ByT5: Towards a token-free future with pre-trained byte-to-byte models

Related tags

Text Data & NLP byt5
Overview

ByT5: Towards a token-free future with pre-trained byte-to-byte models

ByT5 is a tokenizer-free extension of the mT5 model. Instead of using a subword vocabulary like most other pretrained language models (BERT, XLM-R, T5, GPT-3), our ByT5 model operates directly on UTF-8 bytes, removing the need for any text preprocessing. Beyond the reduction in system complexity, we find that parameter-matched ByT5 models are competitive with mT5 across a range of tasks, and outperform mT5 on tasks that involve noisy text or are sensitive to spelling and pronunciation. This repo can be used to reproduce the experiments in the ByT5 paper.

Usage

Training

To run this code, you need to install the t5 library. General instructions for training, fine-tuning, evaluation, and exporting models for inference can be found in the t5 repo. In order to use the additional ByT5 tasks provided in this library with the t5_mesh_transformer command, run from this directory and add the flag --module_import="byt5.tasks".

To train a ByT5-Large model on the mc4 task from scratch as described in the paper:

export PROJECT=yourproject
export ZONE=yourzone
export BUCKET=yourbucket
export TPU=yourtpu

ctpu up --name=$TPU --project=$PROJECT --zone=$ZONE --tpu-size=v3-256 --tpu-only --noconf

TASK=byt5_mc4
MODEL_DIR="${BUCKET}${TASK}"

python -m t5.models.mesh_transformer_main \
  --tpu="${TPU}" \
  --gcp_project="${PROJECT}" \
  --tpu_zone="${ZONE}" \
  --model_dir="${MODEL_DIR}" \
  --gin_file="models/byt5.large.gin" \
  --gin_param="MIXTURE_NAME = '${TASK}'" \
  --gin_param="utils.run.sequence_length = {'inputs': 1024, 'targets': 189}" \
  --gin_param="utils.run.batch_size = ('tokens_per_batch', 1048576)" \
  --gin_param="utils.run.learning_rate_schedule=@learning_rate_schedules.rsqrt_no_ramp_down" \
  --gin_param="run.train_steps = 1000000" \
  --gin_param="utils.tpu_mesh_shape.model_parallelism = 1" \
  --gin_param="utils.tpu_mesh_shape.tpu_topology = 'v3-256'" \
  --eval_mode="perplexity_eval" \
  --eval_gin_param="mesh_eval_dataset_fn.num_eval_examples = 10000" \
  --t5_tfds_data_dir="${BUCKET}/t5-tfds" \
  --module_import="byt5.tasks"

Fine-Tuning

The example below shows how to finetune the ByT5-Large model on the XNLI zeroshot task.

export PROJECT=yourproject
export ZONE=yourzone
export BUCKET=yourbucket
export TPU=yourtpu

ctpu up --name=$TPU --project=$PROJECT --zone=$ZONE --tpu-size=v3-256 --tpu-only --noconf

TASK=byt5_xnli_zeroshot
PRETRAINED_DIR=gs://t5-data/pretrained_models/byt5/large
PRETRAINED_STEPS=1000000
FINETUNE_STEPS=262144
MODEL_DIR="${BUCKET}${TASK}"

# Run fine-tuning
python -m t5.models.mesh_transformer_main \
  --tpu="${TPU}" \
  --gcp_project="${PROJECT}" \
  --tpu_zone="${ZONE}" \
  --model_dir="${MODEL_DIR}" \
  --gin_file="${PRETRAINED_DIR}/operative_config.gin" \
  --gin_param="utils.tpu_mesh_shape.tpu_topology = 'v3-256'" \
  --gin_param="MIXTURE_NAME = '${TASK}'" \
  --gin_param="utils.run.train_steps=$((PRETRAINED_STEPS+FINETUNE_STEPS))" \
  --gin_param="utils.run.init_checkpoint='${PRETRAINED_DIR}/model.ckpt-${PRETRAINED_STEPS}'" \
  --t5_tfds_data_dir="${BUCKET}/t5-tfds" \
  --module_import="byt5.tasks"
  --gin_param="utils.run.batch_size = ('tokens_per_batch', 1048576)" \
  --gin_param="utils.run.sequence_length = {'inputs': 2048, 'targets': 56}"
  --eval_gin_param="Bitransformer.decode.max_decode_length = 56" \

The remaining experiments are shown in the tasks.py file.

Released Model Checkpoints

We have released the following checkpoints for pre-trained models described in our paper:

How to Cite

If you extend or use this work, please cite the paper where it was introduced:

@misc{xue2021byt5,
    title={ByT5: Towards a token-free future with pre-trained byte-to-byte models},
    author={Linting Xue and Aditya Barua and Noah Constant and Rami Al-Rfou and Sharan Narang and Mihir Kale and Adam Roberts and Colin Raffel},
    year={2021},
    eprint={2105.13626},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

This is not an officially supported Google product.

