Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/bonifatius94/pytorch-keras-example
Example on how to run PyTorch with Keras 3.0
https://github.com/bonifatius94/pytorch-keras-example
Last synced: 4 days ago
JSON representation
Example on how to run PyTorch with Keras 3.0
- Host: GitHub
- URL: https://github.com/bonifatius94/pytorch-keras-example
- Owner: Bonifatius94
- Created: 2024-02-24T20:41:28.000Z (9 months ago)
- Default Branch: main
- Last Pushed: 2024-05-19T10:05:08.000Z (6 months ago)
- Last Synced: 2024-05-19T11:24:39.724Z (6 months ago)
- Language: Python
- Size: 5.86 KB
- Stars: 1
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# PyTorch Keras Example
## About
This repository outlines how to use PyTorch as backend for the Keras 3.0 API.
As PyTorch has a painless GPU setup for AI trainings, it might be interesting
to use PyTorch under the hood, but with the familiar syntax from TensorFlow Keras.
## GPU Setup
First, install a proprietary NVIDIA driver for your GPU (version 535 in this case).
```sh
sudo apt-get update && sudo apt-get install -y nvidia-driver-535 nvidia-dkms-535
```
GPU driver changes usually require a reboot to take effect.
```sh
sudo reboot
```
According to the "getting started" [page](https://pytorch.org/get-started/locally/)
of PyTorch, execute following command to set up PyTorch to use GPUs.
```sh
python -m pip install torch torchvision torchaudio
```
Additionally, install the Keras Core 3.0 package to make use of Keras in PyTorch.
Unlike TensorFlow, Keras is not automatically shipped with PyTorch. It needs to be
installed separately.
```sh
python -m pip install keras-core
```
## Training
Now, it's time to check if the setup was successful. Open a console to monitor the GPU.
```sh
watch nvidia-smi
```
Launch another console to run the training script and see if it utilizes the GPU as expected.
```sh
python keras.py
```
Outputs should be something like this:
```text
Using PyTorch backend.
Epoch 1/10
938/938 ━━━━━━━━━━━━━━━━━━━━ 4s 4ms/step - loss: 0.4897 - sparse_categorical_accuracy: 0.8403 - val_loss: 0.0517 - val_sparse_categorical_accuracy: 0.9824
Epoch 2/10
938/938 ━━━━━━━━━━━━━━━━━━━━ 4s 4ms/step - loss: 0.0926 - sparse_categorical_accuracy: 0.9717 - val_loss: 0.0438 - val_sparse_categorical_accuracy: 0.9863
Epoch 3/10
938/938 ━━━━━━━━━━━━━━━━━━━━ 4s 4ms/step - loss: 0.0756 - sparse_categorical_accuracy: 0.9766 - val_loss: 0.0373 - val_sparse_categorical_accuracy: 0.9863
Epoch 4/10
938/938 ━━━━━━━━━━━━━━━━━━━━ 4s 4ms/step - loss: 0.0628 - sparse_categorical_accuracy: 0.9808 - val_loss: 0.0334 - val_sparse_categorical_accuracy: 0.9876
Epoch 5/10
938/938 ━━━━━━━━━━━━━━━━━━━━ 4s 4ms/step - loss: 0.0520 - sparse_categorical_accuracy: 0.9835 - val_loss: 0.0252 - val_sparse_categorical_accuracy: 0.9913
Epoch 6/10
938/938 ━━━━━━━━━━━━━━━━━━━━ 4s 4ms/step - loss: 0.0498 - sparse_categorical_accuracy: 0.9854 - val_loss: 0.0304 - val_sparse_categorical_accuracy: 0.9897
Epoch 7/10
938/938 ━━━━━━━━━━━━━━━━━━━━ 4s 4ms/step - loss: 0.0499 - sparse_categorical_accuracy: 0.9848 - val_loss: 0.0263 - val_sparse_categorical_accuracy: 0.9913
Epoch 8/10
938/938 ━━━━━━━━━━━━━━━━━━━━ 4s 4ms/step - loss: 0.0396 - sparse_categorical_accuracy: 0.9875 - val_loss: 0.0279 - val_sparse_categorical_accuracy: 0.9906
Epoch 9/10
938/938 ━━━━━━━━━━━━━━━━━━━━ 4s 4ms/step - loss: 0.0392 - sparse_categorical_accuracy: 0.9877 - val_loss: 0.0242 - val_sparse_categorical_accuracy: 0.9914
Epoch 10/10
938/938 ━━━━━━━━━━━━━━━━━━━━ 4s 4ms/step - loss: 0.0366 - sparse_categorical_accuracy: 0.9882 - val_loss: 0.0286 - val_sparse_categorical_accuracy: 0.9914
```
```text
Sat Feb 24 21:08:26 2024
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.154.05 Driver Version: 535.154.05 CUDA Version: 12.2 |
|-----------------------------------------+----------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+======================+======================|
| 0 NVIDIA GeForce RTX 3060 Off | 00000000:01:00.0 Off | N/A |
| 0% 34C P2 63W / 170W | 216MiB / 12288MiB | 39% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
+---------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
| 0 N/A N/A 38167 C python 210MiB |
+---------------------------------------------------------------------------------------+
```
In this case NVIDIA schedules a compute task (Type C) for our python training with 39% GPU utilization.