LoRA-Drop: Temporal LoRA Decoding for Efficient LLM Inference

• URL: http://arxiv.org/abs/2601.02569v1
• Date: Mon, 05 Jan 2026 21:47:47 GMT
• Title: LoRA-Drop: Temporal LoRA Decoding for Efficient LLM Inference
• Abstract summary: We present textbfLoRA-Drop, a plug-and-play inference framework that accelerates decoding by applying a emphtemporal compute schedule to a fixed subset of intermediate layers.<n>LoRA-Drop requires no routing network, is compatible with standard KV caching, and can reduce KV-cache footprint by skipping KV updates in droppable layers during LoRA steps and refreshing periodically.
• Score: 4.771191010592548
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Autoregressive large language models (LLMs) are bottlenecked by sequential decoding, where each new token typically requires executing all transformer layers. Existing dynamic-depth and layer-skipping methods reduce this cost, but often rely on auxiliary routing mechanisms or incur accuracy degradation when bypassed layers are left uncompensated. We present \textbf{LoRA-Drop}, a plug-and-play inference framework that accelerates decoding by applying a \emph{temporal compute schedule} to a fixed subset of intermediate layers: on most decoding steps, selected layers reuse the previous-token hidden state and apply a low-rank LoRA correction, while periodic \emph{refresh} steps execute the full model to prevent drift. LoRA-Drop requires no routing network, is compatible with standard KV caching, and can reduce KV-cache footprint by skipping KV updates in droppable layers during LoRA steps and refreshing periodically. Across \textbf{LLaMA2-7B}, \textbf{LLaMA3-8B}, \textbf{Qwen2.5-7B}, and \textbf{Qwen2.5-14B}, LoRA-Drop achieves up to \textbf{2.6$\times$ faster decoding} and \textbf{45--55\% KV-cache reduction} while staying within \textbf{0.5 percentage points (pp)} of baseline accuracy. Evaluations on reasoning (GSM8K, MATH, BBH), code generation (HumanEval, MBPP), and long-context/multilingual benchmarks (LongBench, XNLI, XCOPA) identify a consistent \emph{safe zone} of scheduling configurations that preserves quality while delivering substantial efficiency gains, providing a simple path toward adaptive-capacity inference in LLMs. Codes are available at https://github.com/hosseinbv/LoRA-Drop.git.

