Challenge: In language model training, it is difficult to obtain the right data mixtures for various tasks as the relationship between data and tasks is difficult.
Approach: They propose to identify checkpoint models based on their respective capabilities and leverage them as data mixers by using their aggregated first-order influence approximation over source data.
Outcome: The proposed framework shows significant improvements on eight reasoning benchmarks, with accuracy increases of up to 1.93%.

Similar Papers

Practical Guidelines for Model Merging in LLMs Pre-Training (2026.acl-industry)

Copied to clipboard

Challenge: Existing studies on model merging have focused on stable learning rate regimes, but its effectiveness during LLM pre-training remains underexplored.
Approach: They systematically investigate model merging across training phases, focusing on the transition from stable to decaying learning rates.
Outcome: The proposed methods improve performance during stable learning rate regimes but diminish under decay, a phe-nomenon that is linked to reduced checkpoint diversity and lower parameter-space variability.
Leveraging Pre-trained Checkpoints for Sequence Generation Tasks (2020.tacl-1)

Copied to clipboard

Challenge: Unsupervised pre-training of large neural models has revolutionized Natural Language Processing.
Approach: They propose to use pre-trained checkpoints for Sequence Generation to initialize a Transformer-based sequence-to-sequence model that is compatible with these checkpoint.
Outcome: The proposed model is compatible with pre-trained BERT, GPT-2, and RoBERTa checkpoints and achieves state-of-the-art results on Machine Translation, Text Summarization, Sentence Splitting, and Sentance Fusion.
MixtureKit: A General Framework for Composing, Training, and Visualizing Mixture-of-Experts Models (2026.acl-demo)

Copied to clipboard

Challenge: MixtureKit is a modular open-source framework for constructing, training, and analyzing Mixture-of-Experts (MoE) models from arbitrary pre-trained or fine-tuned checkpoints.
Approach: They propose a modular open-source framework for constructing, training, and analyzing Mixture-of-Experts (MoE) models from arbitrary pre-trained or fine-tuned checkpoints.
Outcome: Experiments on multilingual code-switched (Arabic–Latin) show that BTX models built with MixtureKit outperform dense baselines across multiple benchmarks.
Enhancing Language Generation with Effective Checkpoints of Pre-trained Language Model (2021.findings-acl)

Copied to clipboard

Challenge: Existing methods to exploit PrLMs for NLG tasks do not get as much performance gain as in the NLU task.
Approach: They propose a method to integrate public checkpoints of PrLMs for the most convenience.
Outcome: The proposed method significantly improves the quality of the language generation tasks on 6 different kinds of PrLMs.
Free Lunch: Robust Cross-Lingual Transfer via Model Checkpoint Averaging (2023.acl-long)

Copied to clipboard

Challenge: Massively multilingual language models have shown strong performance in zero-shot (ZS-XLT) and few-shot cross-lingual transfer setups where models are fine-tuned on task data in a source language are transferred without any or with only a few annotated instances to the target language(s).
Approach: They propose a method that averages different checkpoints during task fine-tuning to improve model robustness.
Outcome: The proposed method overestimates model performance in cross-lingual transfer setups where models are evaluated at checkpoints that generalize best to validation instances in the target languages.
Revisiting Checkpoint Averaging for Neural Machine Translation (2022.findings-aacl)

Copied to clipboard

Challenge: Checkpoint averaging is a simple and effective method to boost the performance of converged neural machine translation models.
Approach: They propose to use checkpoint averaging to increase model performance . they also propose to calculate weighted average instead of simple mean .
Outcome: The proposed method is widely adopted in neural machine translation research.
How Can We Know What Language Models Know? (2020.tacl-1)

Copied to clipboard

Challenge: Recent work examines knowledge contained in language models by having the LM fill in the blanks of prompts such as “Obama is a __ by profession”.
Approach: They propose mining-based and paraphrasing-based methods to automatically generate high-quality and diverse prompts, as well as ensemble methods to combine answers from different prompts.
Outcome: The proposed methods improve accuracy from 31.1% to 39.6% on the LAMA benchmark for extracting relational knowledge from LMs.
MergeME: Model Merging Techniques for Homogeneous and Heterogeneous MoEs (2025.naacl-long)

Copied to clipboard

Challenge: State-of-the-art methods for merging expert models with different architectures do not address parameter interference and require extensive fine-tuning to restore performance.
Approach: They propose a method for merging experts with different architectures into a unified Mixture-of-Experts model with a goal of enhancing performance in each domain while retaining effectiveness on general tasks.
Outcome: Experiments across multiple domains show that the proposed methods reduce fine-tuning costs and improve performance over state-of-the-art methods.
AMRs Assemble! Learning to Ensemble with Autoregressive Models for AMR Parsing (2023.acl-short)

Copied to clipboard

Challenge: et al., 2013) examines the current state-of-the-art in AMR parsing . current models violate structural constraints, but they can corrupt graphs .
Approach: They propose two new ensemble strategies to improve AMR parsing robustness and reduce computational time.
Outcome: The proposed methods improve robustness to structural constraints while reducing computational time.
Online Distilling from Checkpoints for Neural Machine Translation (N19-1)

Copied to clipboard

Challenge: Existing neural machine translation models have a deep structure with large amounts of parameters, making them hard to train.
Approach: They propose an online method to generate a teacher model from checkpoints . they show steady improvement over a strong self-attention-based baseline system .
Outcome: The proposed method improves on-the-fly on several datasets and language pairs.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations