Ecomgpt: Instruction-tuning large language models with chain-of-task tasks for e-commerce

Y Li, S Ma, X Wang, S Huang, C Jiang… - Proceedings of the …, 2024‏ - ojs.aaai.org
Recently, instruction-following Large Language Models (LLMs), represented by ChatGPT,
have exhibited exceptional performance in general Natural Language Processing (NLP) …

Localizing task information for improved model merging and compression

K Wang, N Dimitriadis, G Ortiz-Jimenez… - arxiv preprint arxiv …, 2024‏ - arxiv.org
Model merging and task arithmetic have emerged as promising scalable approaches to
merge multiple single-task checkpoints to one multi-task model, but their applicability is …

Learning to route among specialized experts for zero-shot generalization

M Muqeeth, H Liu, Y Liu, C Raffel - arxiv preprint arxiv:2402.05859, 2024‏ - arxiv.org
Recently, there has been a widespread proliferation of" expert" language models that are
specialized to a specific task or domain through parameter-efficient fine-tuning. How can we …

Active instruction tuning: Improving cross-task generalization by training on prompt sensitive tasks

PN Kung, F Yin, D Wu, KW Chang, N Peng - arxiv preprint arxiv …, 2023‏ - arxiv.org
Instruction tuning (IT) achieves impressive zero-shot generalization results by training large
language models (LLMs) on a massive amount of diverse tasks with instructions. However …

What Matters for Model Merging at Scale?

P Yadav, T Vu, J Lai, A Chronopoulou… - arxiv preprint arxiv …, 2024‏ - arxiv.org
Model merging aims to combine multiple expert models into a more capable single model,
offering benefits such as reduced storage and serving costs, improved generalization, and …

Merging by matching models in task parameter subspaces

D Tam, M Bansal, C Raffel - arxiv preprint arxiv:2312.04339, 2023‏ - arxiv.org
Model merging aims to cheaply combine individual task-specific models into a single
multitask model. In this work, we view past merging methods as leveraging different notions …

Zero-shot generalization during instruction tuning: insights from similarity and granularity

B He, N Ding, C Qian, J Deng, G Cui, L Yuan… - arxiv preprint arxiv …, 2024‏ - arxiv.org
Understanding alignment techniques begins with comprehending zero-shot generalization
brought by instruction tuning, but little of the mechanism has been understood. Existing work …

Lines: Post-training layer scaling prevents forgetting and enhances model merging

K Wang, N Dimitriadis, A Favero… - arxiv preprint arxiv …, 2024‏ - arxiv.org
Large pre-trained models exhibit impressive zero-shot performance across diverse tasks,
but fine-tuning often leads to catastrophic forgetting, where improvements on a target …

DGTRL: Deep graph transfer reinforcement learning method based on fusion of knowledge and data

G Chen, J Qi, Y Gao, X Zhu, Z Dong, Y Sun - Information Sciences, 2024‏ - Elsevier
Deep reinforcement learning has shown promising application effects in many fields.
However, issues such as low sample efficiency and weak knowledge transfer and …

SparseCL: Sparse Contrastive Learning for Contradiction Retrieval

H Xu, Z Lin, Y Sun, KW Chang, P Indyk - arxiv preprint arxiv:2406.10746, 2024‏ - arxiv.org
Contradiction retrieval refers to identifying and extracting documents that explicitly disagree
with or refute the content of a query, which is important to many downstream applications …