YAML Config Manager

A robust, "LLM-friendly" Python library for managing Deep Learning configurations. It seamlessly combines YAML files with command-line overrides, supporting dotted notation, scientific notation, and flexible argument styles.

Features

• YAML + CLI: Load defaults from YAML, override via CLI.
• Dotted Access: config.model.layer_count instead of config['model']['layer_count'].
• Robust Parsing:
  • Supports scientific notation (1e-4 is a float, not string).
  • Smart boolean parsing (yes, on, true -> True).
• Flexible CLI: Supports both --param=value and --param value.
• Safety: Raises AttributeError for missing keys (no silent failures).

Installation

pip install .

Quick Start

1. Create config.yaml

training:
  lr: 1e-4
  batch_size: 32
model:
  type: transformer
  layers: 12

2. Use in Python

from yaml_config_manager import load_config

# Automatically parses sys.argv
config = load_config() 

print(f"Model: {config.model.type}")
print(f"LR: {config.training.lr} (type: {type(config.training.lr)})")

3. Run with Overrides

Mixed styles supported:

python train.py --config config.yaml --training.lr 2e-5 --model.layers=24 --debug_mode

Advanced Usage

• Manual Args: load_config(args=['--param', 'val'])
• Default Path: load_config(config_path='default.yaml')
• Missing Keys: Accessing config.missing raises AttributeError to catch typos early.