POP: Prefill-Only Pruning for Efficient Large Model Inference

• URL: http://arxiv.org/abs/2602.03295v1
• Date: Tue, 03 Feb 2026 09:22:26 GMT
• Title: POP: Prefill-Only Pruning for Efficient Large Model Inference
• Authors: Junhui He, Zhihui Fu, Jun Wang, Qingan Li,
• Abstract summary: Large Language Models (LLMs) and Vision-Language Models (VLMs) have demonstrated remarkable capabilities.<n>Existing structured pruning methods, while hardware-efficient, often suffer from significant accuracy degradation.<n>We argue that this failure stems from a stage-agnostic pruning approach that overlooks the asymmetric roles between the prefill and decode stages.
• Score: 5.743318651374061
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Language Models (LLMs) and Vision-Language Models (VLMs) have demonstrated remarkable capabilities. However, their deployment is hindered by significant computational costs. Existing structured pruning methods, while hardware-efficient, often suffer from significant accuracy degradation. In this paper, we argue that this failure stems from a stage-agnostic pruning approach that overlooks the asymmetric roles between the prefill and decode stages. By introducing a virtual gate mechanism, our importance analysis reveals that deep layers are critical for next-token prediction (decode) but largely redundant for context encoding (prefill). Leveraging this insight, we propose Prefill-Only Pruning (POP), a stage-aware inference strategy that safely omits deep layers during the computationally intensive prefill stage while retaining the full model for the sensitive decode stage. To enable the transition between stages, we introduce independent Key-Value (KV) projections to maintain cache integrity, and a boundary handling strategy to ensure the accuracy of the first generated token. Extensive experiments on Llama-3.1, Qwen3-VL, and Gemma-3 across diverse modalities demonstrate that POP achieves up to 1.37$\times$ speedup in prefill latency with minimal performance loss, effectively overcoming the accuracy-efficiency trade-off limitations of existing structured pruning methods.

