Conditional random field. This module implements a conditional random field [LMP01]_. The forward computation of this class computes the log l

用于评价模型，计算每个标签的精确率，召回率，F1分数

Multi-class Focal loss implementation

A single training/test example for token classification.

A single training/test example for token classification.

custom progress bar Example: >>> pbar = ProgressBar(n_total=30,desc='Training') >>> step = 2 >>> pbar(step=step,info=

Implements Adam algorithm with weight decay fix. Parameters: lr (float): learning rate. Default 1e-3. betas (tuple of 2 floats):

A single set of features of data.

A single set of features of data.

Implements AdaBound algorithm. It has been proposed in `Adaptive Gradient Methods with Dynamic Bound of Learning Rate`_. Arguments: pa

# Code below is an implementation of https://arxiv.org/pdf/1804.04235.pdf # inspired but modified from https://github.com/DeadAt0m/adafactor-

computes and stores the average and current value Example: >>> loss = AverageMeter() >>> for step,batch in enumerate(train_da

Bert模型内定的学习率变化机制 Example: >>> scheduler = BertLR(optimizer) >>> for epoch in range(100): >>> scheduler.step()

Processor for the chinese ner data set.

Processor for the chinese ner data set.

Processor for the chinese ner data set.

Processor for the chinese ner data set.

Decays learning rate with cosine annealing, normalizes weight decay hyperparameter value, implements restarts. https://arxiv.org/abs/1711.0510

自定义学习率变化机制 Example: >>> scheduler = CustomDecayLR(optimizer) >>> for epoch in range(100): >>> scheduler.epoch

Cyclical learning rates for training neural networks Example: >>> scheduler = CyclicLR(optimizer) >>> for epoch in range(100)

Base class for data converters for sequence classification data sets.

Example # 初始化 fgm = FGM(model,epsilon=1,emb_name='word_embeddings.') for batch_input, batch_label in data: # 正常训练 los

r"""Implements Lamb algorithm. It has been proposed in `Large Batch Optimization for Deep Learning: Training BERT in 76 minutes`_. Arguments:

r"""Implements the LARS optimizer from https://arxiv.org/pdf/1708.03888.pdf Args: params (iterable): iterable of parameters to optimize o

PyTorch implementation of the lookahead wrapper. Lookahead Optimizer: https://arxiv.org/abs/1907.08610 We found that evaluation performa

模型保存，两种模式： 1. 直接保存最好模型 2. 按照epoch频率保存模型

Implements Nadam algorithm (a variant of Adam based on Nesterov momentum). It has been proposed in `Incorporating Nesterov Momentum into Adam`__.

主要参考论文<< Attention Is All You Need>>中的学习更新方式 Example: >>> scheduler = NoamLR(d_model,factor,warm_up,optimizer) >>> for epoch

Implements NovoGrad algorithm. Arguments: params (iterable): iterable of parameters to optimize or dicts defining parameter gr

Example pgd = PGD(model,emb_name='word_embeddings.',epsilon=1.0,alpha=0.3) K = 3 for batch_input, batch_label in data: # 正常训练

Implements the RAdam optimizer from https://arxiv.org/pdf/1908.03265.pdf Args: params (iterable): iterable of parameters to optimize or di

Implements the RAdam optimizer from https://arxiv.org/pdf/1908.03265.pdf with optional Layer-wise adaptive Scaling from https://arxiv.org/pdf/1708

RAdam + LARS Example: >>> model = ResNet() >>> optimizer = Ralamb(model.parameters(), lr=0.001)

Reduce learning rate when a metric has stopped improving. Models often benefit from reducing the learning rate by a factor of 2-10 once learni

Reduce learning rate and weight decay when a metric has stopped improving. Models often benefit from reducing the learning rate by a factor of

r"""Implements stochastic gradient descent (optionally with momentum) with weight decay from the paper `Fixing Weight Decay Regularization in Adam