp dir path, so that it's easy to monitor?\r\n\r\nyes, this would be useful if you can do it in a way that doesn't add overhead for people trying to add new tokenizers.\r\n\r\nI didn't look at the code.", "I will close it for now and revisit the next time I deal with this unless someone beats me to it." ]

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

[]

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

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6813?src=pr&el=h1) Report\n> Merging [#6813](https://codecov.io/gh/huggingface/transformers/pull/6813?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/244e1b5ba331cb4c1ed96d88d0895c252567f7f3?el=desc) will **decrease** coverage by `0.85%`.\n> The diff coverage is `82.89%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6813?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6813 +/- ##\n==========================================\n- Coverage 78.81% 77.95% -0.86% \n==========================================\n Files 174 178 +4 \n Lines 33670 34125 +455 \n==========================================\n+ Hits 26537 26603 +66 \n- Misses 7133 7522 +389 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6813?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/tokenization\\_rag.py](https://codecov.io/gh/huggingface/transformers/pull/6813/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fcmFnLnB5) | `69.76% <69.76%> (ø)` | |\n| [src/transformers/modeling\\_rag.py](https://codecov.io/gh/huggingface/transformers/pull/6813/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19yYWcucHk=) | `76.98% <76.98%> (ø)` | |\n| [src/transformers/testing\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6813/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90ZXN0aW5nX3V0aWxzLnB5) | `67.28% <77.77%> (+0.40%)` | :arrow_up: |\n| [src/transformers/retrieval\\_rag.py](https://codecov.io/gh/huggingface/transformers/pull/6813/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9yZXRyaWV2YWxfcmFnLnB5) | `91.27% <91.27%> (ø)` | |\n| [src/transformers/\\_\\_init\\_\\_.py](https://codecov.io/gh/huggingface/transformers/pull/6813/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9fX2luaXRfXy5weQ==) | `99.37% <100.00%> (+0.01%)` | :arrow_up: |\n| [src/transformers/configuration\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/6813/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX2F1dG8ucHk=) | `96.25% <100.00%> (+0.09%)` | :arrow_up: |\n| [src/transformers/configuration\\_dpr.py](https://codecov.io/gh/huggingface/transformers/pull/6813/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX2Rwci5weQ==) | `100.00% <100.00%> (ø)` | |\n| [src/transformers/configuration\\_rag.py](https://codecov.io/gh/huggingface/transformers/pull/6813/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX3JhZy5weQ==) | `100.00% <100.00%> (ø)` | |\n| [src/transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6813/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9maWxlX3V0aWxzLnB5) | `83.37% <100.00%> (+0.53%)` | :arrow_up: |\n| [src/transformers/modeling\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/6813/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19hdXRvLnB5) | `83.96% <100.00%> (+1.58%)` | :arrow_up: |\n| ... and [24 more](https://codecov.io/gh/huggingface/transformers/pull/6813/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6813?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/6813?src=pr&el=footer). Last update [3ebb1b3...db3e5e0](https://codecov.io/gh/huggingface/transformers/pull/6813?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "@lhoestq thanks for the comments! I did consider moving retrieval outside of the model - the benefit of this that I see would be that we would move all training-related logic (e.g. handling distributed processing in Retriever) from `transformers` to `examples`.\r\n\r\nThat said, I'm still in favor of keeping the call to `contextualize` as part of the forward pass. Here's my thinking:\r\n- retrieval is more than just data pre-processing step, it is a core part of the model's architecture. E.g. we can't pre-compute retrieved docs for a batch of data beforehand as the question encoder will be updated at every step of training, so the set of retrieved docs would be changing dynamically. If we move retrieval outside of the model people may be tempted to do that.\r\n- we would need to call `contextualize` before every forward pass on the model, so not only in finetuning, but also e.g. in evaluation code. On top of that anyone who would want to run the model for demo purposes would have to instantiate the retriever first and remember to call `contextualize`, instead of doing the two simple steps that other HF models require (encoding the sequence and running the model) - we could potentially consider making contextualization a part of tokenizer's `encode` method (not sure this would be intuitive for people used to HF's APIs) - however, the retrieval logic would still remain in `transformers` then\r\n- In terms of flexibility - I think with the current approach it'd still be possible for people to build different retrievers and pass them to the model\r\n\r\nWhat do you think? I'd be curious to know what others think about it, cc @patrick-s-h-lewis, @thomwolf", "So we've been brainstorming with @patrickvonplaten and @lhoestq on this yesterday and here is a proposal.\r\n\r\nThe general context is that we expect to include more hybrid models with a retrieval component in the future (like REALM, MARGE, knn-LM) so it's nice if we can setup an API which is general enough to include a diversity of models.\r\n\r\nHere is the general idea: we can add a new base class to the library (in addition to the `Tokenizer` and `Model`) which we can call a `Retriever` for instance. The `Retriever` class:\r\n- won't include trainable components and will thus be framework independent like the tokenizer,\r\n- does both the retrieval of the documents given an encoding and the post processing (in our case the retokenization),\r\n- it is sub-classed to be model specific (`RAGRetriever`) like the models and tokenizers,\r\n- we can add an `AutoRetriever` version later.\r\n\r\nWe probably want to keep it's API fairly generic for now since this field is only beginning to be explored. The `Retriever` can just have a `__call__` method (like the tokenizers) which expect an encoding and does retrieval and postprocessing.\r\n\r\nThen (copying @patrickvonplaten's idea) we would be to have a `RetrievalGenerationModel` that comprises the trainable (and PyTorch/TF specific) elements, i.e. the encoder and generator models. In its forward pass the `RetrievalGenerationModel` would have besides the usual `input_ids` one more input argument which is of class `Retriever`. The RetrivalGenerationModel would just call Retriever.forward(encoded_input) and expect an encoding that the self.generator could then be conditioned on.\r\n\r\nWhat do you think? We would help you implement this option of course since it impacts more significantly the library.", "I'm just wondering if TF will like a model which accept a class as input. What do you think @jplu @patrickvonplaten?\r\nWe could also have a method in the class to set the retriever instead of dynamically providing it. ", "Never tried, but I doubt it should be possible in compiled mode. I think what you propose would be a better way to go.", "> Never tried, but I doubt it should be possible in compiled mode. I think what you propose would be a better way to go.\r\n\r\n@jplu - I think there was a slight misunderstanding. \r\nWith the proposed approach we actually would pass an instantiation of a class as an argument to the forward pass of the `RetrievalGenerationModel` -> so before following this path we should check if this can nicely be done with TF...", "Oh ok! I thought the question was about to pass one class that contains all the arguments. My bad 😢 \r\n\r\nSo, after reading your explanation I can say, yes it is doable!", "Hey @ola13,\r\n\r\nThanks for your comment, this is indeed a very important aspect that I didn't really think of before. \r\nWith @lhoestq, we have been brainstorming a bit and thought maybe a slighly different design could make sense:\r\n\r\n```python\r\n#!/usr/bin/env python3\r\n\r\nclass RetrievalGenerationModel(PretrainedModel):\r\n \r\n def __init__(self, config: RetrievalGenerationConfig, encoder: PretrainedModel, retrieval_generator: PretrainedModel):\r\n if encoder is not None and retrieval_generator is not None: \r\n self.encoder = encoder\r\n self.retrieval_generator = retrieval_generator\r\n self.config = RetrievalGenerationConfig.from_encoder_generator_config(self.encoder.config, self.retrieval_generator.config)\r\n\r\n assert config is not None\r\n super().__init__(config)\r\n\r\n if encoder is None:\r\n self.encoder = AutoModel.from_config(config.encoder)\r\n if retrieval_generator is None:\r\n self.retrieval_generator.from_config(config.generator)\r\n\r\n @classmethod\r\n def from_pretrained_encoder_generator(cls, encoder_model_id, generator_model_id):\r\n encoder = AutoModel.from_pretrained(...) # load any query encoding model\r\n retrieval_generator = AutoRetrievalGeneratorModel.from_pretrained(...) # this would be a new class that contains any model that can be used as the `retrieval_generator` model.\r\n return cls(encoder=encoder, retrieval_generator=retrieval_generator)\r\n\r\n\r\n def forward(input_ids, retriever: PretrainedRetriever):\r\n # 1. Every retriever model encodes the query -> any `AutoModel` can be used here\r\n input_ids_encodings = self.encoder(input_ids) # model with weights\r\n\r\n # 2. Use costumized retriever (tokenizer-like) class instance, like `RAGRetriever` that \r\n # - query the index\r\n # - reformats the document outputs\r\n # - tokenizes the document outpus\r\n retrieved_docs_input_ids, retrieved_docs_encodings = retriever(input_ids_encodings, input_ids) # tokenizer like postprocessor that returns the tokenized docs input and the docs encodings\r\n\r\n # 3. Now the retrieval_generator requires a specific forward pass which accepts at least four kinds of tensors: 1) the input_ids (query), 2) the encoded_input_ids (encoded query), 3) retrieved_docs_input (tokenized context) and 4) retrieved_docs_encodings\r\n output_ids = self.retrieval_generator(input_ids, encoded_query, retrieved_docs_input_ids, retrieved_docs_encodings) # any `AutoRetrievalGeneratorModel` can be used here\r\n\r\nclass RagRetrievalGenerator(PretrainedModel):\r\n\r\n def __init__(self):\r\n self.generator = AutoModelForSeq2Seq.from_pretrained(...) # e.g. Bart\r\n\r\n def forward(input_ids, encodings, docs_input_ids, docs_encodings):\r\n doc_scores = torch.bmm(encodings.unsqueeze(1), docs_encodings.transpose(1, 2)).squeeze(1)\r\n ....\r\n output_ids = self.generator.generate(...)\r\n\r\nclass RAGRetriever(PretrainedRetriever)\r\n \"\"\"\r\n This retriever is framework independant (for both TF and PT) \r\n similar to a tokenizer\r\n \"\"\"\r\n\r\n def __init__(self):\r\n self.docs = nlp.load_dataset(...)\r\n ...\r\n\r\n def __call__(input_ids_encodings, input_ids):\r\n # no tensor operations happen here\r\n ...\r\n\r\nclass DPRRetrivalGenerator(PretrainedModel):\r\n\r\n def __init__(self):\r\n self.genator = AutoModelForQuestionsAnswering.from_pretrained(...) # QA model \r\n\r\n\r\n def forward(input_ids, encodings, docs_input_ids, docs_encodings):\r\n \r\n concated_qa_input = torch.cat([input_ids, docs_input_ids], dim=-1)\r\n output_ids = self.generator(concated_qa_input)\r\n\r\n\r\nclass DPRRetriever(PretrainedRetriever)\r\n \"\"\"\r\n This retriever is framework independant (for both TF and PT) \r\n similar to a tokenizer\r\n \"\"\"\r\n\r\n def __init__(self):\r\n self.docs = nlp.load_dataset(...)\r\n ...\r\n\r\n def __call__(input_ids_encodings, input_ids):\r\n # no tensor operations happen here\r\n ...\r\n```\r\n\r\nHopefully this is somewhat understandable @ola13 @thomwolf ...\r\n\r\n@lhoestq and I think that for each RetrivalAugmentedModel we need 2 specific parts:\r\n\r\n1) A specific Retriever: how are documents retrieved and formated and tokenized -> e.g. `RAGRetriever`\r\n2) A specific Generator: Here we can also have multiple possibilities: DPR uses a `AutoModelForQuestionAnswering` while RAG uses a `AutoModelForSeq2Seq`\r\n\r\nSo with this framework we would have to introduce 1 general class that would be used for all RetrievalAugementedModels, called `RetrievalGenerationModel` (or whatever name fits better) and 2 architecture specific classes `RAGRetriever` and `RagRetrievalGenerator`.\r\n\r\nWould be keen to hear your thoughts :-) ", "Hey @patrickvonplaten, makes sense and in fact it's not very different from how we structured the code already the key differences that I see are:\r\n- we move re-tokenization between query_encoder and generator to the Retriever (so respective tokenizers will be encapsulated by the Retriever not a model class as we currently do it)\r\n- we move retrieval score calculation to the model so that no tensor operations happen in the retriever\r\n\r\nwhich both should be pretty straightforward to implement.\r\n\r\nThe one thing that I'm still on the fence about is passing a `retriever` to each `forward` pass on a `RetrievalGenerationModel`, instead of making it a member of `RetrievalGenerationModel` class. Why do you feel the former is preferable over the latter?", "Yeah, good point! It's a bit weird to pass a class instance just to make a forward pass with it. \r\n\r\nMy main reason is the following: \r\n\r\nCurrently, the library makes a very clear distinction between `config`, `tokenizer` and `model` which are all independent of each other. Each of these classes have a seperate `.from_pretrained()` and `.save_pretrained()` method where as the `PretrainedModel.save_pretrained(...)` and `PretrainedModel.from_pretrained(...)` internally call `PretrainedConfig.save_pretrained(...)` and `PretrainedConfig.save_pretrained(...)`, but **never** the `PretrainedTokenizer.from_pretrained(...)` an d`PretrainedTokenizer.save_pretrained(...)` methods. For a `RetrievalGenerationModel` I would like to reuse `PretrainedModel`'s `from_pretrained(...)` and `save_pretrained(...)` methods which means that a tokenizer instance should not be part of the model because other wise we would have to tweak this function (which I don't think is a good idea). \r\nAlso, this will make the `RetrievalGenerationModel` a \"clean\" and relatively light `Model` object without any string processing logic in it whatsoever which is more in line with other `PretrainedModel` classes. ", "@patrickvonplaten, got it, yeah makes sense! We would still want to call `PretrainedTokenizer.from_pretrained(...)` when initializing `RagRetriever` but I guess this should be fine?\r\n\r\nOkay, so I would propose to do the following - I will refactor this PR to follow the design we discussed. It seems though that implementing the generic `Retriever` logic as discussed earlier by @thomwolf would require extra effort and time, and is not necessarily within the scope of this PR. In the interest of time, we could land this PR and then proceed with generalizing the retrieval logic? I'm then happy to work with the RAG implementation to make it compatible.", "Exactly! I was thinking that we either create a genereric `PretrainedRetriever` class with a `from_pretrained()` method that calls the tokenizer `from_pretrained()` methods or add `from_pretrained()` method directly to `RagRetriever`. Maybe @lhoestq and @thomwolf have better insight on the \"tokenizer\" side. \r\n\r\n@ola13 maybe we can wait quickly if @lhoestq and @thomwolf are fine with the design as discussed above :-) ", "Sounds awesome to me!", "Hey I just refactored the model following suggestions above. One point is that I had to modify `generation_utils.py` to account for a model which takes a `retriever` as an argument to the encoder. Let me know what you think!", "Hi, a question - to use RAG I need a couple of non-standard dependencies (faiss, psutil, nlp) - can I define a special test environment which would install those for rag tests? any pointers on how to handle this?", "> Hey I just refactored the model following suggestions above. One point is that I had to modify `generation_utils.py` to account for a model which takes a `retriever` as an argument to the encoder. Let me know what you think!\r\n\r\nAwesome ! I'll take a look. Also cc @patrickvonplaten \r\n\r\n> Hi, a question - to use RAG I need a couple of non-standard dependencies (faiss, psutil, nlp) - can I define a special test environment which would install those for rag tests? any pointers on how to handle this?\r\n\r\nMaybe @LysandreJik knows more about how to handle tests with dependencies ?\r\n", "Hey @ola13, \r\n\r\nI think the general code design is exactly what we have imagined to go for, defining a `RagRetriever` and passing the `retriever` to the forward pass, so this is great! ", "Regarding the test dependencies, you can add the libraries here: https://github.com/huggingface/transformers/blob/d6c08b07a087e83915b4b3156bbf464cebc7b9b5/setup.py#L92 and it should automatically be installed for testing on circle ci :-) `psutil` is already in the test dependency", "@ola13 - it would be awesome if you could add one \"full\" integration test with hardcoded input and output under @slow \r\n\r\nBy that I mean, *e.g.* hardcoding an input question \"Why does it rain\", loading a relevant dataset using the `HfIndex` and the full pretrained encoder and generator model and hardcoding the expected output answer in thet test. I think all operations are deterministic (beam search, etc...), so no random seeds have to be set.\r\n\r\n This way we have one test where we can be sure that the model works as expected and every change to the model in the future can be checked against that.\r\n\r\nThe tests you have in `test_modeling_rag.py` so far look great. We could also add a full `RagModel` test by defining a dummy dataset that will be instantiated from a hardcoded dict at test time and instantiating a very light `RagRetriever` at test time this way. But we can manually add those tests later, they are not super important.\r\n\r\nIn terms of a timeline, it would be be awesome if you manage to make the `test_modeling_rag.py` tests pass and if you could add one \"full\" integration test showing reasonable results. After this is finished, I think the best idea is if we add some changes on top of your PR (this should take another 1,2 days) and then merge the model into the lib :-) \r\n\r\nThanks a mille for your awesome work so far!!!", "Hey @patrickvonplaten, sounds good! yes definitely adding an integration test was on my agenda, right now having merged the `master` I'm also dealing with some issues arising after the refactor from https://github.com/huggingface/transformers/commit/afc4ece462ad83a090af620ff4da099a0272e171#diff-72b038fcff0de4ae5e094e3cde9471f1 as we were relying on the old structure of `past`. I'm hoping to be done with both of these things by tomorrow :) ", "Hi, I just added an integration test for RAG using the dummy variant of `wiki_dpr`. However, I had to locally hack `datasets` to make it run locally, as there seems to be a discrepancy between the dummy index name hardcoded in `wiki_dpr.py` here: https://github.com/huggingface/datasets/blob/37d4840a39eeff5d472beb890c8f850dc7723bb8/datasets/wiki_dpr/wiki_dpr.py#L72 (expecting `dummy.psgs_w100.nq.IndexHNSWFlat-IP-train.faiss`) and what's available on HF's google cloud bucket:\r\n```\r\n~$ gsutil ls -r gs://huggingface-nlp/datasets/wiki_dpr/*\r\ngs://huggingface-nlp/datasets/wiki_dpr/\r\ngs://huggingface-nlp/datasets/wiki_dpr/dummy_psgs_w100_with_nq_embeddings_IndexFlatIP-train.faiss\r\ngs://huggingface-nlp/datasets/wiki_dpr/psgs_w100.nq.IVFPQ4096_HNSW32_PQ64-IP-train.faiss\r\ngs://huggingface-nlp/datasets/wiki_dpr/psgs_w100_with_nq_embeddings_IVFPQ4096_HNSW32,PQ64-IP-train.faiss\r\n```\r\n\r\ncc @lhoestq - this would have to be fixed quickly, alternatively I could use full `wiki_dpr` in tests, but that's 78GB, not sure if it makes sense.\r\n\r\nLet me know what you think!", "> cc @lhoestq - this would have to be fixed quickly, alternatively I could use full `wiki_dpr` in tests, but that's 78GB, not sure if it makes sense.\r\n\r\nI fixed it, dummy.psgs_w100.nq.IndexHNSWFlat-IP-train.faiss is now available on gcs\r\n\r\n", "Previous RAG code is now saved in this PR: #7200", "Last fail is due to time-out. All import tests are passing => merging to master." ]

