On Giant's Shoulders: Effortless Weak to Strong by Dynamic Logits Fusion

• URL: http://arxiv.org/abs/2406.15480v2
• Date: Mon, 14 Oct 2024 11:31:06 GMT
• Title: On Giant's Shoulders: Effortless Weak to Strong by Dynamic Logits Fusion
• Authors: Chenghao Fan, Zhenyi Lu, Wei Wei, Jie Tian, Xiaoye Qu, Dangyang Chen, Yu Cheng,
• Abstract summary: Existing weak-to-strong methods often employ a static knowledge transfer ratio and a single small model for transferring complex knowledge. We propose a dynamic logit fusion approach that works with a series of task-specific small models, each specialized in a different task. Our method closes the performance gap by 96.4% in single-task scenarios and by 86.3% in multi-task scenarios.
• Score: 23.63688816017186
• License:
• Abstract: Efficient fine-tuning of large language models for task-specific applications is imperative, yet the vast number of parameters in these models makes their training increasingly challenging. Despite numerous proposals for effective methods, a substantial memory overhead remains for gradient computations during updates. \thm{Can we fine-tune a series of task-specific small models and transfer their knowledge directly to a much larger model without additional training?} In this paper, we explore weak-to-strong specialization using logit arithmetic, facilitating a direct answer to this question. Existing weak-to-strong methods often employ a static knowledge transfer ratio and a single small model for transferring complex knowledge, which leads to suboptimal performance. % To address this, To surmount these limitations, we propose a dynamic logit fusion approach that works with a series of task-specific small models, each specialized in a different task. This method adaptively allocates weights among these models at each decoding step, learning the weights through Kullback-Leibler divergence constrained optimization problems. We conduct extensive experiments across various benchmarks in both single-task and multi-task settings, achieving leading results. By transferring expertise from the 7B model to the 13B model, our method closes the performance gap by 96.4\% in single-task scenarios and by 86.3\% in multi-task scenarios compared to full fine-tuning of the 13B model. Notably, we achieve surpassing performance on unseen tasks. Moreover, we further demonstrate that our method can effortlessly integrate in-context learning for single tasks and task arithmetic for multi-task scenarios.

Related papers

• Merging Multi-Task Models via Weight-Ensembling Mixture of Experts [64.94129594112557]
  Merging Transformer-based models trained on different tasks into a single unified model can execute all the tasks concurrently. Previous methods, exemplified by task arithmetic, have been proven to be both effective and scalable. We propose to merge most of the parameters while upscaling the Transformer layers to a weight-ensembling mixture of experts (MoE) module.
  arXiv  Detail & Related papers  (2024-02-01T08:58:57Z)
• Concrete Subspace Learning based Interference Elimination for Multi-task Model Fusion [86.6191592951269]
  Merging models fine-tuned from common extensively pretrained large model but specialized for different tasks has been demonstrated as a cheap and scalable strategy to construct a multitask model that performs well across diverse tasks. We propose the CONtinuous relaxation dis (Concrete) subspace learning method to identify a common lowdimensional subspace and utilize its shared information track interference problem without sacrificing performance.
  arXiv  Detail & Related papers  (2023-12-11T07:24:54Z)
• Scalarization for Multi-Task and Multi-Domain Learning at Scale [15.545810422759295]
  Training a single model on multiple input domains and/or output tasks allows for compressing information from multiple sources into a unified backbone. However, optimizing such networks is a challenge due to discrepancies between the different tasks or domains.
  arXiv  Detail & Related papers  (2023-10-13T07:31:04Z)
• Multi-Objective Optimization for Sparse Deep Multi-Task Learning [0.0]
  We present a Multi-Objective Optimization algorithm using a modified Weighted Chebyshev scalarization for training Deep Neural Networks (DNNs) Our work aims to address the (economical and also ecological) sustainability issue of DNN models, with particular focus on Deep Multi-Task models.
  arXiv  Detail & Related papers  (2023-08-23T16:42:27Z)
• An Evolutionary Approach to Dynamic Introduction of Tasks in Large-scale Multitask Learning Systems [4.675744559395732]
  Multitask learning assumes that models capable of learning from multiple tasks can achieve better quality and efficiency via knowledge transfer. State of the art ML models rely on high customization for each task and leverage size and data scale rather than scaling the number of tasks. We propose an evolutionary method that can generate a large scale multitask model and can support the dynamic and continuous addition of new tasks.
  arXiv  Detail & Related papers  (2022-05-25T13:10:47Z)
• Sparsely Activated Mixture-of-Experts are Robust Multi-Task Learners [67.5865966762559]
  We study whether sparsely activated Mixture-of-Experts (MoE) improve multi-task learning. We devise task-aware gating functions to route examples from different tasks to specialized experts. This results in a sparsely activated multi-task model with a large number of parameters, but with the same computational cost as that of a dense model.
  arXiv  Detail & Related papers  (2022-04-16T00:56:12Z)
• Task Adaptive Parameter Sharing for Multi-Task Learning [114.80350786535952]
  Adaptive Task Adapting Sharing (TAPS) is a method for tuning a base model to a new task by adaptively modifying a small, task-specific subset of layers. Compared to other methods, TAPS retains high accuracy on downstream tasks while introducing few task-specific parameters. We evaluate our method on a suite of fine-tuning tasks and architectures (ResNet, DenseNet, ViT) and show that it achieves state-of-the-art performance while being simple to implement.
  arXiv  Detail & Related papers  (2022-03-30T23:16:07Z)
• Controllable Dynamic Multi-Task Architectures [92.74372912009127]
  We propose a controllable multi-task network that dynamically adjusts its architecture and weights to match the desired task preference as well as the resource constraints. We propose a disentangled training of two hypernetworks, by exploiting task affinity and a novel branching regularized loss, to take input preferences and accordingly predict tree-structured models with adapted weights.
  arXiv  Detail & Related papers  (2022-03-28T17:56:40Z)
• XtremeDistilTransformers: Task Transfer for Task-agnostic Distillation [80.18830380517753]
  We develop a new task-agnostic distillation framework XtremeDistilTransformers. We study the transferability of several source tasks, augmentation resources and model architecture for distillation.
  arXiv  Detail & Related papers  (2021-06-08T17:49:33Z)
• Conditionally Adaptive Multi-Task Learning: Improving Transfer Learning in NLP Using Fewer Parameters & Less Data [5.689320790746046]
  Multi-Task Learning (MTL) networks have emerged as a promising method for transferring learned knowledge across different tasks. However, MTL must deal with challenges such as: overfitting to low resource tasks, catastrophic forgetting, and negative task transfer. We propose a novel Transformer architecture consisting of a new conditional attention mechanism and a set of task-conditioned modules.
  arXiv  Detail & Related papers  (2020-09-19T02:04:34Z)

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.