Frontis-MA1：訓練一個AI4AI模型以實現機器學習工程中的遞迴自我改進

摘要

遞迴自我改進（RSI）需要能改進 AI 建構過程的 AI 系統（即 AI4AI）；機器學習工程（MLE）為研究此能力提供了具體且可執行的測試平台。我們提出 OpenMLE，這是一個用於 MLE 中 RSI 研究的開放式全端系統，涵蓋具執行回饋的可驗證任務環境（OpenMLE-Gym）、算子學習（OpenMLE-RL），以及長時程搜尋（OpenMLE-Evo）。在此技術堆疊上，我們對 Frontis-MA1（35B）進行後訓練，使之成為用於 MLE 的後設演化代理，並使後訓練與推論圍繞四個原子程式演化算子（Draft、Improve、Debug、Crossover）對齊：這些算子透過以執行結果為基礎的 SFT 與 RL，在與所有評測基準去重後的資料上進行訓練，接著組合成長時程搜尋，在單一迴圈中耦合學習與演化。在 MLE-Bench Lite 上，於單張 RTX 4090（VRAM 上限 12 GB）、每項任務 12 小時的預算下，Frontis-MA1（35B）使用 OpenMLE-Evo 時，將 Medal Average 從其基礎模型的 39.39% 提升至 60.61%；使用 OpenMLE-Evo-Max（與基準無關的經驗先驗與非同步搜尋）時則達到 71.21%，超越 GPT-5.5 + Codex，並接近 GPT-5.6 Sol 與 2.8T Kimi K3。在保留的 NatureBench Lite 上，兩個組成部分皆可遷移：框架固定時，換用訓練後的模型可將 Match-SOTA 從 50% 提升至 70%；模型固定時，換用 OpenMLE-Evo 可將其從 20% 提升至 50%。我們釋出模型權重與完整 OpenMLE 技術堆疊，以促成朝向 RSI 之可執行 AI4AI 的可重現研究。程式碼：https://github.com/FrontisAI/OpenRSI

English

Recursive self-improvement (RSI) requires AI systems that improve the process of building AI (i.e., AI4AI); machine learning engineering (MLE) offers a concrete, executable testbed for studying this capability. We introduce OpenMLE, an open full-stack system for RSI research in MLE, spanning verifiable task environments with execution feedback (OpenMLE-Gym), operator learning (OpenMLE-RL), and long-horizon search (OpenMLE-Evo). On this stack we post-train Frontis-MA1 (35B) as a meta-evolution agent for MLE, aligning post-training and inference around four atomic program-evolution operators (Draft, Improve, Debug, Crossover): the same operators are trained via execution-grounded SFT and RL on data deduplicated against all evaluation benchmarks, then composed into long-horizon search, coupling learning and evolution in a single loop. On MLE-Bench Lite under a 12-hour per-task budget on one RTX 4090 capped at 12 GB VRAM, Frontis-MA1 (35B) improves Medal Average from 39.39% to 60.61% over its base model with OpenMLE-Evo, and reaches 71.21% with OpenMLE-Evo-Max (benchmark-independent experience priors and asynchronous search), exceeding GPT-5.5 + Codex and approaching GPT-5.6 Sol and the 2.8T Kimi K3. On held-out NatureBench Lite, both components transfer: with the framework fixed, swapping in the trained model raises Match-SOTA from 50% to 70%; with the model fixed, swapping in OpenMLE-Evo raises it from 20% to 50%. We release the model weights and the full OpenMLE stack to enable reproducible research on executable AI4AI toward RSI. Code: https://github.com/FrontisAI/OpenRSI