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

[]

[ "I think this is correct. It should be replaced by an `|`. Do you get a better perplexity if you change this line?", "Thanks a lot for opening this issue @HarshTrivedi ! I also agree that the logic should be OR and not AND. @shngt - can you maybe comment here as well?", "Thank you for confirming this!\r\n\r\nIf I remember correctly, changing `&` to `|` didn't fix the high zero-shot perplexity for me. I'll try it again later today or tomorrow and report back the numbers with `&` vs `|`.", "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 agree - the logic should be OR and not AND. Could you please confirm if the numbers change @HarshTrivedi?\r\n\r\nSorry for the delay - I missed the notification at the time. I'll submit a PR for AND -> OR fix asap, and try to do some more stringent testing to catch the reason for the perplexity difference. How can I proceed with the latter @patrickvonplaten @LysandreJik ?", "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", "(Resolved by https://github.com/huggingface/transformers/pull/8409 I believe)" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6811?src=pr&el=h1) Report\n> Merging [#6811](https://codecov.io/gh/huggingface/transformers/pull/6811?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/5ab21b072fa2a122da930386381d23f95de06e28?el=desc) will **decrease** coverage by `0.10%`.\n> The diff coverage is `100.00%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6811?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6811 +/- ##\n==========================================\n- Coverage 79.58% 79.47% -0.11% \n==========================================\n Files 157 157 \n Lines 28588 28586 -2 \n==========================================\n- Hits 22752 22719 -33 \n- Misses 5836 5867 +31 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6811?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/configuration\\_pegasus.py](https://codecov.io/gh/huggingface/transformers/pull/6811/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX3BlZ2FzdXMucHk=) | `100.00% <100.00%> (ø)` | |\n| [src/transformers/configuration\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6811/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX29wZW5haS5weQ==) | `34.28% <0.00%> (-62.86%)` | :arrow_down: |\n| [src/transformers/tokenization\\_albert.py](https://codecov.io/gh/huggingface/transformers/pull/6811/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fYWxiZXJ0LnB5) | `28.84% <0.00%> (-58.66%)` | :arrow_down: |\n| [src/transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6811/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19vcGVuYWkucHk=) | `23.87% <0.00%> (-57.10%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_distilbert.py](https://codecov.io/gh/huggingface/transformers/pull/6811/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9kaXN0aWxiZXJ0LnB5) | `64.47% <0.00%> (-34.36%)` | :arrow_down: |\n| [src/transformers/tokenization\\_marian.py](https://codecov.io/gh/huggingface/transformers/pull/6811/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fbWFyaWFuLnB5) | `67.79% <0.00%> (-31.36%)` | :arrow_down: |\n| [src/transformers/tokenization\\_transfo\\_xl.py](https://codecov.io/gh/huggingface/transformers/pull/6811/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdHJhbnNmb194bC5weQ==) | `20.53% <0.00%> (-21.21%)` | :arrow_down: |\n| [src/transformers/generation\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6811/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9nZW5lcmF0aW9uX3V0aWxzLnB5) | `96.66% <0.00%> (-0.28%)` | :arrow_down: |\n| [src/transformers/tokenization\\_utils\\_base.py](https://codecov.io/gh/huggingface/transformers/pull/6811/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHNfYmFzZS5weQ==) | `93.76% <0.00%> (+0.27%)` | :arrow_up: |\n| [src/transformers/modeling\\_tf\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6811/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl91dGlscy5weQ==) | `87.29% <0.00%> (+0.32%)` | :arrow_up: |\n| ... and [8 more](https://codecov.io/gh/huggingface/transformers/pull/6811/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6811?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/6811?src=pr&el=footer). Last update [5ab21b0...67322db](https://codecov.io/gh/huggingface/transformers/pull/6811?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

[]

[]

[]

[ { "id": 1834088753, "node_id": "MDU6TGFiZWwxODM0MDg4NzUz", "url": "https://api.github.com/repos/huggingface/transformers/labels/Tests", "name": "Tests", "color": "a6fcca", "default": false, "description": "Related to tests" } ]

[]

[]

[]

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

[ "Also noticed this! I have convinced myself that it's a scoring difference because the summaries generated are the same between this model and the fairseq implementation.\r\n", "This might help:\r\nhttps://github.com/google-research/google-research/issues/168\r\nI used pyrogue and R1, R2, RL = 44.32, 21.15, 37.53\r\n", "@yxyzzz can you tell me how you're using it? I get similar scores with py-rogue\r\n\r\n```\r\ndef calculate_rouge(output_lns: List[str], reference_lns: List[str], use_stemmer=True) -> Dict:\r\n scorer = rouge_scorer.RougeScorer(ROUGE_KEYS, use_stemmer=use_stemmer)\r\n aggregator = scoring.BootstrapAggregator()\r\n\r\n for reference_ln, output_ln in zip(reference_lns, output_lns):\r\n scores = scorer.score(reference_ln, output_ln)\r\n aggregator.add_scores(scores)\r\n\r\n result = aggregator.aggregate()\r\n\r\n\r\n import rouge\r\n import nltk\r\n nltk.download('punkt')\r\n\r\n evaluator = rouge.Rouge(metrics=['rouge-n', 'rouge-l'],\r\n max_n=2,\r\n limit_length=False,\r\n apply_avg=True)\r\n scores = evaluator.get_scores(reference_lns, output_lns)\r\n print(\"py-rogue\", scores)\r\n\r\n print(\"rogue_scorer\", {k: round(v.mid.fmeasure * 100, 4) for k, v in result.items()})\r\n```\r\n\r\nResults in:\r\n\r\n```\r\npy-rogue {'rouge-1': {'f': 0.44335299665102107, 'p': 0.5174289830764615, 'r': 0.40466586165106366}, 'rouge-2': {'f': 0.21133693864752542, 'p': 0.2465209393822732, 'r': 0.19324181648769206}, 'rouge-l': {'f': 0.3073058732169781, 'p': 0.35988134598642835, 'r': 0.2798097075410874}}\r\n\r\nrogue_scorer {'rouge1': 44.0698, 'rouge2': 21.0711, 'rougeLsum': 30.6233}\r\n```", "1. rouge_score split sentences by '\\n'. You can add a '\\n' to separate sentences in the summaries and evaluate. The summary level rougeL (rougeLsum) should be a lot higher and close to the one in the literature. \r\n'{'rouge1': 44.0536, 'rouge2': 21.0711, 'rougeL': 30.6157, 'rougeLsum': 40.9812}'\r\n```\r\noutput_ln2 = []\r\nfor o in `output_ln:\r\n s = sent_tokenize(p)\r\n output_ln2.append('\\n'.join(s))\r\n```\r\n2. Use pyrouge -> https://pypi.org/project/pyrouge/ ", "replacing \r\n```\r\noutput_lns = [x.rstrip() for x in open(args.save_path).readlines()]\r\nreference_lns = [x.rstrip() for x in open(args.reference_path).readlines()][: len(output_lns)]\r\n```\r\nwith works for rouge_score\r\n\r\n```\r\noutput_lns = [\" . \\n\".join(x.rstrip().split('. ')) for x in open(args.save_path).readlines()]\r\nreference_lns = [\" . \\n\".join(x.rstrip().split(' . ')) for x in open(args.reference_path).readlines()][: len(output_lns)]\r\n```\r\n\r\nThanks @yxyzzz !", "should we change run_eval.py ?\n", "Opened a PR at #7356 that fixes this issue @sshleifer " ]

