GitHub - karpathy/autoresearch: AI agents running research on single-GPU nanochat training automatically · GitHub

Karpathy's autonomous ML research agent. You point it at a single-GPU training script; it runs 5-minute experiments overnight, tweaking architecture, hyperparameters, and optimizer based on validation metrics. No human in the loop between iterations. Three files do the work: a fixed data prep script, a train.py the agent modifies, and a program.md where humans set research direction. Validation bits-per-byte keeps comparisons fair across architectures.

Visit Original Article →