Model parallel transformers in JAX and Haiku

Overview

Table of contents

  1. Mesh Transformer JAX
    1. Updates
  2. Pretrained Models
    1. GPT-J-6B
      1. Links
      2. Acknowledgments
      3. License
      4. Model Details
      5. Zero-Shot Evaluations
  3. Architecture and Usage
    1. Fine-tuning
    2. JAX Dependency
  4. TODO

Mesh Transformer JAX

A haiku library using the xmap/pjit operators in JAX for model parallelism of transformers.

The parallelism scheme is similar to the original Megatron-LM, which is efficient on TPUs due to the high speed 2d mesh network. There is also an experimental model version which implements ZeRo style sharding.

This library is designed for scalability up to approximately 40B parameters on TPUv3s, beyond which different parallelism strategies should be used. See other implementations such as GPT-NeoX or DeepSpeed for that.

One future direction for research is integrating this codebase with swarm-jax, to achieve further scalability with pipeline parallelism.

Updates

12-07-21: Added guide to fine tuning

Pretrained Models

GPT-J-6B

A 6 billion parameter, autoregressive text generation model trained on The Pile.

Links

Slim weights (bf16 weights only, for inference, 9GB)

Full weights (including optimizer params, 61GB)

Colab demo

Web demo

Aran's blog post

Acknowledgments

This project would not have been possible without compute generously provided by the TPU Research Cloud with assistance from EleutherAI.

Thanks to the Cloud TPU team at Google for providing early access to the Cloud TPU VM alpha (now publicly available!)

Thanks to everyone who have helped out one way or another (listed alphabetically):

  • Aran Komatsuzaki for advice with experiment design and writing the blog posts.
  • James Bradbury for valuable assistance with debugging JAX issues.
  • Janko Prester for creating the web demo frontend.
  • Laurence Golding for adding some features to the web demo.
  • Leo Gao for running zero shot evaluations for the baseline models for the table.

License

The weights of GPT-J-6B are licensed under version 2.0 of the Apache License.

Model Details

Hyperparameter Value
n_parameters 6,053,381,344
n_layers 28*
d_model 4,096
d_ff 16,384
n_heads 16
d_head 256
n_ctx 2,048
n_vocab 50,257 (same tokenizer as GPT-2/3)
position encoding Rotary position encodings (RoPE)
RoPE dimensions 64

* each layer consists of one feedforward block and one self attention block

The model consists of 28 layers with a model dimension of 4096, and a feedforward dimension of 16384. The model dimension is split into 16 heads, each with a dimension of 256. Rotary position encodings (RoPE) was applied to 64 dimensions of each head. The model is trained with a tokenization vocabulary of 50257, using the same set of BPEs as GPT-2/GPT-3.

Zero-Shot Evaluations

Models roughly sorted by performance, or by FLOPs if not available.

Model Weights Training FLOPs LAMBADA PPL ↓ LAMBADA Acc ↑ Winogrande ↑ Hellaswag ↑ PIQA ↑ Dataset Size (GB)
Chance 0 ~a lot ~0% 50% 25% 25% 0
GPT-3-Ada‡ ----- 9.95 51.6% 52.9% 43.4% 70.5% -----
GPT-2-1.5B ----- 10.63 51.21% 59.4% 50.9% 70.8% 40
GPTNeo-1.3B‡ 3.0e21 7.50 57.2% 55.0% 48.9% 71.1% 825
Megatron-2.5B* 2.4e21 ----- 61.7% ----- ----- ----- 174
GPTNeo-2.7B‡ 6.8e21 5.63 62.2% 56.5% 55.8% 73.0% 825
GPT-3-1.3B*‡ 2.4e21 5.44 63.6% 58.7% 54.7% 75.1% ~800
GPT-3-Babbage‡ ----- 5.58 62.4% 59.0% 54.5% 75.5% -----
Megatron-8.3B* 7.8e21 ----- 66.5% ----- ----- ----- 174
GPT-3-2.7B*‡ 4.8e21 4.60 67.1% 62.3% 62.8% 75.6% ~800
Megatron-11B† 1.0e22 ----- ----- ----- ----- ----- 161
GPT-J-6B 1.5e22 3.99 69.7% 65.3% 66.1% 76.5% 825
GPT-3-6.7B*‡ 1.2e22 4.00 70.3% 64.5% 67.4% 78.0% ~800
GPT-3-Curie‡ ----- 4.00 69.3% 65.6% 68.5% 77.9% -----
GPT-3-13B*‡ 2.3e22 3.56 72.5% 67.9% 70.9% 78.5% ~800
GPT-3-175B*‡ 3.1e23 3.00 76.2% 70.2% 78.9% 81.0% ~800
GPT-3-Davinci‡ ----- 3.0 75% 72% 78% 80% -----
Gopher 230B* 6.31E+23 ----- 74.50% 70.10% 79.20% 81.80% 1344
MT-NLG 530B*‡ ----- ----- 76.6% 73.0% 80.2% 82.0% -----

* represents evaluation numbers reported by their respective authors, all other numbers are provided by running the lm-evaluation-harness either with the released weights or with API access. Due to subtle implementation differences as well as different zero shot task framing, these might not be directly comparable. See this blog post for more details.