[ { "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 } ]

[]

[ "Fixed, will add a check. \r\nAre you running distillation experiments!? FYI that model is not trained.", "> Are you running distillation experiments!? FYI that model is not trained.\r\n\r\nYes, I know. Reproducing the results, then planning to run a few experiments with it.\r\n\r\nWasn't able to use `--fp16 `, kept getting OOM errors (using 4 2080TIs).", "Cool! \r\nre: fp16:\r\nAre you in torch 1.6?\r\nTry torch 1.5.1 with apex installed.\r\n\r\nI haven't run anything successfully in torch 1.6 and am very suspicious of native amp.", "> Try torch 1.5.1 with apex installed.\r\n> \r\n> I haven't run anything successfully in torch 1.6 and am very suspicious of native amp.\r\n\r\nThanks, I will try that.\r\nAlso, did you use `run_eval.py` for the results [here](https://docs.google.com/spreadsheets/d/1EkhDMwVO02m8jCD1cG3RoFPLicpcL1GQHTQjfvDYgIM/edit#gid=0)? \r\nI tried using `sshleifer/distilbart-cnn-12-6` as well as one I finetuned from `sshleifer/student_cnn_12_6`, but got comparatively lower results.\r\n", "Yes I did, what were your results?", "Validation - `{'rouge1': 36.902390083382635, 'rouge2': 15.98520126771937, 'rougeL': 25.75566724592724} `\r\nTest -` {'rouge1': 33.980893339399074, 'rouge2': 13.925809496977044, 'rougeL': 23.731267594610095} `", "That's awful! Can I see your command?\r\n", "```\r\npython run_eval.py distilbart-cnn-12-6/best_tfmr $DATA_DIR/val.source dbart_val_generations.txt \\\r\n --reference_path $DATA_DIR/val.target \\\r\n --score_path distilbart-cnn-12-6/cnn_rouge.json \\\r\n --task summarization \\\r\n --n_obs 100 \\\r\n --device cuda \\\r\n --bs 32 \\\r\n```", "On 100 observations that might not be so bad. \r\nThe 21.26 Rouge 2 is from the following command (a few months ago):\r\n\r\n```bash\r\npython run_eval.py sshleifer/distilbart-cnn-12-6 \\\r\ncnn_dm/test.source \\\r\ndbart_cnn_12_6_test_gens.txt \\\r\n--reference_path cnn_dm/test.target \\\r\n--score_path dbart_cnn_12_6_test_rouge.json \\\r\n--task summarization --bs 32 --fp16\r\n```\r\nin torch 1.5.1.\r\n\r\nReran Today (it took an hour)\r\n```\r\n{'rouge1': 44.2503, 'rouge2': 21.2586, 'rougeL': 30.3729, 'n_obs': 11490, 'runtime': 3569, 'seconds_per_sample': 0.3106}\r\n```\r\n\r\n\r\n", "I had tried with 1000 (based on the [comment](https://docs.google.com/spreadsheets/d/1EkhDMwVO02m8jCD1cG3RoFPLicpcL1GQHTQjfvDYgIM/edit#gid=0)), had similar results. I wouldn't have expected the result to change that much, my bad. Thanks for your help!", "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/6804?src=pr&el=h1) Report\n> Merging [#6804](https://codecov.io/gh/huggingface/transformers/pull/6804?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/3cac867fac3f8717b25e3026b97b456a4e748039?el=desc) will **increase** coverage by `0.03%`.\n> The diff coverage is `100.00%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6804?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6804 +/- ##\n==========================================\n+ Coverage 79.21% 79.25% +0.03% \n==========================================\n Files 157 157 \n Lines 28588 28588 \n==========================================\n+ Hits 22646 22656 +10 \n+ Misses 5942 5932 -10 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6804?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/modeling\\_bart.py](https://codecov.io/gh/huggingface/transformers/pull/6804/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19iYXJ0LnB5) | `95.06% <100.00%> (-0.52%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_electra.py](https://codecov.io/gh/huggingface/transformers/pull/6804/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9lbGVjdHJhLnB5) | `25.13% <0.00%> (-73.83%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6804/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9vcGVuYWkucHk=) | `22.58% <0.00%> (-72.26%)` | :arrow_down: |\n| [src/transformers/tokenization\\_bart.py](https://codecov.io/gh/huggingface/transformers/pull/6804/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fYmFydC5weQ==) | `42.10% <0.00%> (-57.90%)` | :arrow_down: |\n| [src/transformers/trainer.py](https://codecov.io/gh/huggingface/transformers/pull/6804/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90cmFpbmVyLnB5) | `13.21% <0.00%> (-40.45%)` | :arrow_down: |\n| [src/transformers/modeling\\_marian.py](https://codecov.io/gh/huggingface/transformers/pull/6804/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19tYXJpYW4ucHk=) | `60.00% <0.00%> (-30.00%)` | :arrow_down: |\n| [src/transformers/training\\_args.py](https://codecov.io/gh/huggingface/transformers/pull/6804/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90cmFpbmluZ19hcmdzLnB5) | `66.66% <0.00%> (-25.00%)` | :arrow_down: |\n| [src/transformers/trainer\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6804/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90cmFpbmVyX3V0aWxzLnB5) | `64.44% <0.00%> (-20.00%)` | :arrow_down: |\n| [src/transformers/activations.py](https://codecov.io/gh/huggingface/transformers/pull/6804/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9hY3RpdmF0aW9ucy5weQ==) | `85.00% <0.00%> (-5.00%)` | :arrow_down: |\n| [src/transformers/tokenization\\_dpr.py](https://codecov.io/gh/huggingface/transformers/pull/6804/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fZHByLnB5) | `53.15% <0.00%> (-4.51%)` | :arrow_down: |\n| ... and [14 more](https://codecov.io/gh/huggingface/transformers/pull/6804/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6804?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/6804?src=pr&el=footer). Last update [3cac867...7a6bf5a](https://codecov.io/gh/huggingface/transformers/pull/6804?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "What do you mean default in config?\r\nThe default `ignore_index` is -100 for CrossEntropyLoss.\r\n`pad_token_id` is overwritten by BartConfig.", "Sorry mixed things up. This would make `BartForConditionalGeneration` behave differently from all the others models (all ModelForMaskedLM and T5ForConditionalGeneration use -100) so I think this is pretty breaking. Users probably have special code to changed padded token to -100, plus you may want to mask other things than the padding for loss computation (more relevant for masked LM than seq2seq but still).\r\n\r\nI think this is some preprocessing work to do on the labels, for instance the `DataCollatorForLanguageModeling` replaces all non-masked tokens by -100 in the labels." ]

[]

[]

[]

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

[]

[ "Style issue will be solved with merge @sgugger ", "> Thanks for fixing this! I'd remove the first change in the logs though.\r\n\r\nThanks for the review! Done.", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6802?src=pr&el=h1) Report\n> Merging [#6802](https://codecov.io/gh/huggingface/transformers/pull/6802?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/9336086ab5d232cccd9512333518cf4299528882?el=desc) will **decrease** coverage by `0.42%`.\n> The diff coverage is `89.47%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6802?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6802 +/- ##\n==========================================\n- Coverage 80.32% 79.89% -0.43% \n==========================================\n Files 157 157 \n Lines 28589 28739 +150 \n==========================================\n- Hits 22963 22960 -3 \n- Misses 5626 5779 +153 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6802?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/\\_\\_init\\_\\_.py](https://codecov.io/gh/huggingface/transformers/pull/6802/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9fX2luaXRfXy5weQ==) | `99.28% <ø> (ø)` | |\n| [src/transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6802/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9maWxlX3V0aWxzLnB5) | `82.66% <ø> (+0.25%)` | :arrow_up: |\n| [src/transformers/modeling\\_tf\\_flaubert.py](https://codecov.io/gh/huggingface/transformers/pull/6802/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9mbGF1YmVydC5weQ==) | `88.34% <ø> (+63.80%)` | :arrow_up: |\n| [src/transformers/optimization.py](https://codecov.io/gh/huggingface/transformers/pull/6802/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9vcHRpbWl6YXRpb24ucHk=) | `82.28% <ø> (ø)` | |\n| [src/transformers/tokenization\\_t5.py](https://codecov.io/gh/huggingface/transformers/pull/6802/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdDUucHk=) | `95.28% <ø> (-0.05%)` | :arrow_down: |\n| [src/transformers/trainer.py](https://codecov.io/gh/huggingface/transformers/pull/6802/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90cmFpbmVyLnB5) | `53.23% <46.66%> (-0.43%)` | :arrow_down: |\n| [...rc/transformers/data/datasets/language\\_modeling.py](https://codecov.io/gh/huggingface/transformers/pull/6802/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9kYXRhL2RhdGFzZXRzL2xhbmd1YWdlX21vZGVsaW5nLnB5) | `90.69% <89.18%> (-1.14%)` | :arrow_down: |\n| [src/transformers/data/data\\_collator.py](https://codecov.io/gh/huggingface/transformers/pull/6802/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9kYXRhL2RhdGFfY29sbGF0b3IucHk=) | `91.90% <94.59%> (+2.19%)` | :arrow_up: |\n| [src/transformers/configuration\\_pegasus.py](https://codecov.io/gh/huggingface/transformers/pull/6802/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX3BlZ2FzdXMucHk=) | `100.00% <100.00%> (ø)` | |\n| [src/transformers/data/datasets/\\_\\_init\\_\\_.py](https://codecov.io/gh/huggingface/transformers/pull/6802/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9kYXRhL2RhdGFzZXRzL19faW5pdF9fLnB5) | `100.00% <100.00%> (ø)` | |\n| ... and [25 more](https://codecov.io/gh/huggingface/transformers/pull/6802/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6802?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/6802?src=pr&el=footer). Last update [9336086...2b981cd](https://codecov.io/gh/huggingface/transformers/pull/6802?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

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

[]

[]

[]

[]

[]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6799?src=pr&el=h1) Report\n> Merging [#6799](https://codecov.io/gh/huggingface/transformers/pull/6799?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/02d09c8fcc6bda2c345c84cec53289abbe7532ac?el=desc) will **increase** coverage by `0.10%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6799?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6799 +/- ##\n==========================================\n+ Coverage 79.01% 79.11% +0.10% \n==========================================\n Files 157 157 \n Lines 28739 28739 \n==========================================\n+ Hits 22707 22736 +29 \n+ Misses 6032 6003 -29 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6799?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/modeling\\_tf\\_electra.py](https://codecov.io/gh/huggingface/transformers/pull/6799/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9lbGVjdHJhLnB5) | `25.13% <0.00%> (-73.83%)` | :arrow_down: |\n| [src/transformers/tokenization\\_marian.py](https://codecov.io/gh/huggingface/transformers/pull/6799/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fbWFyaWFuLnB5) | `32.20% <0.00%> (-66.95%)` | :arrow_down: |\n| [src/transformers/modeling\\_marian.py](https://codecov.io/gh/huggingface/transformers/pull/6799/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19tYXJpYW4ucHk=) | `60.00% <0.00%> (-30.00%)` | :arrow_down: |\n| [src/transformers/tokenization\\_reformer.py](https://codecov.io/gh/huggingface/transformers/pull/6799/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fcmVmb3JtZXIucHk=) | `81.66% <0.00%> (-13.34%)` | :arrow_down: |\n| [src/transformers/activations.py](https://codecov.io/gh/huggingface/transformers/pull/6799/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9hY3RpdmF0aW9ucy5weQ==) | `85.00% <0.00%> (-5.00%)` | :arrow_down: |\n| [src/transformers/tokenization\\_auto.py](https://codecov.io/gh/huggingface/transformers/pull/6799/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fYXV0by5weQ==) | `95.55% <0.00%> (-2.23%)` | :arrow_down: |\n| [src/transformers/configuration\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6799/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX3V0aWxzLnB5) | `96.00% <0.00%> (-0.67%)` | :arrow_down: |\n| [src/transformers/modeling\\_bart.py](https://codecov.io/gh/huggingface/transformers/pull/6799/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19iYXJ0LnB5) | `95.06% <0.00%> (-0.52%)` | :arrow_down: |\n| [src/transformers/data/data\\_collator.py](https://codecov.io/gh/huggingface/transformers/pull/6799/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9kYXRhL2RhdGFfY29sbGF0b3IucHk=) | `91.90% <0.00%> (-0.41%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6799/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl91dGlscy5weQ==) | `86.97% <0.00%> (-0.33%)` | :arrow_down: |\n| ... and [8 more](https://codecov.io/gh/huggingface/transformers/pull/6799/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6799?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/6799?src=pr&el=footer). Last update [02d09c8...d6e38c4](https://codecov.io/gh/huggingface/transformers/pull/6799?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

