github
/
ColossalAI
mirror of https://github.com/hpcaitech/ColossalAI
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
ColossalAI/examples/tutorial/fastfold
History
LuGY ecbad93b65
[example] Add fastfold tutorial (#2528) 
* add fastfold example

* pre-commit polish

* pre-commit polish readme and add empty test ci

* Add test_ci and reduce the default sequence length
 		2023-01-30 17:08:18 +08:00
..
README.md [example] Add fastfold tutorial (#2528) 2023-01-30 17:08:18 +08:00
inference.py [example] Add fastfold tutorial (#2528) 2023-01-30 17:08:18 +08:00
test_ci.sh [example] Add fastfold tutorial (#2528) 2023-01-30 17:08:18 +08:00

README.md

FastFold Inference

Table of contents

📚 Overview

This example lets you to quickly try out the inference of FastFold.

NOTE: We use random data and random parameters in this example.

🚀 Quick Start

  1. Install FastFold

We highly recommend installing an Anaconda or Miniconda environment and install PyTorch with conda.

git clone https://github.com/hpcaitech/FastFold
cd FastFold
conda env create --name=fastfold -f environment.yml
conda activate fastfold
python setup.py install
  1. Run the inference scripts.
python inference.py --gpus=1 --n_res=256 --chunk_size=None --inplace
  • gpus means the DAP size
  • n_res means the length of residue sequence
  • chunk_size introduces a memory-saving technology at the cost of speed, None means not using, 16 may be a good trade off for long sequences.
  • inplace introduces another memory-saving technology with zero cost, drop --inplace if you do not want it.

🔍 Dive into FastFold

There are another features of FastFold, such as:

  • more excellent kernel based on triton
  • much faster data processing based on ray
  • training supported

More detailed information can be seen here.