Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/thu-ml/vflow
Official code for "VFlow: More Expressive Generative Flows with Variational Data Augmentation" (ICML 2020)
https://github.com/thu-ml/vflow
Last synced: 3 months ago
JSON representation
Official code for "VFlow: More Expressive Generative Flows with Variational Data Augmentation" (ICML 2020)
- Host: GitHub
- URL: https://github.com/thu-ml/vflow
- Owner: thu-ml
- License: mit
- Created: 2020-06-29T11:57:49.000Z (over 4 years ago)
- Default Branch: master
- Last Pushed: 2023-04-25T16:59:25.000Z (over 1 year ago)
- Last Synced: 2024-07-04T01:02:04.763Z (4 months ago)
- Language: Python
- Homepage:
- Size: 30.3 KB
- Stars: 38
- Watchers: 10
- Forks: 4
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# VFlow: More Expressive Generative Flows with Variational Data Augmentation
This repository contains Tensorflow implementation of experiments from the paper [VFlow: More Expressive Generative Flows with Variational Data Augmentation](https://arxiv.org/abs/2002.09741). The implementation is based on [Flow++: Improving Flow-Based Generative Models with Variational Dequantization and Architecture Design](https://github.com/aravindsrinivas/flowpp).
# News
- We've released the checkpoint file for CIFAR-10. See [This link](https://drive.google.com/file/d/1OkgFzOMjpaItZfw0LERwOKTRJS3MYOo4/view?usp=sharing).
# Main Dependencies
* Python >= 3.6
* Tensorflow v1.14.0
* [horovod v0.18.2](https://github.com/uber/horovod)
[Horovod GPU setup instructions](https://github.com/uber/horovod/blob/master/docs/gpus.rst)
# Usage Instructions on One Machine
We trained our models with data-parallelism using Horovod. For simply usage in one machine (e.g. 8 GPUs):
## CIFAR 10
```
horovodrun -np 8 python3 run_cifar.py
```
## ImageNet
### Data for ImageNet Experiments:
Script to create dataset [here](https://github.com/thu-ml/vflow/blob/master/flows_imagenet/create_imagenet_benchmark_datasets.py)
### ImageNet 32x32
```
horovodrun -np 8 python3 imagenet32.py
```
### ImageNet 64x64
```
horovodrun -np 8 python3 imagenet64.py
```
# Multi-Machine Usage
## Network with Infiniband
Create a file named as `hostfile`:
(Modify `$ip_for_another_machine_for_ib_interface`)
```
localhost slots=8
$ip_for_another_machine_for_ib_interface slots=8
```
Then create a script named as `run.sh`:
(Modify `$ib_network_interface`, `ib_device`, e.g. `ib0`, `mlx5_0`)
```
mpirun -np 16 \
--hostfile hostfile \
-bind-to none -map-by slot \
-x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH -x PATH -x NCCL_SOCKET_IFNAME=$ib_network_interface \
-mca pml ob1 -mca btl self,openib -mca btl_openib_if_include $ib_device \
bash -c "python /dir_to_vflow/flows/run_cifar10.py"
```
## Network with TCP
Create a file named as `hostfile`:
(Modify `$ip_for_another_machine_for_tcp_interface`)
```
localhost slots=8
$ip_for_another_machine_for_tcp_interface slots=8
```
Then create a script named as `run.sh`:
(Modify `$tcp_network_interface`, e.g. `enp94s0f0`)
```
mpirun -np 16 \
--hostfile hostfile \
-bind-to none -map-by slot \
-x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH -x PATH -x NCCL_SOCKET_IFNAME=$tcp_network_interface \
-mca pml ob1 -mca btl ^openib -mca btl_tcp_if_include $tcp_network_interface \
bash -c "python /dir_to_vflow/flows/run_cifar10.py"
```
# Contact
Please open an issue.
# Cite
Please cite our paper if you use this code in your own work:
```
@inproceedings{chen2020vflow,
title={VFlow: More Expressive Generative Flows with Variational Data Augmentation},
author={Chen, Jianfei and Lu, Cheng and Chenli, Biqi and Zhu, Jun and Tian, Tian},
booktitle={International Conference on Machine Learning},
year={2020}
}
```