[]

[]

[]

[ { "id": 1834088753, "node_id": "MDU6TGFiZWwxODM0MDg4NzUz", "url": "https://api.github.com/repos/huggingface/transformers/labels/Tests", "name": "Tests", "color": "a6fcca", "default": false, "description": "Related to tests" } ]

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

[]

[ { "id": 2039044877, "node_id": "MDU6TGFiZWwyMDM5MDQ0ODc3", "url": "https://api.github.com/repos/huggingface/transformers/labels/marian", "name": "marian", "color": "30cc95", "default": false, "description": "" } ]

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

[ "Which tutorial? its called `prepare_seq2seq_batch` now.", "\"prepare_seq2seq_batch\" works. Thanks a lot.\r\n\r\nI follow this tutorial, https://huggingface.co/transformers/model_doc/marian.html\r\n\r\nWhere can I find the new user manual for MarianMT model? Thank you.\r\n", "https://huggingface.co/transformers/master/model_doc/marian.html", "Have they change it again and add a maximum length?", "I am getting: `AttributeError: 'MarianTokenizer' object has no attribute 'prepare_seq2seq_batch'`\r\n\r\nI changed it to `prepare_translation_batch` and it works", "It fails again... and changing to `prepare_seq2seq_batch` throws the deprecation warning..." ]

[ { "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 } ]

[]

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

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

[]

[]

[]

[]

[ "Fixed by #6858 " ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6791?src=pr&el=h1) Report\n> Merging [#6791](https://codecov.io/gh/huggingface/transformers/pull/6791?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/930153e7d2d658267b7630a047a4bfc85b86042d?el=desc) will **increase** coverage by `0.41%`.\n> The diff coverage is `0.00%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6791?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6791 +/- ##\n==========================================\n+ Coverage 79.36% 79.78% +0.41% \n==========================================\n Files 157 157 \n Lines 28569 28578 +9 \n==========================================\n+ Hits 22675 22800 +125 \n+ Misses 5894 5778 -116 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6791?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/trainer.py](https://codecov.io/gh/huggingface/transformers/pull/6791/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90cmFpbmVyLnB5) | `52.90% <0.00%> (+39.68%)` | :arrow_up: |\n| [src/transformers/modeling\\_tf\\_mobilebert.py](https://codecov.io/gh/huggingface/transformers/pull/6791/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9tb2JpbGViZXJ0LnB5) | `24.55% <0.00%> (-72.36%)` | :arrow_down: |\n| [src/transformers/tokenization\\_mbart.py](https://codecov.io/gh/huggingface/transformers/pull/6791/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fbWJhcnQucHk=) | `35.93% <0.00%> (-59.38%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_distilbert.py](https://codecov.io/gh/huggingface/transformers/pull/6791/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9kaXN0aWxiZXJ0LnB5) | `64.47% <0.00%> (-34.36%)` | :arrow_down: |\n| [src/transformers/tokenization\\_xlnet.py](https://codecov.io/gh/huggingface/transformers/pull/6791/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25feGxuZXQucHk=) | `66.66% <0.00%> (-23.43%)` | :arrow_down: |\n| [src/transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6791/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19vcGVuYWkucHk=) | `72.25% <0.00%> (-8.71%)` | :arrow_down: |\n| [src/transformers/configuration\\_bart.py](https://codecov.io/gh/huggingface/transformers/pull/6791/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX2JhcnQucHk=) | `90.00% <0.00%> (-4.00%)` | :arrow_down: |\n| [src/transformers/generation\\_tf\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6791/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9nZW5lcmF0aW9uX3RmX3V0aWxzLnB5) | `85.21% <0.00%> (-1.26%)` | :arrow_down: |\n| [src/transformers/generation\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6791/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9nZW5lcmF0aW9uX3V0aWxzLnB5) | `96.66% <0.00%> (-0.28%)` | :arrow_down: |\n| [src/transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6791/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9maWxlX3V0aWxzLnB5) | `82.66% <0.00%> (+0.25%)` | :arrow_up: |\n| ... and [14 more](https://codecov.io/gh/huggingface/transformers/pull/6791/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6791?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/6791?src=pr&el=footer). Last update [930153e...001a17f](https://codecov.io/gh/huggingface/transformers/pull/6791?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "making sure that @borisdayma sees this PR.", "Thanks, I've not played too much with ray tune yet but there seems to be 2 ways to integrate through Ray Tune libraries as per [the docs](https://docs.wandb.com/library/integrations/ray-tune).\r\n\r\nHowever ideally, the setup, logging, etc would be handled directly by `Trainer` existing functions for clarity and concision (and also to support all existing loggers). Handling multiple logging runs should be done within `hyperparameter_search` if possible.\r\n\r\nCould the setup methods be wrapped in a new function and called during the search, in order to avoid duplicating the same logic.\r\nFor wandb, forcing a new run just requires `wandb.init(reinit=True)` so it works both in notebooks and scripts.\r\n\r\nNote: Use this argument **only** while using `hyperparameter_search` as users can currently call manually `wandb.init` before (for example when using pytorch-lightning, sweeps, or keras + huggingface), making the call within the `Trainer` a \"noop\" (because it does not have `reinit=True`).", "Thanks for your comments.\r\n\r\n@borisdayma, simply moving the logger setup to `train()` would do the trick in any case, as it is called from the hyperparameter search methods. This should also work for Optuna, not only for Ray Tune.\r\n\r\nI created a PR for that here: #6850. Is this what you meant?", "Closed in favor of #6850." ]

[ { "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" ]

[]

[]

[ "Let's say the tokens you want to add are <some_token_1> and <some_token_2> (including angle brackets)\r\n```\r\nfrom transformers import T5Tokenizer\r\ntokenizer = T5Tokenizer.from_pretrained(\"t5-base\")\r\ntokenizer.add_tokens(['<some_token_1>', '<some_token_2'>])\r\n```", "I want to add this line in addition to modifying the tokenizer for the model to work with the new tokenizer:\r\n\r\n`model.resize_token_embeddings(len(tokenizer))`" ]

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

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6788?src=pr&el=h1) Report\n> Merging [#6788](https://codecov.io/gh/huggingface/transformers/pull/6788?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/930153e7d2d658267b7630a047a4bfc85b86042d?el=desc) will **decrease** coverage by `2.98%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6788?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6788 +/- ##\n==========================================\n- Coverage 79.36% 76.38% -2.99% \n==========================================\n Files 157 157 \n Lines 28569 28569 \n==========================================\n- Hits 22675 21822 -853 \n- Misses 5894 6747 +853 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6788?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/modeling\\_reformer.py](https://codecov.io/gh/huggingface/transformers/pull/6788/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19yZWZvcm1lci5weQ==) | `16.87% <0.00%> (-79.30%)` | :arrow_down: |\n| [src/transformers/configuration\\_reformer.py](https://codecov.io/gh/huggingface/transformers/pull/6788/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX3JlZm9ybWVyLnB5) | `21.62% <0.00%> (-78.38%)` | :arrow_down: |\n| [src/transformers/tokenization\\_xlm.py](https://codecov.io/gh/huggingface/transformers/pull/6788/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25feGxtLnB5) | `16.26% <0.00%> (-66.67%)` | :arrow_down: |\n| [src/transformers/configuration\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6788/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX29wZW5haS5weQ==) | `34.28% <0.00%> (-62.86%)` | :arrow_down: |\n| [src/transformers/tokenization\\_mbart.py](https://codecov.io/gh/huggingface/transformers/pull/6788/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fbWJhcnQucHk=) | `35.93% <0.00%> (-59.38%)` | :arrow_down: |\n| [src/transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6788/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19vcGVuYWkucHk=) | `23.87% <0.00%> (-57.10%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_distilbert.py](https://codecov.io/gh/huggingface/transformers/pull/6788/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9kaXN0aWxiZXJ0LnB5) | `64.47% <0.00%> (-34.36%)` | :arrow_down: |\n| [src/transformers/tokenization\\_ctrl.py](https://codecov.io/gh/huggingface/transformers/pull/6788/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fY3RybC5weQ==) | `78.64% <0.00%> (-17.48%)` | :arrow_down: |\n| [src/transformers/tokenization\\_utils\\_fast.py](https://codecov.io/gh/huggingface/transformers/pull/6788/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHNfZmFzdC5weQ==) | `81.42% <0.00%> (-12.86%)` | :arrow_down: |\n| [src/transformers/tokenization\\_transfo\\_xl.py](https://codecov.io/gh/huggingface/transformers/pull/6788/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdHJhbnNmb194bC5weQ==) | `33.56% <0.00%> (-8.93%)` | :arrow_down: |\n| ... and [18 more](https://codecov.io/gh/huggingface/transformers/pull/6788/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6788?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/6788?src=pr&el=footer). Last update [930153e...3482a3e](https://codecov.io/gh/huggingface/transformers/pull/6788?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

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

[]

[ "Hi,\r\n\r\nHave a look at the following script from Huggingface:\r\n\r\nhttps://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner.py\r\n\r\nIt shows how you can finetune a pretrained BERT-model for NER.\r\nRemember to take a look at the utility code as well, since this is the code preparing and creating your features (and tensors in general).\r\n\r\nhttps://github.com/huggingface/transformers/blob/master/examples/token-classification/utils_ner.py\r\n\r\nRegards", "\r\n> Hi,\r\n> \r\n> Have a look at the following script from Huggingface:\r\n> \r\n> https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner.py\r\n> \r\n> It shows how you can finetune a pretrained BERT-model for NER.\r\n> Remember to take a look at the utility code as well, since this is the code preparing and creating your features (and tensors in general).\r\n> \r\n> https://github.com/huggingface/transformers/blob/master/examples/token-classification/utils_ner.py\r\n> \r\n> Regards\r\n\r\nThanks for your response very much.\r\n\r\nActually I am not good at python programming.\r\n\r\nrun_ner.py could create a new pretrained model or I could us it to fine tune a existed model to my target NER task?\r\n\r\nWhat are the functions of these two programs? And what parameters should I set?\r\nWhat are the formats of train and test data?\r\n\r\nI also could not find these two programs in the download folder. \r\n\r\n\r\nBy the way, could you offer the tutorial of colab format of NER task just like below link?\r\nhttps://www.depends-on-the-definition.com/named-entity-recognition-with-bert/\r\n\r\nIt is complicated and consuming to set virtual environment in PC.\r\n\r\nThank you\r\n Best regards; \r\n\r\n\r\n", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

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

[]

