RustGPT

A character-level GPT in Rust using Candle with Metal acceleration on Apple Silicon.

Requirements

• Rust 1.70+
• macOS for Metal (CPU fallback is automatic)

Quick Start

1. Put your training text in input.txt (TinyShakespeare works: https://github.com/karpathy/char-rnn/blob/master/data/tinyshakespeare/input.txt).
2. Run: cargo run --release

Defaults (src/main.rs)

• batch_size: 32
• block_size: 256
• max_iters: 8000
• eval_interval: 1000
• eval_iters: 100
• learning_rate: 3e-4
• n_embd: 192
• n_head: 6
• n_layer: 4
• dropout: 0.1
• max_new_tokens: 500
• temperature: 0.9
• top_k: 40

Output

• Prints train/val loss during training
• Prints a generated sample
• Writes model.safetensors and model.safetensors.meta.json

Notes

• Uses Metal if available, otherwise CPU.
• Generation is seeded from a random slice of the training data for better coherence.