pyrodigy 0.1.3

No more excuses for bad models. >70 Optimizers. All explained. All pre-configured.

🔥 pyrodigy 🔥

ATTENTION - WORK IN PROGRESS - TEST DEPLOYMENT - NOT COMPLETE

pyrodigy is a Python wrapper around more than 70 optimizers from pytorch_optimizer, along with some additional custom optimizers. Designed for flexibility, pyrodigy offers easy configuration management, history tracking, and a CLI for convenience.

Features

• Access to 70+ Optimizers: Use a variety of optimizers, from well-known ones to niche algorithms.
• Config Management: View, add, set, or remove optimizer configurations directly from the CLI.
• History Tracking: Track optimizer instantiations with detailed history, including timestamps, parameters, and caller information.
• Customizable TTL: Automatically clear history entries older than a specified time-to-live (TTL).
• Rich CLI Interface: Manage configurations, view documentation, and explore history—all from the command line.

Installation

With pip

pip install pyrodigy

or

Clone the repo and install pyrodigy using Poetry:

poetry install

Dependencies

Note: Pyrodigy requires PyTorch to be installed separately. You can install it based on your specific environment (CPU or GPU) and operating system. Follow the PyTorch installation guide for instructions.

Usage

CLI Commands

The CLI commands allow you to list optimizers, manage configurations, and handle history entries.

List Available Optimizers

Displays a list of optimizers with both config and documentation:

pyrodigy list

Show Optimizer Documentation

Prints the Markdown documentation for the specified optimizer:

pyrodigy show <optimizer_name>

Configuration Management

Manage optimizer configurations using the config command with get, set, add, and rm actions.

• View Configuration:

  pyrodigy config <optimizer_name> get
  

• Set Configuration: Update an existing configuration with new values (JSON format).

  pyrodigy config <optimizer_name> set '{"default": {"lr": 0.01, "beta": 0.9}}'
  

• Add New Configuration: Add a new named configuration (JSON format).

  pyrodigy config <optimizer_name> add <config_name> '{"lr": 0.02, "beta": 0.95}'
  

• Remove Configuration: Remove a named configuration.

  pyrodigy config <optimizer_name> rm <config_name>
  

History Management

Each time an optimizer is instantiated, an entry is created in its history. You can review or clear history and apply a TTL to automatically remove old entries.

• Show History: View the history for an optimizer. Specify a TTL to filter entries within a certain timeframe.

  pyrodigy history <optimizer_name> show --TTL 30d
  

  Example with TTL:

  pyrodigy history a2grad show --TTL 60d
  

• Clear History: Remove all history entries for the optimizer.

  pyrodigy history <optimizer_name> clear
  

Example: Using pyrodigy in Code

Instantiate an optimizer with pyrodigy’s Wrapper, which logs the creation details to the optimizer's history.

from pyrodigy.wrapper import Wrapper

# Define model parameters and optimizer configuration
params = model.parameters()
optimizer_name = "AdamP"
config_name = "default"
lr = 0.001

# Initialize the optimizer
optimizer = Wrapper(params, optimizer_name=optimizer_name, config_name=config_name, lr=lr)

History Entries

Every time you create an optimizer instance, the following details are saved:

• Optimizer Name: The name of the optimizer.
• Config Name: The configuration used, if provided.
• Parameters: Any additional parameters such as learning rate.
• Caller Information: File, line number, and function name where the optimizer was instantiated.

License

Licensed under the Apache License 2.0. See the LICENSE file for details.

Contributing

Contributions are welcome! If you find any issues or have suggestions, feel free to open an issue or submit a pull request.

Support

For questions or support, please open an issue on GitHub.