On the Limits of Layer Pruning for Generative Reasoning in Large Language Models

arXiv:2602.01997v3 Announce Type: replace Abstract: Recent work has shown that layer pruning can effectively compress large language models (LLMs) while retaining strong performance on classification benchmarks, often with little or no finetuning. In contrast, generative reasoning tasks, such as GSM8K and HumanEval\textsuperscript{+}, exhibit substantially weaker recovery. We show that beyond surface-level text degradation, pruning leads to a loss of key algorithmic capabilities, including arithmetic computation and balanced parenthesis generation. Under realistic post-training constraints, using a single 80GB GPU and without access to pretraining-scale data or compute, we evaluate a simple recovery strategy based on supervised finetuning with self-generated responses. This approach recovers up to 90\% of baseline performance on classification tasks, but recovery for generative reasoning remains limited. We further find that this gap persists even under a favorable task-aligned recovery setting, where pruned models are fully finetuned on self-generated GSM8K responses, suggesting that the degradation is not merely due to generic instruction data or parameter-efficient tuning. As complementary evidence, we analyze a depth-pruned model trained with nearly 100B post-pruning tokens and find that deficits persist even on simple arithmetic tasks that do not require multi-step generation. Overall, we characterize practical recovery limits of layer pruning for generative reasoning and provide guidance on when depth reduction is effective under constrained post-training regimes.

Abstract:Recent work has shown that layer pruning can effectively compress large language models (LLMs) while retaining strong performance on classification benchmarks, often with little or no finetuning. In contrast, generative reasoning tasks, such as GSM8K and HumanEval\textsuperscript{+}, exhibit substantially weaker recovery. We show that beyond surface-level text degradation, pruning leads to a loss of key algorithmic capabilities, including arithmetic computation and balanced parenthesis generation. Under realistic post-training constraints, using a single 80GB GPU and without access to pretraining-scale data or compute, we evaluate a simple recovery strategy based on supervised finetuning with self-generated responses. This approach recovers up to 90\% of baseline performance on classification tasks, but recovery for generative reasoning remains limited. We further find that this gap persists even under a favorable task-aligned recovery setting, where pruned models are fully finetuned on self-generated GSM8K responses, suggesting that the degradation is not merely due to generic instruction data or parameter-efficient tuning. As complementary evidence, we analyze a depth-pruned model trained with nearly 100B post-pruning tokens and find that deficits persist even on simple arithmetic tasks that do not require multi-step generation. Overall, we characterize practical recovery limits of layer pruning for generative reasoning and provide guidance on when depth reduction is effective under constrained post-training regimes.

Machine Learning (cs.LG); Artificial Intelligence (cs.AI)

arXiv:2602.01997 [cs.LG]

(or arXiv:2602.01997v3 [cs.LG] for this version)

https://doi.org/10.48550/arXiv.2602.01997 Focus to learn more arXiv-issued DOI via DataCite

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

analiza aiqbench

On the Limits of Layer Pruning for Generative Reasoning in Large Language Models

Subiecte conexe