Saved in:
Bibliographic Details
Main Authors: Zhang, Bowen, Wang, Meiyi, Soh, Harold
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2602.06665
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912884562329600
author Zhang, Bowen
Wang, Meiyi
Soh, Harold
author_facet Zhang, Bowen
Wang, Meiyi
Soh, Harold
contents Post-training improves instruction-following and helpfulness of large language models (LLMs) but often reduces generation diversity, which leads to repetitive outputs in open-ended settings, a phenomenon known as mode collapse. Motivated by evidence that LLM layers play distinct functional roles, we hypothesize that mode collapse can be localized to specific layers and that restoring a carefully chosen range of layers to their pre-trained weights can recover diversity while maintaining high output quality. To validate this hypothesis and decide which layers to restore, we design a proxy task -- Constrained Random Character(CRC) -- with an explicit validity set and a natural diversity objective. Results on CRC reveal a clear diversity-validity trade-off across restoration ranges and identify configurations that increase diversity with minimal quality loss. Based on these findings, we propose Selective Layer Restoration (SLR), a training-free method that restores selected layers in a post-trained model to their pre-trained weights, yielding a hybrid model with the same architecture and parameter count, incurring no additional inference cost. Across three different tasks (creative writing, open-ended question answering, and multi-step reasoning) and three different model families (Llama, Qwen, and Gemma), we find SLR can consistently and substantially improve output diversity while maintaining high output quality.
format Preprint
id arxiv_https___arxiv_org_abs_2602_06665
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Not All Layers Need Tuning: Selective Layer Restoration Recovers Diversity
Zhang, Bowen
Wang, Meiyi
Soh, Harold
Computation and Language
Artificial Intelligence
Post-training improves instruction-following and helpfulness of large language models (LLMs) but often reduces generation diversity, which leads to repetitive outputs in open-ended settings, a phenomenon known as mode collapse. Motivated by evidence that LLM layers play distinct functional roles, we hypothesize that mode collapse can be localized to specific layers and that restoring a carefully chosen range of layers to their pre-trained weights can recover diversity while maintaining high output quality. To validate this hypothesis and decide which layers to restore, we design a proxy task -- Constrained Random Character(CRC) -- with an explicit validity set and a natural diversity objective. Results on CRC reveal a clear diversity-validity trade-off across restoration ranges and identify configurations that increase diversity with minimal quality loss. Based on these findings, we propose Selective Layer Restoration (SLR), a training-free method that restores selected layers in a post-trained model to their pre-trained weights, yielding a hybrid model with the same architecture and parameter count, incurring no additional inference cost. Across three different tasks (creative writing, open-ended question answering, and multi-step reasoning) and three different model families (Llama, Qwen, and Gemma), we find SLR can consistently and substantially improve output diversity while maintaining high output quality.
title Not All Layers Need Tuning: Selective Layer Restoration Recovers Diversity
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2602.06665