The Megatron-11B model provides no comparable metrics, and several implementations using the released weights do not reproduce the generation quality and evaluations. (see 1 2 3) Thus, evaluation was not attempted.

These models have been trained with data which contains possible test set contamination. The OpenAI GPT-3 models failed to deduplicate training data for certain test sets, while the GPT-Neo models as well as this one is trained on The Pile, which has not been deduplicated against any test sets.

Architecture and Usage

Most scripts in this repository are designed to be run on TPUs, which under the TPU-VM architecture are virtual machines which can run arbitrary code. Most scripts are designed to spin up a TPU, SSH into it to set up the dependencies and copy code over from the local directory, and then start a Ray worker which can accept RPC calls.

The TPUVMs handles running model training steps and evaluation, checkpoint save and loading, while the driver python program handles data loading and general orchestration (such as when to save checkpoints etc).

This means that most scripts (train.py, eval_harness.py etc) expect to be running on a GCE virtual machine in the same region as the TPUs, to minimize RPC latency and data transfer cost. Other scripts (usually ones which don't take a --tpu argument, such as device_sample.py, device_serve.py or device_train.py) expect to be run directly on a TPUVM. The device_* scripts only work on a v3-8 and not on larger pods.

Furthermore, there is an example (resharding_example.py) of how to convert the provided checkpoints (which have 8 shards in the case of GPT-J-6B) down to a smaller number, such as for when running on GPU(s).

Fine-tuning

To fine-tune the model, run device_train.py on a TPU VM. Using a TPU v3-8, you can fine-tune at a rate of ~5000 tokens/second, which should be sufficient for small-to-medium-size datasets.

Please read the step by step guide for thorough fine-tuning instructions.

JAX Dependency

Note this library has some specific requirements for JAX version. Specifically, to use the v1 models (including GPT-J 6B), jax==0.2.12 is required. This in turn depends on jaxlib==0.1.68. If this is not done, you will get cryptic xmap errors

However, to use the v2 model code (no publicly released weights), the newest JAX version can be used.

Citation

To cite this repository:

@misc{mesh-transformer-jax,
  author = {Wang, Ben},
  title = {{Mesh-Transformer-JAX: Model-Parallel Implementation of Transformer Language Model with JAX}},
  howpublished = {\url{https://github.com/kingoflolz/mesh-transformer-jax}},
  year = 2021,
  month = May
}

To cite the weights of GPT-J-6B:

@misc{gpt-j,
  author = {Wang, Ben and Komatsuzaki, Aran},
  title = {{GPT-J-6B: A 6 Billion Parameter Autoregressive Language Model}},
  howpublished = {\url{https://github.com/kingoflolz/mesh-transformer-jax}},
  year = 2021,
  month = May
}

If you use this repository or any of the pretrained weights to do something cool, we would love to hear about it. Feel free to open a github issue or reach out over email (in profile).

TODO

  • disentangle heads and shards
  • test/benchmark on TPU
  • implement gradient checkpointing
  • fix initialization
  • mixed precision
  • deal with preemptible TPUs
  • test and validate generation
  • shard activations instead of replicating for memory efficiency (in v2)
  • support ZeRO style sharding (in v2)
Owner
Ben Wang
Ben Wang
Simple Annotated implementation of GPT-NeoX in PyTorch

Simple Annotated implementation of GPT-NeoX in PyTorch This is a simpler implementation of GPT-NeoX in PyTorch. We have taken out several optimization

labml.ai 101 Dec 03, 2022
A unified tokenization tool for Images, Chinese and English.

ICE Tokenizer Token id [0, 20000) are image tokens. Token id [20000, 20100) are common tokens, mainly punctuations. E.g., icetk[20000] == 'unk', ice

THUDM 42 Dec 27, 2022
Samantha, A covid-19 information bot which will provide basic information about this pandemic in form of conversation.

