Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Vaina, Sofia Maria Lo Cicero, Chumachenko, Artem, Ryabinin, Max
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2603.10156
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911504145580032
author Vaina, Sofia Maria Lo Cicero
Chumachenko, Artem
Ryabinin, Max
author_facet Vaina, Sofia Maria Lo Cicero
Chumachenko, Artem
Ryabinin, Max
contents Finetuning on domain-specific data is a well-established method for enhancing LLM performance on downstream tasks. Training on each dataset produces a new set of model weights, resulting in a multitude of checkpoints saved in-house or on open-source platforms. However, these training artifacts are rarely reused for subsequent experiments despite containing improved model abilities for potentially similar tasks. In this paper, we propose Mashup Learning, a simple method to leverage the outputs of prior training runs to enhance model adaptation to new tasks. Our procedure identifies the most relevant historical checkpoints for a target dataset, aggregates them with model merging, and uses the result as an improved initialization for training. Across 8 standard LLM benchmarks, four models, and two collections of source checkpoints, Mashup Learning consistently improves average downstream accuracy by 0.5-5 percentage points over training from scratch. It also accelerates convergence, requiring 41-46% fewer training steps and up to 37% less total wall-clock time to match from-scratch accuracy, including all selection and merging overhead.
format Preprint
id arxiv_https___arxiv_org_abs_2603_10156
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Mashup Learning: Faster Finetuning by Remixing Past Checkpoints
Vaina, Sofia Maria Lo Cicero
Chumachenko, Artem
Ryabinin, Max
Machine Learning
Artificial Intelligence
Finetuning on domain-specific data is a well-established method for enhancing LLM performance on downstream tasks. Training on each dataset produces a new set of model weights, resulting in a multitude of checkpoints saved in-house or on open-source platforms. However, these training artifacts are rarely reused for subsequent experiments despite containing improved model abilities for potentially similar tasks. In this paper, we propose Mashup Learning, a simple method to leverage the outputs of prior training runs to enhance model adaptation to new tasks. Our procedure identifies the most relevant historical checkpoints for a target dataset, aggregates them with model merging, and uses the result as an improved initialization for training. Across 8 standard LLM benchmarks, four models, and two collections of source checkpoints, Mashup Learning consistently improves average downstream accuracy by 0.5-5 percentage points over training from scratch. It also accelerates convergence, requiring 41-46% fewer training steps and up to 37% less total wall-clock time to match from-scratch accuracy, including all selection and merging overhead.
title Mashup Learning: Faster Finetuning by Remixing Past Checkpoints
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2603.10156