Frontis-MA1: Training an AI4AI Model towards Recursive Self-Improvement in Machine Learning Engineering
A 35B model trained to evolve machine-learning code beat its base model sharply on MLE-Bench Lite under a constrained 12-hour setup.
The paper introduces OpenMLE, a full-stack testbed for recursive self-improvement research in machine learning engineering. Frontis-MA1 is post-trained around four program-evolution operators: Draft, Improve, Debug, and Crossover. With OpenMLE-Evo, it raises Medal Average from 39.39% to 60.61% over its base model, and reaches 71.21% with OpenMLE-Evo-Max. The authors say the model weights and OpenMLE stack are being released for reproducible AI4AI research. HF Daily Papers' note
The paper introduces OpenMLE, a full-stack testbed for recursive self-improvement research in machine learning engineering. Frontis-MA1 is post-trained around four program-evolution operators: Draft, Improve, Debug, and Crossover. With OpenMLE-Evo, it raises Medal Average from 39.39% to 60.61% over its base model, and reaches 71.21% with OpenMLE-Evo-Max. The authors say the model weights and OpenMLE stack are being released for reproducible AI4AI research. HF Daily Papers' note
score 6