Evolution Fine-Tuning: Learning to Discover Across 371 Optimization Tasks

2026-06-27

AI Summary

This work addresses the challenge of enabling large language models (LLMs) to leverage prior evolutionary search experience for new optimization tasks, thereby avoiding costly retraining from scratch. The authors propose Evolutionary Fine-Tuning (EFT), a method that transforms cross-task evolutionary trajectories into supervised signals to train LLMs to internalize generalizable solution evolution capabilities, eliminating reliance on external search scaffolds. Using the Finch Collection—a dataset spanning 10 domains, 371 tasks, and 156K evolutionary trajectories—the approach applies intermediate-stage fine-tuning to open-source LLMs with 2B–9B parameters, augmented by test-time reinforcement learning. Evaluated on 22 held-out tasks, EFT achieves an average performance improvement of 10.22% and matches or surpasses state-of-the-art results on circle packing and the Erdős minimum overlap problem.

Abstract

Would experience designing faster GPU kernels also help close in on a long-standing open mathematical conjecture? Large Language Models (LLMs) integrated into evolutionary search have recently produced state-of-the-art solutions on optimization tasks, including open mathematical conjectures, GPU kernel design, scientific law discovery, and combinatorial puzzles. To achieve this, prior work applied search scaffolds to one target task at a time, so every new problem is approached from scratch and the experience accumulated during search is discarded once the model finishes its attempt. This leaves the capability of iteratively evolving a solution (e.g., knowing which part to mutate and how, deciding when to backtrack) entirely in the scaffold rather than in the model itself. Whether the model itself could acquire this capability and reuse it across different tasks has been largely unexamined. To address this, we introduce Evolution Fine-Tuning (EFT), a mid-training paradigm that teaches LLMs to evolve solutions across tasks by converting evolutionary search trajectories into supervision. We construct Finch Collection, a 156K-trajectory dataset spanning 10 domains and 371 optimization tasks, and fine-tune open-source LLMs from 2B to 9B parameters. Empirically, EFT confers cross-task generalization: across 22 held-out tasks, our models surpass their base counterparts by 10.22% on average. Furthermore, when paired with test-time RL, our model matches state-of-the-art performance on two circle-packing tasks and outperforms its base-model counterpart on the Erdős minimum-overlap problem. EFT thus serves as a "practice phase" for general-purpose discovery agents that do not solve new problems from scratch.

Problem

• evolutionary search
• cross-task generalization
• large language models
• optimization tasks
• iterative evolution

Innovation

• Evolution Fine-Tuning
• cross-task generalization
• evolutionary search
• large language models
• optimization tasks