The Automated LLM Speedrunning Benchmark: Reproducing NanoGPT Improvements

TL;DR

This paper introduces an automated benchmark to evaluate AI agents' ability to reproduce and improve LLM training results, highlighting current limitations of reasoning LLMs in replicating known innovations.

Contribution

The paper presents the Automated LLM Speedrunning Benchmark, a realistic and accessible test for AI agents to reproduce and optimize LLM training improvements.

Findings

Recent reasoning LLMs struggle to reimplement known innovations

The benchmark covers diverse code-level improvements

It provides a measure of AI's ability to automate scientific reproduction

Abstract

Rapid advancements in large language models (LLMs) have the potential to assist in scientific progress. A critical capability toward this endeavor is the ability to reproduce existing work. To evaluate the ability of AI agents to reproduce results in an active research area, we introduce the Automated LLM Speedrunning Benchmark, leveraging the research community contributions on the NanoGPT speedrun, a competition to train a GPT-2 model in the shortest time. Each of the 19 speedrun tasks provides the agent with the previous records training script, optionally paired with one of three hint formats, ranging from pseudocode to paper-like descriptions of the new records improvements. Records execute quickly by design and speedrun improvements encompass diverse code-level changes, ranging from high-level algorithmic advancements to hardware-aware optimizations. These features make the benchmark both accessible and realistic for the frontier problem of improving LLM training. We find that recent reasoning LLMs combined with SoTA scaffolds struggle to reimplement already-known innovations in our benchmark, even when given detailed hints. Our benchmark thus provides a simple, non-saturated measure of an LLMs ability to automate scientific reproduction, a necessary (but not sufficient) skill for an autonomous research agent.