[ "When I add torch.manual_seed(0) at the beginning the outputs will be the same.\r\ndoes the model from tf pretrained has some randomization when loading the model weights?", "Hmm, there sholud always be the same. Could you add your script to a google colab that allows to run your checkpoint, so that we can debug?", "Hi here is the colab script with my checkpoint uploaded:\r\nhttps://colab.research.google.com/drive/1Yx7zRkzpaGMMraTkIeIGnJJLnlpUKzQH?usp=sharing\r\n\r\nyou can download my checkpoint here: https://drive.google.com/drive/folders/1521pvzvkqvEBUvRqZn7CRO-soCXOUZCu?usp=sharing", "I think some layers in your T5 model have not been trained and are therefore not saved in your model checkpoint. At initialization this layer is then randomely initialized. \r\n\r\nOne thing you can do to verify is to load the model as follows, save it as a PyTorch model and then load the PyTorch model as follows.\r\n\r\n```python \r\nconfig = T5Config.from_pretrained('t5-base')\r\nmodel = T5ForConditionalGeneration.from_pretrained(\r\n \"sample_data/model/model.ckpt-1004000\", from_tf=True, config=config)\r\nmodel.save_pretrained(\"./model\")\r\nmodel = T5ForConditionalGeneration.from_pretrained(\"/.model\")\r\n```\r\n\r\nif this command throws a warning that some layers are not initialized, then you know what the problem is.\r\nIf not, I will take a look again :-) \r\n", "hi, I tried what you said, no warning shows up.\r\nso I guess it is not the case?\r\n\r\nDo you think it could be that weights from float64 to float32 or the opposite causing this problem? because the difference in outputs is tiny. I don't know, this still cannot explain why outputs are different every time.", "any updates?", "Hey @ArvinZhuang, \r\n\r\nI just downloaded the weights and tried to reproduce the error using your code snippet. In my case the output is deterministic, as expected. Could you make sure that you are using the newest version of transformers and try again?", "Hi, the outputs still different on my machine.... very strange.\r\n\r\nI'm using transformer v3.3.0\r\ntorch v1.6.0\r\nTensorFlow v2.3.1 \r\n\r\n\r\n\r\nand btw, I cannot directly load t5 config by using T5Config.from_pretrained('t5-base') now, but the 't5-base' is still in the \"https://huggingface.co/models\" list. So I copy and past config.json from \"https://huggingface.co/t5-base\" but the results show this time is very different from the post above, which I think should not be the case because the tf checkpoint and input string are exactly the same as before....\r\n\r\n\r\nupdates: can directly load t5 config by T5Config.from_pretrained('t5-base') now, however, the output logits still very different from my first post above.....", "+1 on the issue\r\nI'm using transformer ==3.4.0\r\ntorch==1.6.0\r\n\r\nI run:\r\n\r\n```\r\nfrom transformers import AutoTokenizer, AutoModelForSeq2SeqLM, AutoConfig\r\n\r\nconfig = AutoConfig.from_pretrained(\"Vamsi/T5_Paraphrase_Paws\", output_hidden_states=True) \r\ntokenizer = AutoTokenizer.from_pretrained(\"Vamsi/T5_Paraphrase_Paws\") \r\nmodel = AutoModelForSeq2SeqLM.from_pretrained(\"Vamsi/T5_Paraphrase_Paws\", config=config).to('cuda')\r\n\r\ndef prediction(documents, query):\r\n querys = [query] * len(documents)\r\n encoded_decoder_inputs = tokenizer(documents, padding=True, truncation=True, return_tensors=\"pt\").to('cuda')\r\n encoded_encoder_inputs = tokenizer(querys, padding=True, truncation=True, return_tensors=\"pt\").to('cuda')\r\n with torch.no_grad():\r\n outputs = model(input_ids=encoded_encoder_inputs[\"input_ids\"],\r\n labels=encoded_decoder_inputs[\"input_ids\"],\r\n attention_mask=encoded_encoder_inputs[\"attention_mask\"])\r\n batch_logits = outputs[1]\r\n print(batch_logits)\r\n\r\ndocuments = ['a', 'b']\r\nquery = \"who am I?\"\r\nprediction(documents, query) \r\n```\r\n\r\nand got:\r\n```\r\ntensor([[[-21.6500, -9.8658, -13.6561, ..., -43.1233, -43.0788, -43.0745],\r\n [-30.3906, -12.7200, -1.2460, ..., -41.7208, -41.6774, -41.6465],\r\n [-15.7073, -5.9496, -5.9364, ..., -36.8553, -36.8221, -36.8052]],\r\n\r\ntensor([[[-21.6500, -9.8658, -13.6561, ..., -43.1233, -43.0788, -43.0745],\r\n [-30.3906, -12.7200, -1.2460, ..., -41.7208, -41.6774, -41.6465],\r\n [-20.1459, -5.3198, -4.7644, ..., -37.7978, -37.7850, -37.8202]]],\r\n device='cuda:0')\r\n```\r\nNote: rerunning ` prediction(documents, query)` produces same deterministic results, suggesting that the inputs to `labels` do affect the `logits` outcome.", "Hi ednussi, \r\nyes, rerunning prediction(documents, query) will give deterministic results. \r\nHowever, my issue is rerunning the above chunk of code twice (reload everything, including model.). the outputs are different.\r\nDoes this also happen to you?\r\n\r\nUpdate:\r\nI tried the model provided by ednussi, and the outputs of running the code twice are the same. But my tf model still gives two different results, suggesting that loading model from tf gives different outcomes.", "Hi @ArvinZhuang,\r\nYes, similar to you when I run the model call with same input_ids but different labels I got different outcomes (using pytorch). Realized I missed a part of the print so edited my comment above to match the output.\r\nHoping @patrickvonplaten or someone from the `huggingface` team can take a second look, so we can get to the bottom of this.", "Hey @ednussi, if you look into the code, you can see that T5 uses the `labels` to create the `decoder_input_ids` which necessarily do affect the outputs. You can try using deterministic `input_ids` and `decoder_input_ids` and no labels and see if the output stays deterministic (it should).", "Thanks @patrickvonplaten.\r\nWas following your suggestion, and after reading through the documentation and code of how the `decoder_input_ids` is used, it became clear why it affects the `logits` and helped clear my confusion.", "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", "@ednussi - wanted to ask for your help if you could provide a brief explanation from your reading? Thanks." ]

[ { "id": 1108649053, "node_id": "MDU6TGFiZWwxMTA4NjQ5MDUz", "url": "https://api.github.com/repos/huggingface/transformers/labels/Help%20wanted", "name": "Help wanted", "color": "008672", "default": false, "description": "Extra attention is needed, help appreciated" }, { "id": 1845609017, "node_id": "MDU6TGFiZWwxODQ1NjA5MDE3", "url": "https://api.github.com/repos/huggingface/transformers/labels/seq2seq", "name": "seq2seq", "color": "fef2c0", "default": false, "description": "" }, { "id": 1936351150, "node_id": "MDU6TGFiZWwxOTM2MzUxMTUw", "url": "https://api.github.com/repos/huggingface/transformers/labels/Examples", "name": "Examples", "color": "d4c5f9", "default": false, "description": "Which is related to examples in general" }, { "id": 2009457320, "node_id": "MDU6TGFiZWwyMDA5NDU3MzIw", "url": "https://api.github.com/repos/huggingface/transformers/labels/translation", "name": "translation", "color": "b2d2f4", "default": false, "description": "machine translation utilities and models" } ]

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

[]

[]

[]

[]

[ { "id": 1845609017, "node_id": "MDU6TGFiZWwxODQ1NjA5MDE3", "url": "https://api.github.com/repos/huggingface/transformers/labels/seq2seq", "name": "seq2seq", "color": "fef2c0", "default": false, "description": "" }, { "id": 1936351150, "node_id": "MDU6TGFiZWwxOTM2MzUxMTUw", "url": "https://api.github.com/repos/huggingface/transformers/labels/Examples", "name": "Examples", "color": "d4c5f9", "default": false, "description": "Which is related to examples in general" } ]

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

[ "@sshleifer I can help out here. What exactly are you looking for?", "I'll take care of it actually, sorry! There is a harder one at #6785 if you're interested.", "Fixed, using sortish sampler for val. Much faster!" ]

[]

[]

[ "weird, I thought I had a newer version, but it was old for some reason. updating to `pip install black==20.8b1` resolved this." ]

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

[]

[ "Good catch, thanks a lot! We will fix it ASAP.", "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" ]

[ { "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 } ]

[]

[ "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/6777?src=pr&el=h1) Report\n> Merging [#6777](https://codecov.io/gh/huggingface/transformers/pull/6777?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/42fddacd1cac3cc57c3326aa51a409f5090b1261?el=desc) will **increase** coverage by `1.90%`.\n> The diff coverage is `100.00%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6777?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6777 +/- ##\n==========================================\n+ Coverage 78.47% 80.37% +1.90% \n==========================================\n Files 157 157 \n Lines 28569 28569 \n==========================================\n+ Hits 22420 22963 +543 \n+ Misses 6149 5606 -543 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6777?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/commands/serving.py](https://codecov.io/gh/huggingface/transformers/pull/6777/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb21tYW5kcy9zZXJ2aW5nLnB5) | `55.88% <100.00%> (+55.88%)` | :arrow_up: |\n| [src/transformers/modeling\\_tf\\_t5.py](https://codecov.io/gh/huggingface/transformers/pull/6777/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl90NS5weQ==) | `26.84% <0.00%> (-64.10%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_flaubert.py](https://codecov.io/gh/huggingface/transformers/pull/6777/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9mbGF1YmVydC5weQ==) | `24.53% <0.00%> (-63.81%)` | :arrow_down: |\n| [src/transformers/modeling\\_marian.py](https://codecov.io/gh/huggingface/transformers/pull/6777/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19tYXJpYW4ucHk=) | `60.00% <0.00%> (-30.00%)` | :arrow_down: |\n| [src/transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/6777/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19yb2JlcnRhLnB5) | `77.37% <0.00%> (-19.71%)` | :arrow_down: |\n| [src/transformers/activations.py](https://codecov.io/gh/huggingface/transformers/pull/6777/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9hY3RpdmF0aW9ucy5weQ==) | `85.00% <0.00%> (-5.00%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6777/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl91dGlscy5weQ==) | `85.01% <0.00%> (-2.29%)` | :arrow_down: |\n| [src/transformers/modeling\\_bart.py](https://codecov.io/gh/huggingface/transformers/pull/6777/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19iYXJ0LnB5) | `95.05% <0.00%> (-0.35%)` | :arrow_down: |\n| [src/transformers/tokenization\\_utils\\_base.py](https://codecov.io/gh/huggingface/transformers/pull/6777/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHNfYmFzZS5weQ==) | `93.49% <0.00%> (-0.28%)` | :arrow_down: |\n| [src/transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6777/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9maWxlX3V0aWxzLnB5) | `82.41% <0.00%> (-0.26%)` | :arrow_down: |\n| ... and [21 more](https://codecov.io/gh/huggingface/transformers/pull/6777/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6777?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/6777?src=pr&el=footer). Last update [42fddac...6fd85c4](https://codecov.io/gh/huggingface/transformers/pull/6777?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6776?src=pr&el=h1) Report\n> Merging [#6776](https://codecov.io/gh/huggingface/transformers/pull/6776?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/42fddacd1cac3cc57c3326aa51a409f5090b1261?el=desc) will **decrease** coverage by `0.99%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6776?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6776 +/- ##\n==========================================\n- Coverage 78.47% 77.48% -1.00% \n==========================================\n Files 157 157 \n Lines 28569 28569 \n==========================================\n- Hits 22420 22137 -283 \n- Misses 6149 6432 +283 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6776?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/modeling\\_reformer.py](https://codecov.io/gh/huggingface/transformers/pull/6776/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19yZWZvcm1lci5weQ==) | `16.87% <0.00%> (-79.30%)` | :arrow_down: |\n| [src/transformers/configuration\\_reformer.py](https://codecov.io/gh/huggingface/transformers/pull/6776/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX3JlZm9ybWVyLnB5) | `21.62% <0.00%> (-78.38%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6776/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9vcGVuYWkucHk=) | `22.58% <0.00%> (-72.26%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_flaubert.py](https://codecov.io/gh/huggingface/transformers/pull/6776/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9mbGF1YmVydC5weQ==) | `24.53% <0.00%> (-63.81%)` | :arrow_down: |\n| [src/transformers/modeling\\_marian.py](https://codecov.io/gh/huggingface/transformers/pull/6776/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19tYXJpYW4ucHk=) | `60.00% <0.00%> (-30.00%)` | :arrow_down: |\n| [src/transformers/tokenization\\_ctrl.py](https://codecov.io/gh/huggingface/transformers/pull/6776/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fY3RybC5weQ==) | `78.64% <0.00%> (-17.48%)` | :arrow_down: |\n| [src/transformers/tokenization\\_xlm\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/6776/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25feGxtX3JvYmVydGEucHk=) | `84.52% <0.00%> (-10.72%)` | :arrow_down: |\n| [src/transformers/activations.py](https://codecov.io/gh/huggingface/transformers/pull/6776/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9hY3RpdmF0aW9ucy5weQ==) | `85.00% <0.00%> (-5.00%)` | :arrow_down: |\n| [src/transformers/tokenization\\_dpr.py](https://codecov.io/gh/huggingface/transformers/pull/6776/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fZHByLnB5) | `53.15% <0.00%> (-4.51%)` | :arrow_down: |\n| [src/transformers/modeling\\_bart.py](https://codecov.io/gh/huggingface/transformers/pull/6776/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19iYXJ0LnB5) | `95.05% <0.00%> (-0.35%)` | :arrow_down: |\n| ... and [14 more](https://codecov.io/gh/huggingface/transformers/pull/6776/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6776?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/6776?src=pr&el=footer). Last update [42fddac...c769370](https://codecov.io/gh/huggingface/transformers/pull/6776?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