Covid-19-BOT Samantha, A covid-19 information bot which will provide basic information about this pandemic in form of conversation. This bot uses torc

Neeraj Majhi 2 Nov 05, 2021
Shared code for training sentence embeddings with Flax / JAX

flax-sentence-embeddings This repository will be used to share code for the Flax / JAX community event to train sentence embeddings on 1B+ training pa

Nils Reimers 23 Dec 30, 2022
A framework for implementing federated learning

This is partly the reproduction of the paper of [Privacy-Preserving Federated Learning in Fog Computing](DOI: 10.1109/JIOT.2020.2987958. 2020)

DavidChen 46 Sep 23, 2022
Datasets of Automatic Keyphrase Extraction

This repository contains 20 annotated datasets of Automatic Keyphrase Extraction made available by the research community. Following are the datasets and the original papers that proposed them. If yo

LIAAD - Laboratory of Artificial Intelligence and Decision Support 163 Dec 23, 2022
A library for finding knowledge neurons in pretrained transformer models.

knowledge-neurons An open source repository replicating the 2021 paper Knowledge Neurons in Pretrained Transformers by Dai et al., and extending the t

EleutherAI 96 Dec 21, 2022
基于GRU网络的句子判断程序/A program based on GRU network for judging sentences

SentencesJudger SentencesJudger 是一个基于GRU神经网络的句子判断程序,基本的功能是判断文章中的某一句话是否为一个优美的句子。 English 如何使用SentencesJudger 确认Python运行环境 安装pyTorch与LTP python3 -m pip

8 Mar 24, 2022
Mysticbbs-rjam - rJAM splitscreen message reader for MysticBBS A46+

rJAM splitscreen message reader for MysticBBS A46+

Robbert Langezaal 4 Nov 22, 2022
Revisiting Pre-trained Models for Chinese Natural Language Processing (Findings of EMNLP 2020)

This repository contains the resources in our paper "Revisiting Pre-trained Models for Chinese Natural Language Processing", which will be published i

Yiming Cui 463 Dec 30, 2022
Repository for the paper "Optimal Subarchitecture Extraction for BERT"

Bort Companion code for the paper "Optimal Subarchitecture Extraction for BERT." Bort is an optimal subset of architectural parameters for the BERT ar

Alexa 461 Nov 21, 2022
:hot_pepper: R²SQL: "Dynamic Hybrid Relation Network for Cross-Domain Context-Dependent Semantic Parsing." (AAAI 2021)

R²SQL The PyTorch implementation of paper Dynamic Hybrid Relation Network for Cross-Domain Context-Dependent Semantic Parsing. (AAAI 2021) Requirement

huybery 60 Dec 31, 2022
Spam filtering made easy for you

spammy Author: Tasdik Rahman Latest version: 1.0.3 Contents 1 Overview 2 Features 3 Example 3.1 Accuracy of the classifier 4 Installation 4.1 Upgradin

Tasdik Rahman 137 Dec 18, 2022
GooAQ 🥑 : Google Answers to Google Questions!

This repository contains the code/data accompanying our recent work on long-form question answering.

AI2 112 Nov 06, 2022
Local cross-platform machine translation GUI, based on CTranslate2

DesktopTranslator Local cross-platform machine translation GUI, based on CTranslate2 Download Windows Installer You can either download a ready-made W

Yasmin Moslem 29 Jan 05, 2023
A PyTorch-based model pruning toolkit for pre-trained language models

English | 中文说明 TextPruner是一个为预训练语言模型设计的模型裁剪工具包,通过轻量、快速的裁剪方法对模型进行结构化剪枝,从而实现压缩模型体积、提升模型速度。 其他相关资源: 知识蒸馏工具TextBrewer:https://github.com/airaria/TextBrewe

Ziqing Yang 231 Jan 08, 2023
Multilingual finetuning of Machine Translation model on low-resource languages. Project for Deep Natural Language Processing course.

Low-resource-Machine-Translation This repository contains the code for the project relative to the course Deep Natural Language Processing. The goal o

Andrea Cavallo 3 Jun 22, 2022
NLP codes implemented with Pytorch (w/o library such as huggingface)

NLP_scratch NLP codes implemented with Pytorch (w/o library such as huggingface) scripts ├── models: Neural Network models ├── data: codes for dataloa

3 Dec 28, 2021
🏆 • 5050 most frequent words in 109 languages

🏆 Most Common Words Multilingual 5000 most frequent words in 109 languages. Uses wordfrequency.info as a source. 🔗 License source code license data

14 Nov 24, 2022
code for "AttentiveNAS Improving Neural Architecture Search via Attentive Sampling"

AttentiveNAS: Improving Neural Architecture Search via Attentive Sampling This repository contains PyTorch evaluation code, training code and pretrain

Facebook Research 94 Oct 26, 2022