Making large AI models cheaper, faster and more accessible
 
 
 
 
 
 
Go to file
Jie Zhu 73d36618a6
[profiler] add MemProfiler (#356)
* add memory trainer hook

* fix bug

* add memory trainer hook

* fix import bug

* fix import bug

* add trainer hook

* fix #370 git log bug

* modify `to_tensorboard` function to support better output

* remove useless output

* change the name of `MemProfiler`

* complete memory profiler

* replace error with warning

* finish trainer hook

* modify interface of MemProfiler

* modify `__init__.py` in profiler

* remove unnecessary pass statement

* add usage to doc string

* add usage to trainer hook

* new location to store temp data file
2022-03-29 12:48:34 +08:00
.github [devops] recover tsinghua pip source due to proxy issue (#509) 2022-03-24 16:11:49 +08:00
benchmark@9757a13749 Automated submodule synchronization 2022-03-14 00:01:12 +00:00
colossalai [profiler] add MemProfiler (#356) 2022-03-29 12:48:34 +08:00
docker update setup and workflow (#222) 2022-02-15 11:31:13 +08:00
docs [doc] update apidoc (#530) 2022-03-25 18:29:43 +08:00
examples@f4751fc792 Automated submodule synchronization (#483) 2022-03-22 09:34:26 +08:00
model_zoo [MOE] add MOEGPT model (#510) 2022-03-24 17:39:21 +08:00
requirements [log] better logging display with rich (#426) 2022-03-16 09:51:15 +08:00
tests [zero] polish ZeroInitContext (#540) 2022-03-29 09:09:04 +08:00
.flake8 added flake8 config (#219) 2022-02-15 11:31:13 +08:00
.gitignore moved env variables to global variables; (#215) 2022-02-15 11:31:13 +08:00
.gitmodules fixed submodule url (#167) 2022-01-19 22:33:39 +08:00
.pre-commit-config.yaml [zero] find miss code (#378) 2022-03-11 15:50:28 +08:00
.readthedocs.yaml update doc requirements and rtd conf (#165) 2022-01-19 19:46:43 +08:00
.style.yapf fixed mkdir conflict and align yapf config with flake (#220) 2022-02-15 11:31:13 +08:00
CHANGE_LOG.md fix typo in CHANGE_LOG.md 2022-03-13 23:34:34 +09:00
CONTRIBUTING.md update contributing.md with the current workflow (#440) 2022-03-17 10:28:04 +08:00
LICENSE polish license (#300) 2022-03-11 15:50:28 +08:00
MANIFEST.in refactor kernel (#142) 2022-01-13 16:47:17 +08:00
README-zh-Hans.md [profiler] add MemProfiler (#356) 2022-03-29 12:48:34 +08:00
README.md [profiler] add MemProfiler (#356) 2022-03-29 12:48:34 +08:00
pytest.ini Migrated project 2021-10-28 18:21:23 +02:00
setup.py [polish] add license meta to setup.py (#427) 2022-03-16 12:05:56 +08:00
version.txt update version (#533) 2022-03-26 12:34:28 +08:00

README.md

Colossal-AI

logo

An integrated large-scale model training system with efficient parallelization techniques.

Paper | Documentation | Examples | Forum | Blog

Build Documentation CodeFactor HuggingFace badge slack badge WeChat badge

| English | 中文 |

Table of Contents

Features

Colossal-AI provides a collection of parallel training components for you. We aim to support you to write your distributed deep learning models just like how you write your model on your laptop. We provide user-friendly tools to kickstart distributed training in a few lines.

  • Data Parallelism
  • Pipeline Parallelism
  • 1D, 2D, 2.5D, 3D tensor parallelism
  • Sequence parallelism
  • Friendly trainer and engine
  • Extensible for new parallelism
  • Mixed Precision Training
  • Zero Redundancy Optimizer (ZeRO)

(back to top)

Demo

ViT

  • 14x larger batch size, and 5x faster training for Tensor Parallelism = 64

GPT-3

  • Save 50% GPU resources, and 10.7% acceleration

GPT-2

  • 11x lower GPU memory consumption, and superlinear scaling efficiency with Tensor Parallelism
  • 10.7x larger model size on the same hardware

BERT

  • 2x faster training, or 50% longer sequence length

Please visit our documentation and tutorials for more details.