You might also like...
Beyond Masking: Demystifying Token-Based Pre-Training for Vision Transformers
Beyond Masking: Demystifying Token-Based Pre-Training for Vision Transformers

beyond masking Beyond Masking: Demystifying Token-Based Pre-Training for Vision Transformers The code is coming Figure 1: Pipeline of token-based pre-

Yunjie Tian 23 Sep 27, 2022
TunBERT is the first release of a pre-trained BERT model for the Tunisian dialect using a Tunisian Common-Crawl-based dataset.
TunBERT is the first release of a pre-trained BERT model for the Tunisian dialect using a Tunisian Common-Crawl-based dataset.

TunBERT is the first release of a pre-trained BERT model for the Tunisian dialect using a Tunisian Common-Crawl-based dataset. TunBERT was applied to three NLP downstream tasks: Sentiment Analysis (SA), Tunisian Dialect Identification (TDI) and Reading Comprehension Question-Answering (RCQA)

InstaDeep Ltd 72 Dec 9, 2022
DziriBERT: a Pre-trained Language Model for the Algerian Dialect
DziriBERT: a Pre-trained Language Model for the Algerian Dialect

DziriBERT is the first Transformer-based Language Model that has been pre-trained specifically for the Algerian Dialect.

null 117 Jan 7, 2023
Google and Stanford University released a new pre-trained model called ELECTRA
Google and Stanford University released a new pre-trained model called ELECTRA

Google and Stanford University released a new pre-trained model called ELECTRA, which has a much compact model size and relatively competitive performance compared to BERT and its variants. For further accelerating the research of the Chinese pre-trained model, the Joint Laboratory of HIT and iFLYTEK Research (HFL) has released the Chinese ELECTRA models based on the official code of ELECTRA. ELECTRA-small could reach similar or even higher scores on several NLP tasks with only 1/10 parameters compared to BERT and its variants.

Yiming Cui 1.2k Dec 30, 2022
Code for CodeT5: a new code-aware pre-trained encoder-decoder model.
Code for CodeT5: a new code-aware pre-trained encoder-decoder model.

CodeT5: Identifier-aware Unified Pre-trained Encoder-Decoder Models for Code Understanding and Generation This is the official PyTorch implementation

Salesforce 564 Jan 8, 2023
Use PaddlePaddle to reproduce the paper：mT5: A Massively Multilingual Pre-trained Text-to-Text Transformer
Use PaddlePaddle to reproduce the paper：mT5: A Massively Multilingual Pre-trained Text-to-Text Transformer

MT5_paddle Use PaddlePaddle to reproduce the paper：mT5: A Massively Multilingual Pre-trained Text-to-Text Transformer English | 简体中文 mT5: A Massively

null 2 Oct 17, 2021
Implementation of Natural Language Code Search in the project CodeBERT: A Pre-Trained Model for Programming and Natural Languages.

CodeBERT-Implementation In this repo we have replicated the paper CodeBERT: A Pre-Trained Model for Programming and Natural Languages. We are interest

Tanuj Sur 4 Jul 1, 2022
ElasticBERT: A pre-trained model with multi-exit transformer architecture.

This repository contains finetuning code and checkpoints for ElasticBERT. Towards Efficient NLP: A Standard Evaluation and A Strong Baseli

fastNLP 48 Dec 14, 2022
KakaoBrain KoGPT (Korean Generative Pre-trained Transformer)

KoGPT KoGPT (Korean Generative Pre-trained Transformer) https://github.com/kakaobrain/kogpt https://huggingface.co/kakaobrain/kogpt Model Descriptions

Kakao Brain 797 Dec 26, 2022
Comments
  • Using ByT5 for NER

    Using ByT5 for NER

    I wanted to use ByT5 to finetune the pretrained model on a NER task. I can see that a sample NER task has been added to the SeqIO library in tasks.py. But can you explain it to me in detail about how I can go about it, because I am still unsure about the format in which we have to feed in the train data for NER. I'm new to SeqIO and I am still trying to figure out how to use the scripts mentioned in the readme file.

    opened by Akshay0799 0
  • Issue with span_corruption preprocessors

    Issue with span_corruption preprocessors

    Hi, I'm trying to pretrain byt5 on the custom corpora (of short texts), but I'm stuck with the data pipeline (the code is below). When I decode the outputs, inputs and targets are merged from the different examples, and both are noised.