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

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6775?src=pr&el=h1) Report\n> Merging [#6775](https://codecov.io/gh/huggingface/transformers/pull/6775?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/42fddacd1cac3cc57c3326aa51a409f5090b1261?el=desc) will **increase** coverage by `1.22%`.\n> The diff coverage is `13.33%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6775?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6775 +/- ##\n==========================================\n+ Coverage 78.47% 79.70% +1.22% \n==========================================\n Files 157 157 \n Lines 28569 28579 +10 \n==========================================\n+ Hits 22420 22779 +359 \n+ Misses 6149 5800 -349 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6775?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/data/processors/squad.py](https://codecov.io/gh/huggingface/transformers/pull/6775/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9kYXRhL3Byb2Nlc3NvcnMvc3F1YWQucHk=) | `27.59% <13.33%> (-0.54%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_flaubert.py](https://codecov.io/gh/huggingface/transformers/pull/6775/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9mbGF1YmVydC5weQ==) | `24.53% <0.00%> (-63.81%)` | :arrow_down: |\n| [src/transformers/modeling\\_marian.py](https://codecov.io/gh/huggingface/transformers/pull/6775/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19tYXJpYW4ucHk=) | `60.00% <0.00%> (-30.00%)` | :arrow_down: |\n| [src/transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/6775/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19yb2JlcnRhLnB5) | `77.37% <0.00%> (-19.71%)` | :arrow_down: |\n| [src/transformers/activations.py](https://codecov.io/gh/huggingface/transformers/pull/6775/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9hY3RpdmF0aW9ucy5weQ==) | `85.00% <0.00%> (-5.00%)` | :arrow_down: |\n| [src/transformers/tokenization\\_dpr.py](https://codecov.io/gh/huggingface/transformers/pull/6775/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fZHByLnB5) | `53.15% <0.00%> (-4.51%)` | :arrow_down: |\n| [src/transformers/modeling\\_bart.py](https://codecov.io/gh/huggingface/transformers/pull/6775/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19iYXJ0LnB5) | `95.05% <0.00%> (-0.35%)` | :arrow_down: |\n| [src/transformers/tokenization\\_utils\\_base.py](https://codecov.io/gh/huggingface/transformers/pull/6775/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHNfYmFzZS5weQ==) | `93.49% <0.00%> (-0.28%)` | :arrow_down: |\n| [src/transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6775/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9maWxlX3V0aWxzLnB5) | `82.41% <0.00%> (-0.26%)` | :arrow_down: |\n| [src/transformers/pipelines.py](https://codecov.io/gh/huggingface/transformers/pull/6775/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9waXBlbGluZXMucHk=) | `80.07% <0.00%> (+0.12%)` | :arrow_up: |\n| ... and [10 more](https://codecov.io/gh/huggingface/transformers/pull/6775/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6775?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/6775?src=pr&el=footer). Last update [42fddac...54cbfb1](https://codecov.io/gh/huggingface/transformers/pull/6775?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "@LysandreJik @mfuntowicz \r\nJust checking in to see if this PR is good, or does it need some more improvements?\r\n\r\nThanks", "Hi @bdalal,\n\nWill have a look at it ASAP, sorry for the delay \n\nMorgan", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

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

[ "Done!\r\n```python\r\n# Config values that vary between checkpoints: for testing and conversion\r\ntask_specific_params = {\r\n # These are task specific params for pegasus-large and normal params for finetuned checkpoints\r\n \"summarization_xsum\": {\"length_penalty\": 0.8, \"max_length\": 64, \"max_position_embeddings\": 512},\r\n \"summarization_cnn_dailymail\": {\"length_penalty\": 0.8, \"max_length\": 128, \"max_position_embeddings\": 1024},\r\n \"summarization_newsroom\": {\"length_penalty\": 0.8, \"max_length\": 128, \"max_position_embeddings\": 512},\r\n \"summarization_wikihow\": {\"length_penalty\": 0.6, \"max_length\": 256, \"max_position_embeddings\": 512},\r\n \"summarization_multi_news\": {\"length_penalty\": 0.8, \"max_length\": 256, \"max_position_embeddings\": 1024},\r\n \"summarization_reddit_tifu\": {\"length_penalty\": 0.6, \"max_length\": 128, \"max_position_embeddings\": 512},\r\n \"summarization_big_patent\": {\"length_penalty\": 0.7, \"max_length\": 256, \"max_position_embeddings\": 1024},\r\n \"summarization_arxiv\": {\"length_penalty\": 0.8, \"max_length\": 256, \"max_position_embeddings\": 1024},\r\n \"summarization_pubmed\": {\"length_penalty\": 0.8, \"max_length\": 256, \"max_position_embeddings\": 1024},\r\n \"summarization_gigaword\": {\"length_penalty\": 0.6, \"max_length\": 32, \"max_position_embeddings\": 128},\r\n \"summarization_aeslc\": {\"length_penalty\": 0.6, \"max_length\": 32, \"max_position_embeddings\": 512},\r\n \"summarization_billsum\": {\"length_penalty\": 0.6, \"max_length\": 256, \"max_position_embeddings\": 1024},\r\n # this last entry is useless -- just for consistency\r\n \"summarization_large\": {\"length_penalty\": 0.8, \"max_length\": 256, \"max_position_embeddings\": 1024},\r\n}\r\n```" ]

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

[]

[ "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" ]

[]

[]

[ "Answered in #6752" ]

[]

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

[ "This looks super interesting @jlei2 ! I openen a PR #6877 to see if changing to EinusmDense speeds up the runtime...In the PR I cannot see an improvement - could you take a look and maybe comment on how you changed the TF Bert code in transformers to benchmark your changes? :-) ", "Hi Patrick, thank you for taking time to investigate this issue! Your PR is not completely correct. Here is my draft [PR](https://github.com/jlei2/transformers/pull/1). DenseEinsum actually is the combination of matrix reshape/transpose and dense layer. So we need to be careful when replacing keras.dense with DenseEinsum, otherwise we will not be able to get the same output as we can get before this change.\r\n\r\nI tested on my end that there is no speedup after this change **at runtime**, using the huggingface bechmark command tool given by you. But I did see expected speedup after exporting DenseEinsum models. I used tensorflow profiling tool(https://www.tensorflow.org/tfx/serving/tensorboard) to benchmark the inference time of exported model when serving on TF-serving 2.2.0. So actually I don't have my own benchmark code.\r\n\r\nDevice: 1 GPU-V100\r\nBatch Size: 1\r\nModel: Bert-base FP32\r\n\r\n<img width=\"220\" alt=\"Screen Shot 2020-09-02 at 1 52 08 PM\" src=\"https://user-images.githubusercontent.com/70337521/92035460-981e9200-ed23-11ea-8141-a5f8e600559a.png\"> <img width=\"220\" alt=\"Screen Shot 2020-09-02 at 1 52 28 PM\" src=\"https://user-images.githubusercontent.com/70337521/92035489-a2409080-ed23-11ea-9414-dd2bb6e8ea68.png\">\r\nSo the left picture is from the Huggingface model after applying my PR. The right one is from original Huggingface model using current master. You can see that there is almost 100% speedup. So I suspect this issue only happens on exported model for tf-serving. \r\n\r\nThe code I used to export saved model:\r\n```\r\nimport tensorflow as tf\r\nimport transformers\r\n\r\nfrom transformers import BertModel\r\npt_model_path = 'bert-base-uncased'\r\n\r\nmodel = BertModel.from_pretrained('bert-base-uncased')\r\nmodel.save_pretrained(pt_model_path)\r\n\r\nMAX_LEN = 128\r\nmodel_path = 'saved_model/tmp_model'\r\ninput_ids = tf.keras.layers.Input((MAX_LEN,), dtype=tf.int32, name='input_ids')\r\nattention_mask = tf.keras.layers.Input((MAX_LEN,), dtype=tf.int32, name='attention_mask')\r\ntoken_type_ids = tf.keras.layers.Input((MAX_LEN,), dtype=tf.int32, name='token_type_ids')\r\nbase_model = transformers.TFBertModel.from_pretrained(pt_model_path, output_hidden_states=False, from_pt=True)\r\nbase_output = base_model.bert([input_ids, attention_mask, token_type_ids])\r\nseq_out, pool_out = base_output[0], base_output[1]\r\nbase_model.trainable = False\r\nmodel = tf.keras.models.Model(inputs=[input_ids, attention_mask, token_type_ids], outputs=[pool_out])\r\nmodel.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])\r\nprint(model.summary())\r\nmodel.save(model_path, include_optimizer=False, save_format=\"tf\")\r\n```\r\n\r\nIs this something you will want to solve on your end? though we see that during runtime there is no difference.", "Hey @jlei2, \r\n\r\nThanks a lot for checking and your draft PR. I think we will be able to implement your proposition given that we can keep complete backwards compatibility. Do you have an idea by any chance why we can see the speed up only on TF-serving? \r\n\r\nAlso cc @jplu - this looks very interesting! ", "To be honest I have no idea about the root difference on TFS side. I can only observe that there are some CPU processes wasting time before the Matmul op. My feeling is that the MatMul op triggered by tf.keras.layers.Dense() may not be implemented to be very efficient on TF-serving. Though this issue seems to have more to do with tensorflow keras team or TFS team instead of Huggingface code base imo, it would be helpful to all Huggingface users if you are able to resolve this issue on your end.", "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", "Re-activated cc @jplu ", "Hey @jlei2!\r\n\r\nSorry for this long silence. We are currently working on this performance issue and to be sure to make the proper changes, can you share with us your benchmark script that you used to compute the values in the table?\r\n\r\nThanks!", "Hi @jplu , thanks for keeping an eye on this! Unfortunately I can not share the benchmark script I used to get the numbers in the table with you because it's confidential. But I can share how you can obtain the same slow-down observation using Tensorflow Profiler tool (to get the two Performance Summary Screenshots I uploaded above 15.7ms vs 28.8ms).\r\n\r\n### Export baseline and modified SavedModel\r\n**Baseline transformers repo**: The commit I used is the same as the master branch in my fork (https://github.com/jlei2/transformers)\r\n\r\n**Modified transformers repo using tf.keras.layers.experimental.EinsumDense**: (https://github.com/jlei2/transformers/tree/dense-einsum-tf2.3)\r\n\r\n**The code to export the model**: \r\n```\r\nimport tensorflow as tf\r\nimport transformers\r\n\r\nfrom transformers import BertModel\r\npt_model_path = 'bert-base-uncased'\r\n\r\nmodel = BertModel.from_pretrained('bert-base-uncased')\r\nmodel.save_pretrained(pt_model_path)\r\n\r\nMAX_LEN = 128\r\nmodel_path = 'saved_model/tmp_model'\r\ninput_ids = tf.keras.layers.Input((MAX_LEN,), dtype=tf.int32, name='input_ids')\r\nattention_mask = tf.keras.layers.Input((MAX_LEN,), dtype=tf.int32, name='attention_mask')\r\ntoken_type_ids = tf.keras.layers.Input((MAX_LEN,), dtype=tf.int32, name='token_type_ids')\r\nbase_model = transformers.TFBertModel.from_pretrained(pt_model_path, output_hidden_states=False, from_pt=True)\r\nbase_output = base_model.bert([input_ids, attention_mask, token_type_ids])\r\nseq_out, pool_out = base_output[0], base_output[1]\r\nbase_model.trainable = False\r\nmodel = tf.keras.models.Model(inputs=[input_ids, attention_mask, token_type_ids], outputs=[pool_out])\r\nmodel.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])\r\nprint(model.summary())\r\nmodel.save(model_path, include_optimizer=False, save_format=\"tf\")\r\n```\r\n\r\nThen on a machine with 1 GPU-V100 and TF-Serving-gpu== 2.3.0 and Tensorflow==2.3.0:\r\n### Spin up SavedModel:\r\n```\r\nexport LD_LIBRARY_PATH=/usr/local/cuda/extras/CUPTI/lib64:$LD_LIBRARY_PATH\r\nexport MODEL_DIR=your_model_dir\r\nexport MODEL_NAME=your_model_name\r\nnohup tensorflow_model_server --rest_api_port=8501 --model_name=$MODEL_NAME --model_base_path=$MODEL_DIR >server.log 2>&1\r\n```\r\n\r\n### Spin up Tensorboard\r\n```\r\ntensorboard --logdir ~/logs/inference_demo/ --port your_port_number --bind_all\r\n```\r\n\r\nGo to the profile page on Tensorboard\r\n<img width=\"447\" alt=\"Screen Shot 2020-12-01 at 8 58 18 PM\" src=\"https://user-images.githubusercontent.com/70337521/100830662-03f2c280-3419-11eb-95bc-f0d70caec6db.png\">\r\n\r\nAfter clicking capture, I would send request to TFS to profile the whole process\r\n\r\n**Code to send request to served model**:\r\n```\r\nimport json\r\nimport requests\r\n\r\nBATCH_SIZE = 1\r\nMAX_SEQ_LEN = 128\r\nbatch = []\r\nMODEL_NAME = your_model_name\r\n\r\nfor _ in range(BATCH_SIZE):\r\n batch.append({\"input_ids\":[999] * MAX_SEQ_LEN, \"attention_mask\":[1] * MAX_SEQ_LEN, \"token_type_ids\":[0] * MAX_SEQ_LEN})\r\n\r\ninput_data = {\"instances\": batch}\r\nr = requests.post(\"http://localhost:8501/v1/models/%s:predict\"%(MODEL_NAME), data=json.dumps(input_data))\r\nprint(r.text)\r\n```\r\nI would run this scripts for several times first to warm up the model and then start to profile formally. \r\n\r\nAnd finally you will see profiling results on Tensorboard UI page just like what I uploaded.\r\n\r\nHope this could be helpful to you!", "Actually I write a very simple benchmark script that can show the difference:\r\n\r\n```\r\nimport json\r\nimport requests\r\nimport time\r\nBATCH_SIZE = 1\r\nMAX_SEQ_LEN = 128\r\nbatch = []\r\nMODEL_NAME = your_model_name\r\n\r\nfor _ in range(BATCH_SIZE):\r\n\tbatch.append({\"input_ids\":[999] * MAX_SEQ_LEN, \"attention_mask\":[1] * MAX_SEQ_LEN, \"token_type_ids\":[0] * MAX_SEQ_LEN})\r\n\r\ninput_data = {\"instances\": batch}\r\n\r\nstart = time.time()\r\nfor _ in range(100):\r\n\tr = requests.post(\"http://localhost:8501/v1/models/%s:predict\"%(MODEL_NAME), data=json.dumps(input_data))\r\nend = time.time()\r\nprint(end-start)\r\n```\r\nBaseline time: ~2.8s\r\nMy version's time: ~1.5s.\r\n\r\nSo we can easily see ~2x speed up.", "Awesome!! Thanks a lot @jlei2!! This is already a good start to check the differences. I will let you know here once we have done the changes!", "@jlei2 I have open a PR for integrating this change. Unfortunately, as I'm on Windows, the GPU profiling is not yet available in WSL, can you clone my branch and run it on your side with your own benchmark in order to be sure that it looks ok.\r\n\r\nThere is a small update on the code to create the saved model:\r\n```\r\nimport tensorflow as tf\r\nimport transformers\r\n\r\nMAX_LEN = 128\r\nmodel_path = 'saved_model/tmp_model'\r\ninput_ids = tf.keras.layers.Input((MAX_LEN,), dtype=tf.int32, name='input_ids')\r\nattention_mask = tf.keras.layers.Input((MAX_LEN,), dtype=tf.int32, name='attention_mask')\r\ntoken_type_ids = tf.keras.layers.Input((MAX_LEN,), dtype=tf.int32, name='token_type_ids')\r\nbase_model = transformers.TFBertModel.from_pretrained(\"bert-base-uncased\")\r\ninputs = {\"input_ids\": input_ids, \"attention_mask\": attention_mask, \"token_type_ids\": token_type_ids}\r\nbase_output = base_model.bert(inputs)\r\nseq_out, pool_out = base_output[0], base_output[1]\r\nbase_model.trainable = False\r\nmodel = tf.keras.models.Model(inputs=inputs, outputs=[pool_out])\r\nmodel.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])\r\nprint(model.summary())\r\nmodel.save(model_path, include_optimizer=False, save_format=\"tf\")\r\n```\r\n\r\nNo need anymore to load from the PT model, thanks to some update we have applied on the loading weights mechanism.", "Hi @jplu ,\r\n\r\nThanks for opening the PR to integrate this change. I have cloned your branch and run benchmark on my side. The results are as expected!\r\n\r\n**latency: (ms/batch) on 1 GPU-V100**\r\nBatch Size | current master | tf-einsumdense |\r\n-- | -- | --\r\n1 | 20.9 | 6.26\r\n2 | 24.1 | 8.68\r\n4 | 27.6 | 13.1\r\n8 | 36.3 | 21.5\r\n16 | 58.8 | 42.3\r\n32 | 94.7 | 80.4\r\n64 | 170 | 156\r\n128 | 321 | 309\r\n\r\nAnd on GPU profiling, obtained the same observation like I posted in the very first comment: GPU computation is continuous and compact. It is not cut off by CPU process anymore.\r\n\r\n**current master, batch_size=128**\r\n\r\n\r\n**tf-einsumdense, batch_size=128**\r\n\r\n\r\n\r\n\r\n" ]

