CLAIFeb 6

Not All Layers Need Tuning: Selective Layer Restoration Recovers Diversity

arXiv:2602.06665v1h-index: 4
Originality Incremental advance
AI Analysis

This addresses the issue of repetitive outputs in open-ended settings for users of LLMs, offering a training-free method to enhance diversity, though it is incremental as it builds on existing layer-based insights.

The paper tackled the problem of mode collapse in post-trained large language models, where generation diversity is reduced, and found that restoring specific layers to pre-trained weights can recover diversity with minimal quality loss, achieving consistent improvements across multiple tasks and model families.

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.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes