Skip to content

Latest commit

 

History

History
56 lines (44 loc) · 2.73 KB

README.md

File metadata and controls

56 lines (44 loc) · 2.73 KB
Raw

🌟 Lion Optimizer for TensorFlow 🌟

This is an implementation of the Lion optimizer in TensorFlow, based on the PyTorch implementation 🙏 and the original paper from Xiangning Chen et al. 👍

Modifications

The main modifications for the translation of the PyTorch code to TensorFlow were the replacement of PyTorch-specific functions with their TensorFlow equivalents, as well as some adjustments to match the TensorFlow API.

For example, the PyTorch lerp_ function was replaced with the TensorFlow tf.raw_ops.LinSpace operator to simulate the same functionality. Additionally, the PyTorch torch.no_grad() context manager was replaced with the @tf.function decorator, which is used to wrap the update_fn and step functions to improve performance.

Usage

Here's an example of how to use the Lion optimizer for a simple training task in TensorFlow:

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
from lion_tensorflow import Lion

# Define a simple neural network model
model = keras.Sequential([
    layers.Dense(64, activation='relu'),
    layers.Dense(10, activation='softmax')
])

# Instantiate the Lion optimizer
optimizer = Lion(lr=1e-3)

# Compile the model
model.compile(optimizer=optimizer, loss='sparse_categorical_crossentropy', metrics=['accuracy'])

# Load and preprocess the data
(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()
x_train = x_train.reshape(-1, 28*28) / 255.
x_test = x_test.reshape(-1, 28*28) / 255.

# Train the model
model.fit(x_train, y_train, epochs=10, validation_data=(x_test, y_test))

Reference

The Lion optimizer was introduced in the following paper:

@misc{https://doi.org/10.48550/arxiv.2302.06675,
  doi = {10.48550/ARXIV.2302.06675},  
  url = {https://arxiv.org/abs/2302.06675}, 
  author = {Chen, Xiangning and Liang, Chen and Huang, Da and Real, Esteban and Wang, Kaiyuan and Liu, Yao and Pham, Hieu and Dong, Xuanyi and Luong, Thang and Hsieh, Cho-Jui and Lu, Yifeng and Le, Quoc V.}, 
  keywords = {Machine Learning (cs.LG), Artificial Intelligence (cs.AI), Computation and Language (cs.CL), Computer Vision and Pattern Recognition (cs.CV), Neural and Evolutionary Computing (cs.NE), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {Symbolic Discovery of Optimization Algorithms},
  publisher = {arXiv},
  year = {2023},  
  copyright = {arXiv.org perpetual, non-exclusive license}
}

🚨Wanted!🚨

Contributors and testers to this project are HIGHLY WANTED! If you test it, find any bugs or have ideas for new features, please drop me a message, submit an issue or a pull request.