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
Semantic search through a vectorized Wikipedia (SentenceBERT) with the Weaviate vector search engine

Semantic search through Wikipedia with the Weaviate vector search engine Weaviate is an open source vector search engine with build-in vectorization a

SeMI Technologies 191 Dec 26, 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
🗣️ NALP is a library that covers Natural Adversarial Language Processing.

NALP: Natural Adversarial Language Processing Welcome to NALP. Have you ever wanted to create natural text from raw sources? If yes, NALP is for you!

Gustavo Rosa 21 Aug 12, 2022
Rhyme with AI

Local development Create a conda virtual environment and activate it: conda env create --file environment.yml conda activate rhyme-with-ai Install the

GoDataDriven 28 Nov 21, 2022
Official PyTorch Implementation of paper "NeLF: Neural Light-transport Field for Single Portrait View Synthesis and Relighting", EGSR 2021.

NeLF: Neural Light-transport Field for Single Portrait View Synthesis and Relighting Official PyTorch Implementation of paper "NeLF: Neural Light-tran

Ken Lin 38 Dec 26, 2022
The source code of HeCo

HeCo This repo is for source code of KDD 2021 paper "Self-supervised Heterogeneous Graph Neural Network with Co-contrastive Learning". Paper Link: htt

Nian Liu 106 Dec 27, 2022
This repository details the steps in creating a Part of Speech tagger using Trigram Hidden Markov Models and the Viterbi Algorithm without using external libraries.

POS-Tagger This repository details the creation of a Part-of-Speech tagger using Trigram Hidden Markov Models to predict word tags in a word sequence.

Raihan Ahmed 1 Dec 09, 2021
Python module (C extension and plain python) implementing Aho-Corasick algorithm

pyahocorasick pyahocorasick is a fast and memory efficient library for exact or approximate multi-pattern string search meaning that you can find mult

Wojciech Muła 763 Dec 27, 2022
Official PyTorch code for ClipBERT, an efficient framework for end-to-end learning on image-text and video-text tasks

Official PyTorch code for ClipBERT, an efficient framework for end-to-end learning on image-text and video-text tasks. It takes raw videos/images + text as inputs, and outputs task predictions. ClipB

Jie Lei 雷杰 612 Jan 04, 2023
Open source annotation tool for machine learning practitioners.

doccano doccano is an open source text annotation tool for humans. It provides annotation features for text classification, sequence labeling and sequ

7.1k Jan 01, 2023
Count the frequency of letters or words in a text file and show a graph.

Word Counter By EBUS Coding Club Count the frequency of letters or words in a text file and show a graph. Requirements Python 3.9 or higher matplotlib

EBUS Coding Club 0 Apr 09, 2022
Conditional Transformer Language Model for Controllable Generation

CTRL - A Conditional Transformer Language Model for Controllable Generation Authors: Nitish Shirish Keskar, Bryan McCann, Lav Varshney, Caiming Xiong,

Salesforce 1.7k Dec 28, 2022
Natural Language Processing for Adverse Drug Reaction (ADR) Detection

Natural Language Processing for Adverse Drug Reaction (ADR) Detection This repo contains code from a project to identify ADRs in discharge summaries a

Medicines Optimisation Service - Austin Health 21 Aug 05, 2022
Fast topic modeling platform

The state-of-the-art platform for topic modeling. Full Documentation User Mailing List Download Releases User survey What is BigARTM? BigARTM is a pow

BigARTM 633 Dec 21, 2022
Python package for performing Entity and Text Matching using Deep Learning.

DeepMatcher DeepMatcher is a Python package for performing entity and text matching using deep learning. It provides built-in neural networks and util

461 Dec 28, 2022
Prompt tuning toolkit for GPT-2 and GPT-Neo

mkultra mkultra is a prompt tuning toolkit for GPT-2 and GPT-Neo. Prompt tuning injects a string of 20-100 special tokens into the context in order to

61 Jan 01, 2023
This repository is home to the Optimus data transformation plugins for various data processing needs.

Transformers Optimus's transformation plugins are implementations of Task and Hook interfaces that allows execution of arbitrary jobs in optimus. To i

Open Data Platform 37 Dec 14, 2022
PyTorch Implementation of VAENAR-TTS: Variational Auto-Encoder based Non-AutoRegressive Text-to-Speech Synthesis.

VAENAR-TTS - PyTorch Implementation PyTorch Implementation of VAENAR-TTS: Variational Auto-Encoder based Non-AutoRegressive Text-to-Speech Synthesis.

Keon Lee 67 Nov 14, 2022
SurvTRACE: Transformers for Survival Analysis with Competing Events

⭐ SurvTRACE: Transformers for Survival Analysis with Competing Events This repo provides the implementation of SurvTRACE for survival analysis. It is

Zifeng 13 Oct 06, 2022