Related papers

• POP: Online Structural Pruning Enables Efficient Inference of Large Foundation Models [12.10403234534641]
  POP (Partition-guided Online Pruning) is an efficient online structural pruning framework with minimal computational overhead.<n> POP is a lightweight, plug-and-play method that requires no preprocessing, including offline calibration, retraining, or learning predictors.
  arXiv  Detail & Related papers  (2026-02-06T16:07:42Z)
• Up to 36x Speedup: Mask-based Parallel Inference Paradigm for Key Information Extraction in MLLMs [22.76757502541604]
  We introduce PIP: a Parallel Inference Paradigm for Key Information Extraction (KIE)<n>Our approach reformulates the problem by using "[mask]" tokens as placeholders for all target values, enabling their simultaneous generation in a single forward pass.<n> Experimental results show that our PIP-models achieve a 5-36x inference speedup with negligible performance degradation compared to traditional autoregressive base models.
  arXiv  Detail & Related papers  (2026-01-27T13:45:30Z)
• Fast KVzip: Efficient and Accurate LLM Inference with Gated KV Eviction [50.99402504483692]
  We propose a novel gating-based KV cache eviction method for frozen-weight language models.<n>Our approach integrates seamlessly into both the prefill and decoding stages.<n>Experiments show that our method maintains near-lossless performance while evicting up to 70% of the KV cache.
  arXiv  Detail & Related papers  (2026-01-25T03:07:54Z)
• Smooth Operator: Smooth Verifiable Reward Activates Spatial Reasoning Ability of Vision-Language Model [18.526821056010384]
  Vision-Language Models (VLMs) face a critical bottleneck in achieving precise numerical prediction for 3D scene understanding.<n>Traditional reinforcement learning approaches, primarily based on relative ranking, often suffer from severe reward sparsity and gradient instability.<n>We introduce the Smooth Numerical Reward Activation (SNRA) operator and the Absolute-Preserving GRPO framework.
  arXiv  Detail & Related papers  (2026-01-12T16:26:42Z)
• Training-free Context-adaptive Attention for Efficient Long Context Modeling [57.703159205740185]
  Training-free Context-adaptive Attention (TCA-Attention) is a training-free sparse attention mechanism that selectively attends to only the informative tokens for efficient long-context inference.<n>TCA-Attention achieves a 2.8$times$ speedup and reduces KV cache by 61% at 128K context length while maintaining performance comparable to full attention.
  arXiv  Detail & Related papers  (2025-12-10T01:54:57Z)
• SpecAttn: Speculating Sparse Attention [1.6921396880325779]
  We introduce SpecAttn, a novel training-free approach that seamlessly integrates with speculative decoding techniques.<n>Our key insight is to exploit the attention weights already computed by the draft model during speculative decoding to identify important tokens for the target model.<n>SpecAttn achieves over 75% reduction in key-value cache accesses with a mere 15.29% increase in perplexity on the PG-19 dataset.
  arXiv  Detail & Related papers  (2025-10-31T17:12:34Z)
• READER: Retrieval-Assisted Drafter for Efficient LLM Inference [0.0386965802948046]
  Autoregressive Language Models instantiate a factorized likelihood over token sequences, yet their strictly sequential decoding process imposes an intrinsic lower bound on latency inference.<n>This bottleneck has emerged as a central obstacle to the scalable deployment of large-scale generative models.<n>We present READER, a speculative decoding framework that bypasses the training of the auxiliary draft model.
  arXiv  Detail & Related papers  (2025-08-12T16:47:48Z)
• Pushing the Limits of Low-Bit Optimizers: A Focus on EMA Dynamics [64.62231094774211]
  Statefuls (e.g., Adam) maintain auxiliary information even 2x the model size in order to achieve optimal convergence.<n>SOLO enables Adam-styles to maintain quantized states with precision as low as 3 bits, or even 2 bits.<n>SOLO can thus be seamlessly applied to Adam-styles, leading to substantial memory savings with minimal accuracy loss.
  arXiv  Detail & Related papers  (2025-05-01T06:47:45Z)
• PrefixKV: Adaptive Prefix KV Cache is What Vision Instruction-Following Models Need for Efficient Generation [97.41972925670508]
  Large vision-language models (LVLMs) incur significant computational and memory overhead during inference.<n>We present PrefixKV, where "Prefix" means the top-ranked KV based on importance rather than position in the original sequence.<n>Our method achieves the state-of-the-art performance compared with others.
  arXiv  Detail & Related papers  (2024-12-04T15:48:59Z)
• ALOcc: Adaptive Lifting-Based 3D Semantic Occupancy and Cost Volume-Based Flow Predictions [91.55655961014027]
  3D semantic occupancy and flow prediction are fundamental to understanding scene scene.<n>This paper proposes a vision-based framework with three targeted improvements.<n>Our purely convolutional architecture establishes new SOTA performance on multiple benchmarks for both semantic occupancy and joint semantic-flow prediction.
  arXiv  Detail & Related papers  (2024-11-12T11:32:56Z)
• ZipVL: Efficient Large Vision-Language Models with Dynamic Token Sparsification [29.163757099307553]
  The efficiency of large vision-language models (LVLMs) is constrained by the computational bottleneck of the attention mechanism during the prefill phase.<n>We present ZipVL, an efficient inference framework designed for LVLMs through a dynamic ratio allocation strategy of important tokens.
  arXiv  Detail & Related papers  (2024-10-11T07:24:21Z)
• Sparse is Enough in Fine-tuning Pre-trained Large Language Models [98.46493578509039]
  We propose a gradient-based sparse fine-tuning algorithm, named Sparse Increment Fine-Tuning (SIFT) We validate its effectiveness on a range of tasks including the GLUE Benchmark and Instruction-tuning.
  arXiv  Detail & Related papers  (2023-12-19T06:06:30Z)
• Confident Adaptive Language Modeling [95.45272377648773]
  CALM is a framework for dynamically allocating different amounts of compute per input and generation timestep. We demonstrate the efficacy of our framework in reducing compute -- potential speedup of up to $times 3$ -- while provably maintaining high performance.
  arXiv  Detail & Related papers  (2022-07-14T17:00:19Z)
• BERT Loses Patience: Fast and Robust Inference with Early Exit [91.26199404912019]
  We propose Patience-based Early Exit as a plug-and-play technique to improve the efficiency and robustness of a pretrained language model. Our approach improves inference efficiency as it allows the model to make a prediction with fewer layers.
  arXiv  Detail & Related papers  (2020-06-07T13:38:32Z)

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.