The Automated LLM Speedrunning Benchmark: Reproducing NanoGPT Improvements Paper • 2506.22419 • Published Jun 27 • 14