Related papers

• You Only Index Once: Cross-Layer Sparse Attention with Shared Routing [61.29627714699688]
  Cross-layer sparse attention (A) is built on top of KV-sharing architectures such as YOCO.<n>A single indexer computes token-level top-k selection once and reuses the resulting index across layers.<n>Results show CLSA is both accurate and efficient, achieving up to 7.6x decoding speedup and 17.1x overall throughput improvement at 128K context.
  arXiv  Detail & Related papers  (2026-06-04T17:54:04Z)
• Continual Fine-Tuning of Large Language Models via Program Memory [31.698386161730486]
  Low-Rank Adaptation (LoRA) has become a standard approach for adapting Large Language Models (LLMs) under limited compute.<n>We propose a continual LoRA framework with textbfProgram memory, inspired by textbfComplementary textbfLearning Systems in neuroscience.
  arXiv  Detail & Related papers  (2026-05-13T08:26:37Z)
• Multi-Head Low-Rank Attention [22.28455391125486]
  Multi-Head Latent Attention (MLA) significantly reduces the total KV cache size.<n>Since its single latent head cannot be partitioned, each device is forced to redundantly load the complete KV cache for every token.<n>We propose Multi-Head Low-Rank Attention (MLRA) which enables partitionable latent states for efficient 4-way TP decoding.
  arXiv  Detail & Related papers  (2026-03-02T18:52:38Z)
• RelayCaching: Accelerating LLM Collaboration via Decoding KV Cache Reuse [5.597099794399441]
  RelayCaching is a training-free inference method that directly reuses decoding phase KV caches from previous agents.<n>We show that RelayCaching achieves over 80% KV cache reuse, reduces TTFT by up to $4.7times$ compared to the standard pipeline.
  arXiv  Detail & Related papers  (2026-02-28T04:46:28Z)
• Beyond SGD, Without SVD: Proximal Subspace Iteration LoRA with Diagonal Fractional K-FAC [50.36542772932594]
  Low-Rank Adaptation (LoRA) fine-tunes large models by learning low-rank updates on top of frozen weights.<n>In this work, we address the gap between training with full steps with low-rank projections (SVDLoRA) and LoRA fine-tuning.<n>We propose LoRSum, a memory-efficient subroutine that closes this gap for gradient descent.
  arXiv  Detail & Related papers  (2026-02-18T13:41:41Z)
• Attention Is All You Need for KV Cache in Diffusion LLMs [36.94369617373333]
  Elastic-Cache performs adaptive, layer-aware cache updates for diffusion large language models.<n>Our method achieves significantly higher throughput ($6.8times$ on GSM8K) than existing confidence-based approaches.
  arXiv  Detail & Related papers  (2025-10-16T17:59:48Z)
• Faster Than SVD, Smarter Than SGD: The OPLoRA Alternating Update [50.36542772932594]
  Low-Rank Adaptation (LoRA) fine-tunes large models by learning low-rank updates on top of frozen weights.<n>There is still a gap between full training with low-rank projections (SVDLoRA) and LoRA fine-tuning, indicating that LoRA steps can be further improved.
  arXiv  Detail & Related papers  (2025-09-24T10:32:50Z)
• Drag-and-Drop LLMs: Zero-Shot Prompt-to-Weights [75.83625828306839]
  textbfDrag-and-Drop LLMs (textitDnD) eliminates per-task training by mapping a handful of unlabeled task prompts directly to LoRA weight updates.<n>A lightweight text encoder distills each prompt batch into condition embeddings, which are then transformed by a cascaded hyper-convolutional decoder into the full set of LoRA matrices.
  arXiv  Detail & Related papers  (2025-06-19T15:38:21Z)
• On-the-Fly Adaptive Distillation of Transformer to Dual-State Linear Attention [53.22963042513293]
  Large language models (LLMs) excel at capturing global token dependencies via self-attention but face prohibitive compute and memory costs on lengthy inputs.<n>We first propose dual-state linear attention (A), a novel design that maintains two hidden states-one for preserving historical context and one for tracking recencythereby mitigating the short-range bias typical of linear-attention architectures.<n>We introduce DSLA-Serve, an online adaptive distillation framework that progressively replaces Transformer layers DSLA layers at inference time, guided by a sensitivity-based layer ordering.
  arXiv  Detail & Related papers  (2025-06-11T01:25:06Z)
• Rectified Sparse Attention [61.7702154360081]
  Efficient long-sequence generation is a critical challenge for Large Language Models.<n>We propose Rectified Sparse Attention (ReSA), a simple yet effective method that combines block-sparse attention with periodic dense rectification.<n> Experiments across math reasoning, language modeling, and retrieval tasks demonstrate that ReSA achieves near-lossless generation quality.
  arXiv  Detail & Related papers  (2025-06-04T16:01:48Z)
• QuantSpec: Self-Speculative Decoding with Hierarchical Quantized KV Cache [67.84112700032007]
  Large Language Models (LLMs) are increasingly being deployed on edge devices for long-context settings.<n>In these scenarios, the Key-Value ( KV) cache is the primary bottleneck in terms of both GPU memory and latency.<n>We propose a novel self-speculative decoding framework, QuantSpec, where the draft model shares the architecture of the target model but employs a hierarchical 4-bit quantized KV cache and 4-bit quantized weights for acceleration.
  arXiv  Detail & Related papers  (2025-02-05T20:43:48Z)
• EoRA: Fine-tuning-free Compensation for Compressed LLM with Eigenspace Low-Rank Approximation [84.70637613266835]
  EoRA is a fine-tuning-free method that augments compressed Large Language Models with low-rank matrices.<n>EoRA consistently outperforms prior training-free low rank methods in recovering the accuracy of compressed LLMs.
  arXiv  Detail & Related papers  (2024-10-28T17:59:03Z)

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.