How to use transformers - 10 common examples

train_sampler = RandomSampler(train_dataset) if args.local_rank == -1 else DistributedSampler(train_dataset)
    train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size)

    if args.max_steps > 0:
        t_total = args.max_steps
        args.num_train_epochs = args.max_steps // (len(train_dataloader) // args.gradient_accumulation_steps) + 1
    else:
        t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs

    # Prepare optimizer and schedule (linear warmup and decay)
    no_decay = ['bias', 'LayerNorm.weight']
    optimizer_grouped_parameters = [
        {'params': [p for n, p in model.named_parameters() if p.requires_grad and not any(nd in n for nd in no_decay)], 'weight_decay': args.weight_decay},
        {'params': [p for n, p in model.named_parameters() if p.requires_grad and any(nd in n for nd in no_decay)], 'weight_decay': 0.0}
        ]
    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
    warmup_steps = args.warmup_samples // args.train_batch_size
    if args.lr_decay:
        scheduler = get_cosine_schedule_with_warmup(optimizer, warmup_steps=warmup_steps, t_total=t_total)
    else:
        scheduler = get_constant_schedule(optimizer, warmup_steps=warmup_steps)

    if args.fp16:
        try:
            from apex import amp
        except ImportError:
            raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.")
        model, optimizer = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level)

    # multi-gpu training (should be after apex fp16 initialization)
    if args.n_gpu > 1:
        model = torch.nn.DataParallel(model)

train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args["train_batch_size"])

        t_total = len(train_dataloader) // args["gradient_accumulation_steps"] * args["num_train_epochs"]

        no_decay = ["bias", "LayerNorm.weight"]
        optimizer_grouped_parameters = [
            {"params": [p for n, p in model.named_parameters() if not any(
                nd in n for nd in no_decay)], "weight_decay": args["weight_decay"]},
            {"params": [p for n, p in model.named_parameters() if any(
                nd in n for nd in no_decay)], "weight_decay": 0.0}
        ]

        warmup_steps = math.ceil(t_total * args["warmup_ratio"])
        args["warmup_steps"] = warmup_steps if args["warmup_steps"] == 0 else args["warmup_steps"]

        optimizer = AdamW(optimizer_grouped_parameters, lr=args["learning_rate"], eps=args["adam_epsilon"])
        scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args["warmup_steps"], num_training_steps=t_total)

        if args["fp16"]:
            try:
                from apex import amp
            except ImportError:
                raise ImportError(
                    "Please install apex from https://www.github.com/nvidia/apex to use fp16 training.")

            model, optimizer = amp.initialize(model, optimizer, opt_level=args["fp16_opt_level"])

        if args["n_gpu"] > 1:
            model = torch.nn.DataParallel(model)

        global_step = 0
        tr_loss, logging_loss = 0.0, 0.0

train_sampler = RandomSampler(train_dataset) if args.local_rank == -1 else DistributedSampler(train_dataset)
    train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size)

    if args.max_steps > 0:
        t_total = args.max_steps
        args.num_train_epochs = args.max_steps // (len(train_dataloader) // args.gradient_accumulation_steps) + 1
    else:
        t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs

    # Prepare optimizer and schedule (linear warmup and decay)
    no_decay = ['bias', 'LayerNorm.weight']
    optimizer_grouped_parameters = [
        {'params': [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)], 'weight_decay': args.weight_decay},
        {'params': [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], 'weight_decay': 0.0}
        ]
    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total)
    if args.fp16:
        try:
            from apex import amp
        except ImportError:
            raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.")
        model, optimizer = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level)

    # multi-gpu training (should be after apex fp16 initialization)
    if args.n_gpu > 1:
        model = torch.nn.DataParallel(model)

    # Distributed training (should be after apex fp16 initialization)
    if args.local_rank != -1:
        model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.local_rank],
                                                          output_device=args.local_rank,

params += list(model_tag.parameters())
    if opt.task_sc:
        params += list(model_class.parameters())
    params = list(filter(lambda p: p.requires_grad, params))
    named_params = []
    named_params += list(model_tag.named_parameters())
    if opt.task_sc:
        named_params += list(model_class.named_parameters())
    named_params = list(filter(lambda p: p[1].requires_grad, named_params))
    no_decay = ['bias', 'LayerNorm.bias', 'LayerNorm.weight']
    optimizer_grouped_parameters = [
        {'params': [p for n, p in named_params if not any(nd in n for nd in no_decay)], 'weight_decay': 0.01},
        {'params': [p for n, p in named_params if any(nd in n for nd in no_decay)], 'weight_decay': 0.0}
        ]
    num_train_optimization_steps = len(train_feats['data']) // opt.batchSize * opt.max_epoch
    optimizer = AdamW(optimizer_grouped_parameters, lr=opt.lr, correct_bias=False)  # To reproduce BertAdam specific behavior set correct_bias=False
    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=int(opt.warmup_proportion * num_train_optimization_steps), num_training_steps=num_train_optimization_steps)  # PyTorch scheduler

