Nudging: Inference-time Alignment via Model Collaboration

• URL: http://arxiv.org/abs/2410.09300v2
• Date: Tue, 15 Oct 2024 01:07:01 GMT
• Title: Nudging: Inference-time Alignment via Model Collaboration
• Authors: Yu Fei, Yasaman Razeghi, Sameer Singh,
• Abstract summary: We propose nudging, a plug-and-play algorithm that aligns any base model at inference time using a small aligned model. Nudging is motivated by recent findings that alignment primarily alters the model's behavior on a small subset of stylistic tokens. We evaluate the effectiveness of nudging across 3 model families and 13 tasks, covering reasoning, general knowledge, instruction following, and safety benchmarks.
• Score: 18.530367090350605
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large language models (LLMs) require alignment, such as instruction-tuning or reinforcement learning from human feedback, to effectively and safely follow user instructions. This process necessitates training aligned versions for every model size in each model family, resulting in significant computational overhead. In this work, we propose nudging, a simple, plug-and-play, and training-free algorithm that aligns any base model at inference time using a small aligned model. Nudging is motivated by recent findings that alignment primarily alters the model's behavior on a small subset of stylistic tokens, such as "Sure" or "Thank". We find that base models are significantly more uncertain when generating these tokens. Leveraging this observation, nudging employs a small aligned model to generate nudging tokens to steer the large base model's output toward desired directions when the base model's uncertainty is high. We evaluate the effectiveness of nudging across 3 model families and 13 tasks, covering reasoning, general knowledge, instruction following, and safety benchmarks. Without any additional training, nudging a large base model with a 7x - 14x smaller aligned model achieves zero-shot performance comparable to, and sometimes surpassing, that of large aligned models. For example, nudging OLMo-7b with OLMo-1b-instruct, affecting less than 9% of tokens, achieves a 10% absolute improvement on GSM8K over OLMo-7b-instruct. Unlike prior inference-time tuning methods, nudging enables off-the-shelf collaboration between model families. For instance, nudging Gemma-2-27b with Llama-2-7b-chat outperforms Llama-2-70b-chat on various tasks. Overall, this work introduces a simple yet powerful approach to token-level model collaboration, offering a modular solution to LLM alignment. Our project website: https://fywalter.github.io/nudging/ .

Related papers

• Cross-model Control: Improving Multiple Large Language Models in One-time Training [34.98931804630706]
  Cross-model Control (CMC) is a method that improves multiple large language models in one-time training. Based on this insight, we incorporate a tiny language model with a minimal number of parameters. We propose a novel token mapping strategy named PM-MinED to make this tiny language model applicable to models with different vocabularies.
  arXiv  Detail & Related papers  (2024-10-23T06:52:09Z)
• What Matters for Model Merging at Scale? [94.26607564817786]
  Model merging aims to combine multiple expert models into a more capable single model. Previous studies have primarily focused on merging a few small models. This study systematically evaluates the utility of model merging at scale.
  arXiv  Detail & Related papers  (2024-10-04T17:17:19Z)
• Enabling Small Models for Zero-Shot Classification through Model Label Learning [50.68074833512999]
  We introduce a novel paradigm, Model Label Learning (MLL), which bridges the gap between models and their functionalities. Experiments on seven real-world datasets validate the effectiveness and efficiency of MLL.
  arXiv  Detail & Related papers  (2024-08-21T09:08:26Z)
• EMR-Merging: Tuning-Free High-Performance Model Merging [55.03509900949149]
  We show that Elect, Mask & Rescale-Merging (EMR-Merging) shows outstanding performance compared to existing merging methods. EMR-Merging is tuning-free, thus requiring no data availability or any additional training while showing impressive performance.
  arXiv  Detail & Related papers  (2024-05-23T05:25:45Z)
• First-Step Advantage: Importance of Starting Right in Multi-Step Math Reasoning [11.75364271481855]
  Language models can solve complex reasoning tasks better by learning to generate rationales for their predictions. We observe that smaller models in particular when corrected, can solve a task that they would have otherwise struggled with. We propose QuestCoT, where a smaller model first asks itself how to start, before proceeding with a chain of reasoning.
  arXiv  Detail & Related papers  (2023-11-14T06:45:31Z)
• Zephyr: Direct Distillation of LM Alignment [59.03530095974505]
  We aim to produce a smaller language model that is aligned to user intent. Previous research has shown that applying supervised fine-tuning (dSFT) on larger models significantly improves task accuracy. We apply distilled direct preference optimization (dDPO) to learn a chat model with significantly improved intent alignment.
  arXiv  Detail & Related papers  (2023-10-25T19:25:16Z)
• "Medium" LMs of Code in the Era of LLMs: Lessons From StackOverflow [5.036273913335737]
  We train two models: SOBertBase, with 109M parameters, and SOBertLarge with 762M parameters, at a budget of just $$187$ and $$800$ each. Results demonstrate that pre-training both extensively and properly on in-domain data can yield a powerful and affordable alternative to leveraging closed-source general-purpose models.
  arXiv  Detail & Related papers  (2023-06-05T21:38:30Z)
• eP-ALM: Efficient Perceptual Augmentation of Language Models [70.47962271121389]
  We propose to direct effort to efficient adaptations of existing models, and propose to augment Language Models with perception. Existing approaches for adapting pretrained models for vision-language tasks still rely on several key components that hinder their efficiency. We show that by freezing more than 99% of total parameters, training only one linear projection layer, and prepending only one trainable token, our approach (dubbed eP-ALM) significantly outperforms other baselines on VQA and Captioning.
  arXiv  Detail & Related papers  (2023-03-20T19:20:34Z)
• Deep Model Assembling [31.88606253639418]
  This paper studies a divide-and-conquer strategy to train large models. It divides a large model into smaller modules, training them independently, and reassembling the trained modules to obtain the target model. We introduce a global, shared meta model to implicitly link all the modules together. This enables us to train highly compatible modules that collaborate effectively when they are assembled together.
  arXiv  Detail & Related papers  (2022-12-08T08:04:06Z)
• CAMERO: Consistency Regularized Ensemble of Perturbed Language Models with Weight Sharing [83.63107444454938]
  We propose a consistency-regularized ensemble learning approach based on perturbed models, named CAMERO. Specifically, we share the weights of bottom layers across all models and apply different perturbations to the hidden representations for different models, which can effectively promote the model diversity. Our experiments using large language models demonstrate that CAMERO significantly improves the generalization performance of the ensemble model.
  arXiv  Detail & Related papers  (2022-04-13T19:54:51Z)

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.