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
PocketSphinx is a lightweight speech recognition engine, specifically tuned for handheld and mobile devices, though it works equally well on the desktop

molten A minimal, extensible, fast and productive API framework for Python 3. Changelog: https://moltenframework.com/changelog.html Community: https:/

3.2k Dec 28, 2022
Statistics and Mathematics for Machine Learning, Deep Learning , Deep NLP

Stat4ML Statistics and Mathematics for Machine Learning, Deep Learning , Deep NLP This is the first course from our trio courses: Statistics Foundatio

Omid Safarzadeh 83 Dec 29, 2022
A PyTorch Implementation of End-to-End Models for Speech-to-Text

speech Speech is an open-source package to build end-to-end models for automatic speech recognition. Sequence-to-sequence models with attention, Conne

Awni Hannun 647 Dec 25, 2022
Language-Agnostic SEntence Representations

LASER Language-Agnostic SEntence Representations LASER is a library to calculate and use multilingual sentence embeddings. NEWS 2019/11/08 CCMatrix is

Facebook Research 3.2k Jan 04, 2023
Extract Keywords from sentence or Replace keywords in sentences.

FlashText This module can be used to replace keywords in sentences or extract keywords from sentences. It is based on the FlashText algorithm. Install

Vikash Singh 5.3k Jan 01, 2023
fastai ulmfit - Pretraining the Language Model, Fine-Tuning and training a Classifier

fast.ai ULMFiT with SentencePiece from pretraining to deployment Motivation: Why even bother with a non-BERT / Transformer language model? Short answe

Florian Leuerer 26 May 27, 2022
Voice Assistant inspired by Google Assistant, Cortana, Alexa, Siri, ...

author: @shival_gupta VoiceAI This program is an example of a simple virtual assitant It will listen to you and do accordingly It will begin with wish

Shival Gupta 1 Jan 06, 2022
An IVR Chatbot which can exponentially reduce the burden of companies as well as can improve the consumer/end user experience.

IVR-Chatbot Achievements 🏆 Team Uhtred won the Maverick 2.0 Bot-a-thon 2021 organized by AbInbev India. ❓ Problem Statement As we all know that, lot

ARYAMAAN PANDEY 9 Dec 08, 2022
LeBenchmark: a reproducible framework for assessing SSL from speech

LeBenchmark: a reproducible framework for assessing SSL from speech

11 Nov 30, 2022
Implementation of COCO-LM, Correcting and Contrasting Text Sequences for Language Model Pretraining, in Pytorch

COCO LM Pretraining (wip) Implementation of COCO-LM, Correcting and Contrasting Text Sequences for Language Model Pretraining, in Pytorch. They were a

Phil Wang 44 Jul 28, 2022
Fastseq 基于ONNXRUNTIME的文本生成加速框架

Fastseq 基于ONNXRUNTIME的文本生成加速框架

Jun Gao 9 Nov 09, 2021
Understand Text Summarization and create your own summarizer in python

Automatic summarization is the process of shortening a text document with software, in order to create a summary with the major points of the original document. Technologies that can make a coherent

Sreekanth M 1 Oct 18, 2022
Word2Wave: a framework for generating short audio samples from a text prompt using WaveGAN and COALA.

Word2Wave is a simple method for text-controlled GAN audio generation. You can either follow the setup instructions below and use the source code and CLI provided in this repo or you can have a play

Ilaria Manco 91 Dec 23, 2022
Translate U is capable of translating the text present in an image from one language to the other.

Translate U is capable of translating the text present in an image from one language to the other. The app uses OCR and Google translate to identify and translate across 80+ languages.

Neelanjan Manna 1 Dec 22, 2021
Traditional Chinese Text Recognition Dataset: Synthetic Dataset and Labeled Data

Traditional Chinese Text Recognition Dataset: Synthetic Dataset and Labeled Data Authors: Yi-Chang Chen, Yu-Chuan Chang, Yen-Cheng Chang and Yi-Ren Ye

Yi-Chang Chen 5 Dec 15, 2022
Repository for Project Insight: NLP as a Service

Project Insight NLP as a Service Contents Introduction Features Installation Setup and Documentation Project Details Demonstration Directory Details H

Abhishek Kumar Mishra 286 Dec 06, 2022
Tools, wrappers, etc... for data science with a concentration on text processing

Rosetta Tools for data science with a focus on text processing. Focuses on "medium data", i.e. data too big to fit into memory but too small to necess

207 Nov 22, 2022
Words_And_Phrases - Just a repo for useful words and phrases that might come handy in some scenarios. Feel free to add yours

Words_And_Phrases Just a repo for useful words and phrases that might come handy in some scenarios. Feel free to add yours Abbreviations Abbreviation

Subhadeep Mandal 1 Feb 01, 2022
Unet-TTS: Improving Unseen Speaker and Style Transfer in One-shot Voice Cloning

Unet-TTS: Improving Unseen Speaker and Style Transfer in One-shot Voice Cloning English | 中文 ❗ Now we provide inferencing code and pre-training models

164 Jan 02, 2023
Partially offline multi-language translator built upon Huggingface transformers.

Translate Command-line interface to translation pipelines, powered by Huggingface transformers. This tool can download translation models, and then us

Richard Jarry 8 Oct 25, 2022