# prepare_inputs_for_bert(sentences, word_lengths)

def decode(data_feats, data_tags, data_class, output_path):
    data_index = np.arange(len(data_feats))
    losses = []
    TP, FP, FN, TN = 0.0, 0.0, 0.0, 0.0
    TP2, FP2, FN2, TN2 = 0.0, 0.0, 0.0, 0.0
    with open(output_path, 'w') as f:
        for j in range(0, len(data_index), opt.test_batchSize):
            if opt.testing:
                words, tags, raw_tags, classes, raw_classes, lens, line_nums = data_reader.get_minibatch_with_class(data_feats, data_tags, data_class, tag_to_idx, class_to_idx, data_index, j, opt.test_batchSize, add_start_end=opt.bos_eos, multiClass=opt.multiClass, keep_order=opt.testing, enc_dec_focus=opt.enc_dec, device=opt.device)
            else:
                words, tags, raw_tags, classes, raw_classes, lens = data_reader.get_minibatch_with_class(data_feats, data_tags, data_class, tag_to_idx, class_to_idx, data_index, j, opt.test_batchSize, add_start_end=opt.bos_eos, multiClass=opt.multiClass, keep_order=opt.testing, enc_dec_focus=opt.enc_dec, device=opt.device)

train_sampler = RandomSampler(train_dataset) if args.local_rank == -1 else DistributedSampler(train_dataset)
    train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size)

    if args.max_steps > 0:
        t_total = args.max_steps
        args.num_train_epochs = args.max_steps // (len(train_dataloader) // args.gradient_accumulation_steps) + 1
    else:
        t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs

    # Prepare optimizer and schedule (linear warmup and decay)
    no_decay = ['bias', 'LayerNorm.weight']
    optimizer_grouped_parameters = [
        {'params': [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)], 'weight_decay': args.weight_decay},
        {'params': [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], 'weight_decay': 0.0}
        ]
    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total)
    if args.fp16:
        try:
            from apex import amp
        except ImportError:
            raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.")
        model, optimizer = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level)

    # multi-gpu training (should be after apex fp16 initialization)
    if args.n_gpu > 1:
        model = torch.nn.DataParallel(model)

    # Distributed training (should be after apex fp16 initialization)
    if args.local_rank != -1:
        model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.local_rank],
                                                          output_device=args.local_rank,

train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size)

    if args.max_steps > 0:
        t_total = args.max_steps
        args.num_train_epochs = args.max_steps // (len(train_dataloader) // args.gradient_accumulation_steps) + 1
    else:
        t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs

    # Prepare optimizer and schedule (linear warmup and decay)
    no_decay = ['bias', 'LayerNorm.weight']
    optimizer_grouped_parameters = [
        {'params': [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)], 'weight_decay': args.weight_decay},
        {'params': [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], 'weight_decay': 0.0}
        ]
    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total)
    if args.fp16:
        try:
            from apex import amp
        except ImportError:
            raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.")
        model, optimizer = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level)

    # multi-gpu training (should be after apex fp16 initialization)
    if args.n_gpu > 1:
        model = torch.nn.DataParallel(model)

    # Distributed training (should be after apex fp16 initialization)
    if args.local_rank != -1:
        model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.local_rank],
                                                          output_device=args.local_rank,
                                                          find_unused_parameters=True)

args.num_train_epochs = args.max_steps // (len(train_dataloader) // args.gradient_accumulation_steps) + 1
    else:
        t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs

    # Prepare optimizer and schedule (linear warmup and decay)
    no_decay = ["bias", "LayerNorm.weight"]
    optimizer_grouped_parameters = [
        {
            "params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)],
            "weight_decay": args.weight_decay,
        },
        {"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
    ]

    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
    scheduler = get_linear_schedule_with_warmup(
        optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
    )
    if args.fp16:
        try:
            from apex import amp
        except ImportError:
            raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.")
        model, optimizer = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level)

    # multi-gpu training (should be after apex fp16 initialization)
    if args.n_gpu > 1:
        model = torch.nn.DataParallel(model)