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

[]

[ "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" ]

[]

[]

[ "Command for running `finetune_trainer.py`\r\n\r\n```bash\r\npython finetune_trainer.py \\\r\n --model_name_or_path sshleifer/bart-tiny-random \\\r\n --data_dir xsum \\\r\n --output_dir test \\\r\n --overwrite_output_dir \\\r\n --n_train 8 \\\r\n --n_val 8 \\\r\n --max_source_length 512 \\\r\n --max_target_length 56 \\\r\n --val_max_target_length 56 \\\r\n --do_train \\\r\n --do_eval \\\r\n --num_train_epochs 2 \\\r\n --per_device_train_batch_size 4 \\\r\n --per_device_eval_batch_size 4 \\\r\n --evaluate_during_training \\\r\n --predict_from_generate \\\r\n --logging_steps 2 \\\r\n --save_steps 2 \\\r\n --eval_steps 2 \\\r\n --sortish_sampler \\\r\n```", "Note: Eventually we need to refactor seq2seq/README.md to accommodate this", "This looks great. Very excited to try this out with the `EncoderDecoder` model.", "LGTM pending resolution of padding mystery.", "Great work @patil-suraj !" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6768?src=pr&el=h1) Report\n> Merging [#6768](https://codecov.io/gh/huggingface/transformers/pull/6768?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/42fddacd1cac3cc57c3326aa51a409f5090b1261?el=desc) will **increase** coverage by `1.17%`.\n> The diff coverage is `38.57%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6768?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6768 +/- ##\n==========================================\n+ Coverage 78.47% 79.65% +1.17% \n==========================================\n Files 157 157 \n Lines 28569 28625 +56 \n==========================================\n+ Hits 22420 22800 +380 \n+ Misses 6149 5825 -324 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6768?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/trainer.py](https://codecov.io/gh/huggingface/transformers/pull/6768/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90cmFpbmVyLnB5) | `51.85% <31.48%> (-1.81%)` | :arrow_down: |\n| [src/transformers/modeling\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6768/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ191dGlscy5weQ==) | `86.66% <62.50%> (-0.84%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_xlm.py](https://codecov.io/gh/huggingface/transformers/pull/6768/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl94bG0ucHk=) | `88.42% <0.00%> (-4.85%)` | :arrow_down: |\n| [src/transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6768/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9maWxlX3V0aWxzLnB5) | `82.41% <0.00%> (-0.26%)` | :arrow_down: |\n| [src/transformers/modeling\\_bart.py](https://codecov.io/gh/huggingface/transformers/pull/6768/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19iYXJ0LnB5) | `95.56% <0.00%> (+0.17%)` | :arrow_up: |\n| [src/transformers/configuration\\_bart.py](https://codecov.io/gh/huggingface/transformers/pull/6768/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9jb25maWd1cmF0aW9uX2JhcnQucHk=) | `94.00% <0.00%> (+4.00%)` | :arrow_up: |\n| [src/transformers/tokenization\\_xlnet.py](https://codecov.io/gh/huggingface/transformers/pull/6768/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25feGxuZXQucHk=) | `90.09% <0.00%> (+23.42%)` | :arrow_up: |\n| [src/transformers/modeling\\_tf\\_transfo\\_xl.py](https://codecov.io/gh/huggingface/transformers/pull/6768/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl90cmFuc2ZvX3hsLnB5) | `88.13% <0.00%> (+68.28%)` | :arrow_up: |\n| [...c/transformers/modeling\\_tf\\_transfo\\_xl\\_utilities.py](https://codecov.io/gh/huggingface/transformers/pull/6768/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl90cmFuc2ZvX3hsX3V0aWxpdGllcy5weQ==) | `86.00% <0.00%> (+76.00%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6768?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/6768?src=pr&el=footer). Last update [42fddac...4becfac](https://codecov.io/gh/huggingface/transformers/pull/6768?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "I think even with domain-agnostic models we'd like to keep the configuration, no? I'm not sure the trainer would behave correctly without a configuration, so if we want to remove the dependency towards configurations, we might as well do it all at once, right?\r\n\r\nWould the goal be to have the trainer accept all `nn.Module`s?", "Like agreed upon internally, we will move to Trainer accepting models instantiating a base abstractclass/conforming to some protocol. I think the config will be in the required field but have to work a bit more on this to be sure.\r\n\r\nIn any case, this is work for a subsequent PR :-)" ]

[]

[]

[]

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

[]

[ "Hi! What are you using to launch your ditributed training? What script are you using? Could you show the command you used, and could you paste your environment information as required by the template? Thank you.", "## Environment info\r\n- `transformers` version: 2.11.0\r\n- Platform: Databricks\r\n- Python version: 3.6.9\r\n- PyTorch version: 1.3.1 \r\n- Tensorflow version : 2.0\r\n- Using GPU in script: yes\r\n- Using distributed or parallel set-up in script: try to\r\n\r\n\r\n## Information\r\n\r\nI am using the model xlm-roberta-base. The tasks I am working on is a further train on my own dataset.\r\n\r\nThe problem arises when using the script examples/language_modeling/run_language_modeling.py with the following command :\r\n\r\n```\r\npython transformers/examples/language-modeling/run_language_modeling.py \r\n --model_type xlm-roberta \r\n --model_name_or_path xlm-roberta-base \r\n --train_data_file data/processed/piaf.txt \r\n --output_dir ./output --learning_rate 0.1 \r\n --per_gpu_train_batch_size 2 \r\n --local_rank 0 \r\n --num_train_epochs 1 \r\n --do_train \r\n --mlm\r\n```", "Okay, could you try using the `torch.distributed.launch` utility used to launch a distributed training?\r\n\r\nYour command, given that you have 8 GPUs, would be:\r\n\r\n```\r\npython -m torch.distributed.launch \\\r\n --nproc_per_node 8 transformers/examples/language-modeling/run_language_modeling.py \r\n --model_type xlm-roberta \r\n --model_name_or_path xlm-roberta-base \r\n --train_data_file data/processed/piaf.txt \r\n --output_dir ./output --learning_rate 0.1 \r\n --per_gpu_train_batch_size 2 \r\n --local_rank 0 \r\n --num_train_epochs 1 \r\n --do_train \r\n --mlm\r\n```\r\n\r\nFeel free to modify this part: `--nproc_per_node 8` according to the number of GPUs.", "This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.\n" ]

[]

[]

[]

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

[]

[ "Any idea why there is one failed test case ?\r\nI have checked the code on the notebook and it runs without any issue on Colab.", "This seems completely unrelated, just relaunched the failed test.", "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6764?src=pr&el=h1) Report\n> Merging [#6764](https://codecov.io/gh/huggingface/transformers/pull/6764?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/4bd7be9a4268221d2a0000c7e8033aaeb365c03b?el=desc) will **decrease** coverage by `0.04%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6764?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6764 +/- ##\n==========================================\n- Coverage 79.74% 79.70% -0.05% \n==========================================\n Files 157 157 \n Lines 28479 28479 \n==========================================\n- Hits 22712 22698 -14 \n- Misses 5767 5781 +14 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6764?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/tokenization\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/6764/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fcm9iZXJ0YS5weQ==) | `87.67% <0.00%> (-10.96%)` | :arrow_down: |\n| [src/transformers/tokenization\\_utils\\_base.py](https://codecov.io/gh/huggingface/transformers/pull/6764/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHNfYmFzZS5weQ==) | `86.58% <0.00%> (-7.19%)` | :arrow_down: |\n| [src/transformers/tokenization\\_transfo\\_xl.py](https://codecov.io/gh/huggingface/transformers/pull/6764/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdHJhbnNmb194bC5weQ==) | `38.73% <0.00%> (-3.76%)` | :arrow_down: |\n| [src/transformers/tokenization\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6764/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fb3BlbmFpLnB5) | `82.57% <0.00%> (-1.52%)` | :arrow_down: |\n| [src/transformers/tokenization\\_utils\\_fast.py](https://codecov.io/gh/huggingface/transformers/pull/6764/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHNfZmFzdC5weQ==) | `92.85% <0.00%> (-1.43%)` | :arrow_down: |\n| [src/transformers/tokenization\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/6764/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fYmVydC5weQ==) | `91.07% <0.00%> (-0.45%)` | :arrow_down: |\n| [src/transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6764/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHMucHk=) | `89.45% <0.00%> (-0.40%)` | :arrow_down: |\n| [src/transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6764/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9maWxlX3V0aWxzLnB5) | `82.41% <0.00%> (-0.26%)` | :arrow_down: |\n| [src/transformers/generation\\_tf\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6764/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9nZW5lcmF0aW9uX3RmX3V0aWxzLnB5) | `86.46% <0.00%> (-0.26%)` | :arrow_down: |\n| [src/transformers/tokenization\\_marian.py](https://codecov.io/gh/huggingface/transformers/pull/6764/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fbWFyaWFuLnB5) | `99.16% <0.00%> (+32.50%)` | :arrow_up: |\n| ... and [1 more](https://codecov.io/gh/huggingface/transformers/pull/6764/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6764?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/6764?src=pr&el=footer). Last update [4bd7be9...c7eb12c](https://codecov.io/gh/huggingface/transformers/pull/6764?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "Hi @agemagician, your model https://huggingface.co/Rostlab/prot_bert is not recognized by our Hosted inference API, probably because it doesn't have a `architectures` field in its config.json:\r\n\r\n<img width=\"565\" alt=\"Screenshot 2020-10-21 at 21 52 57\" src=\"https://user-images.githubusercontent.com/326577/96775413-849fb700-13b5-11eb-8075-11d7792548f6.png\">\r\n\r\n\r\nDo you mind if we update the config.json on the model hub? Alternatively, do you prefer doing it yourself?\r\n\r\nThanks!", "Hi @julien-c , No problem.\r\nI already have seen you updated it.\r\n\r\nHowever, it doesn't work properly because we don't lower case tokens:\r\nhttps://huggingface.co/Rostlab/prot_bert?text=A+T+G+%5BMASK%5D+C\r\n\r\nI have uploaded the \"tokenizer_config.json\" file which should fix it, but we have to wait until the model is no longer in memory and reload it.\r\n\r\nI have also did the same for our better bert model \"prot_bert_bfd\", and it is working fine:\r\nhttps://huggingface.co/Rostlab/prot_bert_bfd?text=A+T+G+%5BMASK%5D+C\r\n\r\nThanks for your help.", "Now properly loaded, and looks great!\r\n\r\nhttps://huggingface.co/Rostlab/prot_bert?text=D+L+I+P+T+S+S+K+V+V+%5BMASK%5D+D+T+S+L+Q+V+K+K+A+F+F+A+L+V+T\r\n\r\n<img width=\"693\" alt=\"Screenshot 2020-10-21 at 22 56 44\" src=\"https://user-images.githubusercontent.com/326577/96786965-ff210480-13be-11eb-9133-375bb20e60a1.png\">\r\n\r\nThanks a lot @agemagician " ]

