v0.1.0 - Initial Release

Backpropagate v0.1.0

Headless LLM Fine-Tuning - Making fine-tuning accessible without the complexity.

Features

• Simple Python API with smart defaults (Trainer class)
• Multi-run SLAO training for better results without overfitting
• Dataset loading & filtering with perplexity-based quality control
• GGUF export with Ollama integration
• GPU safety monitoring (temperature, VRAM)
• Optional Gradio web UI
• Modular installation - install only what you need
• Full Windows support (RTX 40/50 series tested)

Installation

pip install backpropagate              # Core only
pip install backpropagate[standard]    # Recommended (unsloth + ui)
pip install backpropagate[full]        # Everything

Quick Start

from backpropagate import Trainer

trainer = Trainer("unsloth/Qwen2.5-7B-Instruct-bnb-4bit")
trainer.train("my_data.jsonl", steps=100)
trainer.save("./my-model")

See README for full documentation.