    # Distributed training (should be after apex fp16 initialization)
    if args.local_rank != -1:
        model = torch.nn.parallel.DistributedDataParallel(

if self.args.max_steps > 0:
            t_total = self.args.max_steps
            self.args.num_train_epochs = self.args.max_steps // (len(train_dataloader) // self.args.gradient_accumulation_steps) + 1
        else:
            t_total = len(train_dataloader) // self.args.gradient_accumulation_steps * self.args.num_train_epochs

        # Prepare optimizer and schedule (linear warmup and decay)
        no_decay = ['bias', 'LayerNorm.weight']
        optimizer_grouped_parameters = [
            {'params': [p for n, p in self.model.named_parameters() if not any(nd in n for nd in no_decay)],
             'weight_decay': self.args.weight_decay},
            {'params': [p for n, p in self.model.named_parameters() if any(nd in n for nd in no_decay)], 'weight_decay': 0.0}
        ]
        optimizer = AdamW(optimizer_grouped_parameters, lr=self.args.learning_rate, eps=self.args.adam_epsilon)
        scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=self.args.warmup_steps, num_training_steps=t_total)

        # Train!
        logger.info("***** Running training *****")
        logger.info("  Num examples = %d", len(self.train_dataset))
        logger.info("  Num Epochs = %d", self.args.num_train_epochs)
        logger.info("  Total train batch size = %d", self.args.train_batch_size)
        logger.info("  Gradient Accumulation steps = %d", self.args.gradient_accumulation_steps)
        logger.info("  Total optimization steps = %d", t_total)
        logger.info("  Logging steps = %d", self.args.logging_steps)
        logger.info("  Save steps = %d", self.args.save_steps)

        global_step = 0
        tr_loss = 0.0
        self.model.zero_grad()

        train_iterator = trange(int(self.args.num_train_epochs), desc="Epoch")

# Prepare optimizer and schedule (linear warmup and decay)
    no_decay = ['bias', 'LayerNorm.weight']
    optimizer_grouped_parameters = [
        {'params': [p for n, p in model.named_parameters() if p.requires_grad and not any(nd in n for nd in no_decay)], 'weight_decay': args.weight_decay},
        {'params': [p for n, p in model.named_parameters() if p.requires_grad and any(nd in n for nd in no_decay)], 'weight_decay': 0.0}
        ]
    # Scale learning rate to num cores
    #args.learning_rate = args.learning_rate * xm.xrt_world_size()
    if args.sgd:
        optimizer = SGD(optimizer_grouped_parameters, lr=args.learning_rate)
    else:
        optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
    warmup_steps = args.warmup_samples // (args.train_batch_size * xm.xrt_world_size())
    if args.lr_decay:
        scheduler = get_linear_schedule_with_warmup(optimizer, warmup_steps=warmup_steps, t_total=t_total)
    elif args.lr_cosine:
        scheduler = get_cosine_with_hard_restarts_schedule_with_warmup(optimizer, warmup_steps=warmup_steps, t_total=t_total, cycles=args.num_train_epochs)
    else:
        scheduler = WarmupZeroSchedule(optimizer, warmup_steps=warmup_steps)

    # Train!
    tracker = xm.RateTracker()
    log_info("***** Running training *****")
    log_info("  Num Epochs = %d", args.num_train_epochs)
    log_info("  Instantaneous batch size per GPU = %d", args.per_gpu_train_batch_size)
    log_info("  Total train batch size (w. parallel, distributed & accumulation) = %d",
                   args.train_batch_size * args.gradient_accumulation_steps * (xm.xrt_world_size() if args.local_rank != -1 else 1))
    log_info("  Gradient Accumulation steps = %d", args.gradient_accumulation_steps)
    log_info("  Total optimization steps = %d", t_total)

    try:

train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size)

    if args.max_steps > 0:
        t_total = args.max_steps
        args.num_train_epochs = args.max_steps // (len(train_dataloader) // args.gradient_accumulation_steps) + 1
    else:
        t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs

    # Prepare optimizer and schedule (linear warmup and decay)
    no_decay = ['bias', 'LayerNorm.weight']
    optimizer_grouped_parameters = [
        {'params': [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)], 'weight_decay': args.weight_decay},
        {'params': [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], 'weight_decay': 0.0}
        ]
    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total)
    if args.fp16:
        try:
            from apex import amp
        except ImportError:
            raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.")
        model, optimizer = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level)

    # multi-gpu training (should be after apex fp16 initialization)
    if args.n_gpu > 1:
        model = torch.nn.DataParallel(model)

    # Distributed training (should be after apex fp16 initialization)
    if args.local_rank != -1:
        model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.local_rank],
                                                          output_device=args.local_rank,
                                                          find_unused_parameters=True)