rannet.pretrain

Pretrain RanNet

Module Contents

Functions

split_sentences(text[, max_length, lang])
cli()	rannet client
corpus(vocab_path, workers, min_length, max_length, ...)
single_corpus(vocab_path, max_length, chunk_size, ...)
pretrain(corpus_path, config_path, log_path, ...)
export_checkpoint(config_path, ckpt_path, target_path)
main()

Attributes

gpus

rannet.pretrain.gpus

rannet.pretrain.split_sentences(text: str, max_length: int = 512, lang: str = 'english')

rannet.pretrain.cli()

rannet client

rannet.pretrain.corpus(vocab_path: str, workers: int, min_length: int, max_length: int, chunk_size: int, corpus_dir: str, save_dir: str, whole_word_tokenizer: str, cased: bool)

rannet.pretrain.single_corpus(vocab_path: str, max_length: int, chunk_size: int, corpus_path: str, save_path: str)

rannet.pretrain.pretrain(corpus_path: str, config_path: str, log_path: str, base_ckpt_path: str, save_dir: str, record_info_path: str, batch_size: int, learning_rate: float, weight_decay: float, sequence_length: int, num_warmup_steps: int, num_train_steps: int, ckpt_save_freq: int, gradient_accumulation_steps: int, distributed: bool, distributed_strategy: str, verbose: int)

rannet.pretrain.export_checkpoint(config_path: str, ckpt_path: str, target_path: str)

rannet.pretrain.main()