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

• URL: http://arxiv.org/abs/2606.29082v1
• Date: Sat, 27 Jun 2026 20:48:18 GMT
• Title: Evolution Fine-Tuning: Learning to Discover Across 371 Optimization Tasks
• Abstract summary: We introduce Evolution Fine-Tuning (EFT), a mid-training paradigm that teaches Large Language Models to evolve solutions across tasks.<n>We construct Finch Collection, a 156K-trajectory dataset spanning 10 domains and 371 optimization tasks.<n>Our models surpass their base counterparts by 10.22% on average across 22 held-out tasks.
• Score: 36.58007538375195
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• 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.

Related papers

• Model Merging: Foundations and Algorithms [4.528573838858818]
  This thesis studies model merging, combining independently trained neural networks directly in weight space.<n>In the single-task setting, we introduce C$2$M$3$, a cycle-consistent merging algorithm based on Frank-Wolfe optimization.<n>In the multi-task setting, we first develop a theoretical account of task vectors as approximate gradients.<n>We then present MASS, an input-adaptive routing method that uses TSV geometry to select task-relevant subspaces at inference time.
  arXiv  Detail & Related papers  (2026-05-02T19:06:35Z)
• ThetaEvolve: Test-time Learning on Open Problems [110.5756538358217]
  We introduce ThetaEvolve, an open-source framework that simplifies and extends AlphaEvolve to efficiently scale both in-context learning and Reinforcement Learning (RL) at test time.<n>We find that ThetaEvolve with RL at test-time consistently outperforms inference-only baselines.
  arXiv  Detail & Related papers  (2025-11-28T18:58:14Z)
• MM-HELIX: Boosting Multimodal Long-Chain Reflective Reasoning with Holistic Platform and Adaptive Hybrid Policy Optimization [103.74675519953898]
  Long-chain reflective reasoning is a prerequisite for solving complex real-world problems.<n>We build a benchmark consisting 1,260 samples of 42 challenging synthetic tasks.<n>We generate post-training data and explore learning paradigms for exploiting such data.
  arXiv  Detail & Related papers  (2025-10-09T17:53:58Z)
• Staying in the Sweet Spot: Responsive Reasoning Evolution via Capability-Adaptive Hint Scaffolding [59.60915947702282]
  Reinforcement learning with verifiable rewards (RLVR) has achieved remarkable success in enhancing the reasoning capabilities of large language models (LLMs)<n>Existing RLVR methods often suffer from exploration inefficiency due to mismatches between the training data's difficulty and the model's capability.<n>We propose SEELE, a novel supervision-aided RLVR framework that dynamically adjusts problem difficulty to stay within the high-efficiency region.
  arXiv  Detail & Related papers  (2025-09-08T17:36:21Z)
• Self-Improving Language Models for Evolutionary Program Synthesis: A Case Study on ARC-AGI [22.148355836548365]
  Many program synthesis tasks prove too challenging for even state-of-the-art language models to solve in single attempts.<n>We propose SOAR, a method that learns program synthesis by integrating language models into a self-improving evolutionary loop.
  arXiv  Detail & Related papers  (2025-07-10T15:42:03Z)
• Modeling Multi-Task Model Merging as Adaptive Projective Gradient Descent [72.10987117380584]
  Merging multiple expert models offers a promising approach for performing multi-task learning without accessing their original data.<n>We find existing methods discard task-specific information that, while causing conflicts, is crucial for performance.<n>Our approach consistently outperforms previous methods, achieving state-of-the-art results across diverse architectures and tasks in both vision and NLP domains.
  arXiv  Detail & Related papers  (2025-01-02T12:45:21Z)
• Probing the Robustness of Theory of Mind in Large Language Models [6.7932860553262415]
  We introduce a novel dataset of 68 tasks for probing ToM in LLMs. We evaluate the ToM performance of four SotA open source LLMs on our dataset and the dataset introduced by (Kosinski, 2023) We find a consistent tendency in all tested LLMs to perform poorly on tasks that require the realization that an agent has knowledge of automatic state changes in its environment.
  arXiv  Detail & Related papers  (2024-10-08T18:13:27Z)
• NEVIS'22: A Stream of 100 Tasks Sampled from 30 Years of Computer Vision Research [96.53307645791179]
  We introduce the Never-Ending VIsual-classification Stream (NEVIS'22), a benchmark consisting of a stream of over 100 visual classification tasks. Despite being limited to classification, the resulting stream has a rich diversity of tasks from OCR, to texture analysis, scene recognition, and so forth. Overall, NEVIS'22 poses an unprecedented challenge for current sequential learning approaches due to the scale and diversity of tasks.
  arXiv  Detail & Related papers  (2022-11-15T18:57:46Z)
• An Automated Question-Answering Framework Based on Evolution Algorithm [19.054115603616513]
  We propose an automated Question-Answering framework, which could adjust network architecture for multiple datasets. Our framework achieves 78.9 EM and 86.1 F1 on SQuAD 1.1, 69.9 EM and 72.5 F1 on SQuAD 2.0.
  arXiv  Detail & Related papers  (2022-01-26T08:13:24Z)
• Lifelong Learning with Searchable Extension Units [21.17631355880764]
  We propose a new lifelong learning framework named Searchable Extension Units (SEU) It breaks down the need for a predefined original model and searches for specific extension units for different tasks. Our approach can obtain a much more compact model without catastrophic forgetting.
  arXiv  Detail & Related papers  (2020-03-19T03:45:51Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.