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

[]

[ "Same issue when running distilbart", "I face the same issue with PEGASUS", "I've run into this with several seq2seq models (Pegasus, BART, T5). Ironically, running with fp16 causes increased GPU memory usage.", "I solved my issue by downgrading Pytorch to 1.5.1 and installing Nvidia/Apex. ", "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 } ]

[]

[ "Yes! This is something we'll build, step by step. I'll post more about our roadmap for this in the coming weeks.", "> Yes! This is something we'll build, step by step. I'll post more about our roadmap for this in the coming weeks.\r\n\r\nMaybe you can create a web-based interface with input fields/drop downs that can generate the model card automatically. Just a thought. ", "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", "posting so that the bot doesn't close this issue.", "@prajjwal1 Have you checked out our new model storage/model versioning features on huggingface.co? You can also now edit your model card directly from the website (or from `git`) and we'll make this workflow more prominent (vs. adding model cards to `transformers`) in the coming days/weeks.\r\n\r\nFeedback welcome, here or on the Forum.", "No, I haven't. I am aware of it. I will check out soon. Loving the direction where HF is headed.", "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 issue has been automatically marked as stale and been closed because it has not had recent activity. Thank you for your contributions.\n\nIf you think this still needs to be addressed please comment on this thread." ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6761?src=pr&el=h1) Report\n> Merging [#6761](https://codecov.io/gh/huggingface/transformers/pull/6761?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/05e7150a53cc6c1571c0e3acb1b4d692737976d9?el=desc) will **decrease** coverage by `0.24%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6761?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6761 +/- ##\n==========================================\n- Coverage 79.70% 79.45% -0.25% \n==========================================\n Files 157 157 \n Lines 28479 28479 \n==========================================\n- Hits 22698 22627 -71 \n- Misses 5781 5852 +71 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6761?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/modeling\\_tf\\_electra.py](https://codecov.io/gh/huggingface/transformers/pull/6761/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9lbGVjdHJhLnB5) | `25.13% <0.00%> (-73.83%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6761/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl9vcGVuYWkucHk=) | `22.58% <0.00%> (-72.26%)` | :arrow_down: |\n| [src/transformers/tokenization\\_xlnet.py](https://codecov.io/gh/huggingface/transformers/pull/6761/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25feGxuZXQucHk=) | `66.66% <0.00%> (-23.43%)` | :arrow_down: |\n| [src/transformers/tokenization\\_ctrl.py](https://codecov.io/gh/huggingface/transformers/pull/6761/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fY3RybC5weQ==) | `78.64% <0.00%> (-17.48%)` | :arrow_down: |\n| [src/transformers/tokenization\\_reformer.py](https://codecov.io/gh/huggingface/transformers/pull/6761/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fcmVmb3JtZXIucHk=) | `81.66% <0.00%> (-13.34%)` | :arrow_down: |\n| [src/transformers/tokenization\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6761/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fb3BlbmFpLnB5) | `71.21% <0.00%> (-11.37%)` | :arrow_down: |\n| [src/transformers/modeling\\_t5.py](https://codecov.io/gh/huggingface/transformers/pull/6761/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190NS5weQ==) | `77.63% <0.00%> (-6.21%)` | :arrow_down: |\n| [src/transformers/modeling\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6761/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19vcGVuYWkucHk=) | `80.96% <0.00%> (-1.30%)` | :arrow_down: |\n| [src/transformers/modeling\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6761/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ191dGlscy5weQ==) | `87.50% <0.00%> (-0.56%)` | :arrow_down: |\n| [src/transformers/modeling\\_tf\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6761/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl91dGlscy5weQ==) | `86.97% <0.00%> (-0.33%)` | :arrow_down: |\n| ... and [9 more](https://codecov.io/gh/huggingface/transformers/pull/6761/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6761?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/6761?src=pr&el=footer). Last update [05e7150...88974cc](https://codecov.io/gh/huggingface/transformers/pull/6761?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

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

[]

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

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

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6759?src=pr&el=h1) Report\n> Merging [#6759](https://codecov.io/gh/huggingface/transformers/pull/6759?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/434936f34a8b3154a79564c87f4cb50f5d57e050?el=desc) will **increase** coverage by `0.03%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6759?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6759 +/- ##\n==========================================\n+ Coverage 79.47% 79.51% +0.03% \n==========================================\n Files 157 157 \n Lines 28479 28479 \n==========================================\n+ Hits 22635 22644 +9 \n+ Misses 5844 5835 -9 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6759?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/tokenization\\_ctrl.py](https://codecov.io/gh/huggingface/transformers/pull/6759/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fY3RybC5weQ==) | `78.64% <0.00%> (-17.48%)` | :arrow_down: |\n| [src/transformers/tokenization\\_reformer.py](https://codecov.io/gh/huggingface/transformers/pull/6759/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fcmVmb3JtZXIucHk=) | `81.66% <0.00%> (-13.34%)` | :arrow_down: |\n| [src/transformers/tokenization\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6759/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fb3BlbmFpLnB5) | `71.21% <0.00%> (-12.88%)` | :arrow_down: |\n| [src/transformers/tokenization\\_xlm\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/6759/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25feGxtX3JvYmVydGEucHk=) | `84.52% <0.00%> (-10.72%)` | :arrow_down: |\n| [src/transformers/tokenization\\_albert.py](https://codecov.io/gh/huggingface/transformers/pull/6759/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fYWxiZXJ0LnB5) | `87.50% <0.00%> (+58.65%)` | :arrow_up: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6759?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/6759?src=pr&el=footer). Last update [434936f...fd0b38d](https://codecov.io/gh/huggingface/transformers/pull/6759?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n", "really cool model card, thanks for sharing", "File path was incorrect but I fixed it in d822ab636b6a14ed50f7bca0797c1de42c19de61. Thanks!", "@julien-c thanks" ]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6758?src=pr&el=h1) Report\n> Merging [#6758](https://codecov.io/gh/huggingface/transformers/pull/6758?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/a75c64d80c76c3dc71f735d9197a4a601847e0cd?el=desc) will **decrease** coverage by `0.00%`.\n> The diff coverage is `100.00%`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6758?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6758 +/- ##\n==========================================\n- Coverage 78.96% 78.96% -0.01% \n==========================================\n Files 157 157 \n Lines 28486 28488 +2 \n==========================================\n+ Hits 22495 22496 +1 \n- Misses 5991 5992 +1 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6758?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/modeling\\_bart.py](https://codecov.io/gh/huggingface/transformers/pull/6758/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19iYXJ0LnB5) | `95.57% <100.00%> (+0.01%)` | :arrow_up: |\n| [src/transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6758/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9maWxlX3V0aWxzLnB5) | `82.41% <0.00%> (-0.26%)` | :arrow_down: |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6758?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/6758?src=pr&el=footer). Last update [a75c64d...06e7cc5](https://codecov.io/gh/huggingface/transformers/pull/6758?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]

[]

[]

[]

[]

[]

[ "# [Codecov](https://codecov.io/gh/huggingface/transformers/pull/6756?src=pr&el=h1) Report\n> Merging [#6756](https://codecov.io/gh/huggingface/transformers/pull/6756?src=pr&el=desc) into [master](https://codecov.io/gh/huggingface/transformers/commit/61b9ed80742f564dc522783a33bf001d6d871a2c?el=desc) will **increase** coverage by `0.16%`.\n> The diff coverage is `n/a`.\n\n[](https://codecov.io/gh/huggingface/transformers/pull/6756?src=pr&el=tree)\n\n```diff\n@@ Coverage Diff @@\n## master #6756 +/- ##\n==========================================\n+ Coverage 79.65% 79.82% +0.16% \n==========================================\n Files 157 157 \n Lines 28479 28479 \n==========================================\n+ Hits 22686 22734 +48 \n+ Misses 5793 5745 -48 \n```\n\n\n| [Impacted Files](https://codecov.io/gh/huggingface/transformers/pull/6756?src=pr&el=tree) | Coverage Δ | |\n|---|---|---|\n| [src/transformers/modeling\\_tf\\_xlm.py](https://codecov.io/gh/huggingface/transformers/pull/6756/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ190Zl94bG0ucHk=) | `18.94% <0.00%> (-74.32%)` | :arrow_down: |\n| [src/transformers/modeling\\_roberta.py](https://codecov.io/gh/huggingface/transformers/pull/6756/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9tb2RlbGluZ19yb2JlcnRhLnB5) | `77.37% <0.00%> (-19.71%)` | :arrow_down: |\n| [src/transformers/file\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6756/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9maWxlX3V0aWxzLnB5) | `82.66% <0.00%> (+0.25%)` | :arrow_up: |\n| [src/transformers/tokenization\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6756/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHMucHk=) | `89.84% <0.00%> (+0.39%)` | :arrow_up: |\n| [src/transformers/tokenization\\_bert.py](https://codecov.io/gh/huggingface/transformers/pull/6756/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fYmVydC5weQ==) | `91.51% <0.00%> (+0.44%)` | :arrow_up: |\n| [src/transformers/tokenization\\_utils\\_fast.py](https://codecov.io/gh/huggingface/transformers/pull/6756/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHNfZmFzdC5weQ==) | `94.28% <0.00%> (+1.42%)` | :arrow_up: |\n| [src/transformers/tokenization\\_openai.py](https://codecov.io/gh/huggingface/transformers/pull/6756/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fb3BlbmFpLnB5) | `84.09% <0.00%> (+1.51%)` | :arrow_up: |\n| [src/transformers/generation\\_tf\\_utils.py](https://codecov.io/gh/huggingface/transformers/pull/6756/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy9nZW5lcmF0aW9uX3RmX3V0aWxzLnB5) | `86.46% <0.00%> (+3.00%)` | :arrow_up: |\n| [src/transformers/tokenization\\_transfo\\_xl.py](https://codecov.io/gh/huggingface/transformers/pull/6756/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdHJhbnNmb194bC5weQ==) | `42.48% <0.00%> (+3.75%)` | :arrow_up: |\n| [src/transformers/tokenization\\_utils\\_base.py](https://codecov.io/gh/huggingface/transformers/pull/6756/diff?src=pr&el=tree#diff-c3JjL3RyYW5zZm9ybWVycy90b2tlbml6YXRpb25fdXRpbHNfYmFzZS5weQ==) | `93.76% <0.00%> (+7.18%)` | :arrow_up: |\n| ... and [2 more](https://codecov.io/gh/huggingface/transformers/pull/6756/diff?src=pr&el=tree-more) | |\n\n------\n\n[Continue to review full report at Codecov](https://codecov.io/gh/huggingface/transformers/pull/6756?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/6756?src=pr&el=footer). Last update [61b9ed8...7ba95ec](https://codecov.io/gh/huggingface/transformers/pull/6756?src=pr&el=lastupdated). Read the [comment docs](https://docs.codecov.io/docs/pull-request-comments).\n" ]