(back to top)

Installation

PyPI

pip install colossalai

This command will install CUDA extension if your have installed CUDA, NVCC and torch.

If you don't want to install CUDA extension, you should add --global-option="--no_cuda_ext", like:

pip install colossalai --global-option="--no_cuda_ext"

If you want to use ZeRO, you can run:

pip install colossalai[zero]

Install From Source

The version of Colossal-AI will be in line with the main branch of the repository. Feel free to create an issue if you encounter any problems. :-)

git clone https://github.com/hpcaitech/ColossalAI.git
cd ColossalAI
# install dependency
pip install -r requirements/requirements.txt

# install colossalai
pip install .

If you don't want to install and enable CUDA kernel fusion (compulsory installation when using fused optimizer):

pip install --global-option="--no_cuda_ext" .

(back to top)

Use Docker

Run the following command to build a docker image from Dockerfile provided.

cd ColossalAI
docker build -t colossalai ./docker

Run the following command to start the docker container in interactive mode.

docker run -ti --gpus all --rm --ipc=host colossalai bash

(back to top)

Community

Join the Colossal-AI community on Forum, Slack, and WeChat to share your suggestions, feedback, and questions with our engineering team.

Contributing

If you wish to contribute to this project, please follow the guideline in Contributing.

Thanks so much to all of our amazing contributors!

The order of contributor avatars is randomly shuffled.

(back to top)

Quick View

Start Distributed Training in Lines

import colossalai
from colossalai.utils import get_dataloader


# my_config can be path to config file or a dictionary obj
# 'localhost' is only for single node, you need to specify
# the node name if using multiple nodes
colossalai.launch(
    config=my_config,
    rank=rank,
    world_size=world_size,
    backend='nccl',
    port=29500,
    host='localhost'
)

# build your model
model = ...

# build you dataset, the dataloader will have distributed data
# sampler by default
train_dataset = ...
train_dataloader = get_dataloader(dataset=dataset,
                                shuffle=True
                                )


# build your optimizer
optimizer = ...

# build your loss function
criterion = ...

# initialize colossalai
engine, train_dataloader, _, _ = colossalai.initialize(
    model=model,
    optimizer=optimizer,
    criterion=criterion,
    train_dataloader=train_dataloader
)

# start training
engine.train()
for epoch in range(NUM_EPOCHS):
    for data, label in train_dataloader:
        engine.zero_grad()
        output = engine(data)
        loss = engine.criterion(output, label)
        engine.backward(loss)
        engine.step()

Write a Simple 2D Parallel Model

Let's say we have a huge MLP model and its very large hidden size makes it difficult to fit into a single GPU. We can then distribute the model weights across GPUs in a 2D mesh while you still write your model in a familiar way.

from colossalai.nn import Linear2D
import torch.nn as nn


class MLP_2D(nn.Module):

    def __init__(self):
        super().__init__()
        self.linear_1 = Linear2D(in_features=1024, out_features=16384)
        self.linear_2 = Linear2D(in_features=16384, out_features=1024)

    def forward(self, x):
        x = self.linear_1(x)
        x = self.linear_2(x)
        return x

(back to top)

Cite Us

@article{bian2021colossal,
  title={Colossal-AI: A Unified Deep Learning System For Large-Scale Parallel Training},
  author={Bian, Zhengda and Liu, Hongxin and Wang, Boxiang and Huang, Haichen and Li, Yongbin and Wang, Chuanrui and Cui, Fan and You, Yang},
  journal={arXiv preprint arXiv:2110.14883},
  year={2021}
}

(back to top)