    DEFAULT_OUTPUT_FEATURES = {
    "inputs": seqio.Feature(vocabulary=seqio.ByteVocabulary(), add_eos=True),
    "targets": seqio.Feature(vocabulary=seqio.ByteVocabulary(), add_eos=True),
}

MEAN_NOISE_SPAN_LENGTH = 5
SEQUENCE_LENGTH = sequence_length={"inputs": 128, "targets": 128}


seqio.TaskRegistry.add(
    name="nelma_byt5",
    source=seqio.TextLineDataSource(split_to_filepattern={
            "train": "/disk1/projekti/mondodb_lm/test.tsv",
        }),
    preprocessors=[
        functools.partial(
          t5.data.preprocessors.parse_tsv,
          field_names=['text','class'],
          field_delim='\t',
        ),
        functools.partial(
              seqio.preprocessors.rekey,
              key_map={"inputs": None, "targets": "text"}
        ),
        seqio.preprocessors.tokenize,
        seqio.CacheDatasetPlaceholder(),
        functools.partial(
          t5.data.preprocessors.span_corruption,
          mean_noise_span_length=MEAN_NOISE_SPAN_LENGTH),
        seqio.preprocessors.append_eos_after_trim,
     ],
      output_features=DEFAULT_OUTPUT_FEATURES,
      metric_fns=[])
    opened by mondonomo 0
  • How to convert ByT5 model to ONNX format?

    How to convert ByT5 model to ONNX format?

    Hi,

    ONNX allows to compress transformers models and speed up the inference time on CPU and GPU.

    Who could share code / notebook to convert mT5 and ByT5 models to ONNX format?

    There is the library fastT5 for T5 conversion (great!) but it has not been updated to the latest version of transformers and therefore, it does not accept mT5 and ByT5 models until today.

    Thanks.

    opened by piegu 0
Owner
Google Research
Google Research
GitHub
Silero Models: pre-trained speech-to-text, text-to-speech models and benchmarks made embarrassingly simple

Silero Models: pre-trained speech-to-text, text-to-speech models and benchmarks made embarrassingly simple

Alexander Veysov 3.2k Dec 31, 2022
Code associated with the "Data Augmentation using Pre-trained Transformer Models" paper

Data Augmentation using Pre-trained Transformer Models Code associated with the Data Augmentation using Pre-trained Transformer Models paper Code cont

null 44 Dec 31, 2022
Must-read papers on improving efficiency for pre-trained language models.

Must-read papers on improving efficiency for pre-trained language models.

Tobias Lee 89 Jan 3, 2023
The repository for the paper: Multilingual Translation via Grafting Pre-trained Language Models

Graformer The repository for the paper: Multilingual Translation via Grafting Pre-trained Language Models Graformer (also named BridgeTransformer in t

null 22 Dec 14, 2022
Prompt-learning is the latest paradigm to adapt pre-trained language models (PLMs) to downstream NLP tasks

Prompt-learning is the latest paradigm to adapt pre-trained language models (PLMs) to downstream NLP tasks, which modifies the input text with a textual template and directly uses PLMs to conduct pre-trained tasks. This library provides a standard, flexible and extensible framework to deploy the prompt-learning pipeline. OpenPrompt supports loading PLMs directly from huggingface transformers. In the future, we will also support PLMs implemented by other libraries.

THUNLP 2.3k Jan 8, 2023
Chinese Pre-Trained Language Models (CPM-LM) Version-I

CPM-Generate 为了促进中文自然语言处理研究的发展，本项目提供了 CPM-LM (2.6B) 模型的文本生成代码，可用于文本生成的本地测试，并以此为基础进一步研究零次学习/少次学习等场景。[项目首页] [模型下载] [技术报告] 若您想使用CPM-1进行推理，我们建议使用高效推理工具BMI

Tsinghua AI 1.4k Jan 3, 2023
TweebankNLP - Pre-trained Tweet NLP Pipeline (NER, tokenization, lemmatization, POS tagging, dependency parsing) + Models + Tweebank-NER

TweebankNLP This repo contains the new Tweebank-NER dataset and Twitter-Stanza p

Laboratory for Social Machines 84 Dec 20, 2022
PyTorch Implementation of "Bridging Pre-trained Language Models and Hand-crafted Features for Unsupervised POS Tagging" (Findings of ACL 2022)

Feature_CRF_AE Feature_CRF_AE provides a implementation of Bridging Pre-trained Language Models and Hand-crafted Features for Unsupervised POS Tagging

Jacob Zhou 6 Apr 29, 2022
Guide to using pre-trained large language models of source code

Large Models of Source Code I occasionally train and publicly release large neural language models on programs, including PolyCoder. Here, I describe

Vincent Hellendoorn 947 Dec 28, 2022
TaCL: Improve BERT Pre-training with Token-aware Contrastive Learning

TaCL: Improve BERT Pre-training with Token-aware Contrastive Learning

Yixuan Su 26 Oct 17, 2022