# LM Bench

**LM Bench** is a command-line tool that benchmarks your system's capability to train large language models (LLMs), starting from 100 million parameters and scaling up. It evaluates your hardware and provides a practical score based on the largest model your system can handle.

## Features

- Benchmarks training capacity from 100M parameters upward
- Scores based on memory usage, training stability, and throughput
- Supports PyTorch with mixed precision (bfloat16/fp16)
- Compatible with both GPU and Apple MPS backends
- Minimal setup, fast results

