PreFT: Prefill-only finetuning for efficient inference

• URL: http://arxiv.org/abs/2605.14217v1
• Date: Thu, 14 May 2026 00:19:41 GMT
• Title: PreFT: Prefill-only finetuning for efficient inference
• Abstract summary: We develop and release an efficient implementation of two prefill-only PEFTs, LoRA and ReFT, on the vLLM inference engine.<n>We show that serving multi-user PreFTs is more efficient than traditional PEFTs ($1.9times$ the throughput when serving $512$ adapters on Llama 3.1 70B)
• Score: 54.58291801311547
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models can now be personalised efficiently at scale using parameter efficient finetuning methods (PEFTs), but serving user-specific PEFTs harms throughput, even with specialised kernels and memory management techniques. This is because, theoretically and empirically, a mismatch exists between prefill (processing a large number of tokens at once) and decode (generating a single token autoregressively): the latter has far lower throughput when serving multiple adapters. Rather than optimising performance relative to parameter count, for efficient multi-adapter serving, we instead ought to optimise performance relative to serving throughput. We therefore propose PreFT (Prefill-only Finetuning), wherein we only apply the adapter to prefill tokens and discard it afterwards. PreFT significantly increases throughput with minimal effect on performance. We develop and release an efficient implementation of two prefill-only PEFTs, LoRA and ReFT, on the vLLM inference engine. We first show that serving multi-user PreFTs is more efficient than traditional PEFTs ($1.9\times$ the throughput when serving $512$ adapters on Llama 3.1 70B). Then, we compare the performance of prefill-only vs. all-token adapters on a variety of supervised finetuning and reinforcement learning tasks with LMs at varying scales. On SFT, we observe that the evaluation loss of PreFTs is higher than PEFTs, but can be compensated by increasing rank with nearly no reduction in throughput. On RL, we consistently find that PreFTs approach parity with standard PEFTs. Together, this work validates prefill-only adaptation of LLMs as a more favourable accuracy-throughput tradeoff than existing PEFTs for personalised serving.

Related papers

• Preserving Pre-trained Representation Space: On Effectiveness of Prefix-tuning for Large Multi-modal Models [24.62337386603331]
  Large Multi-modal Models (LMMs) are revolutionizing the way machines interact with the world. To adapt LMMs for downstream tasks, parameter-efficient fine-tuning (PEFT) has gained popularity. This paper focuses on the strengths and weaknesses of each tuning strategy, shifting the focus from the efficiency typically associated with these approaches.
  arXiv  Detail & Related papers  (2024-10-29T07:55:50Z)
• Lessons and Insights from a Unifying Study of Parameter-Efficient Fine-Tuning (PEFT) in Visual Recognition [36.031972728327894]
  We conduct a unifying empirical study of representative PEFT methods with Vision Transformers.<n>We find that different PEFT methods achieve similar accuracy in the low-shot benchmark VTAB-1K.<n>Despite similar accuracy, we find that PEFT methods make different mistakes and high-confidence predictions, likely due to their different inductive biases.
  arXiv  Detail & Related papers  (2024-09-24T19:57:40Z)
• Forecast-PEFT: Parameter-Efficient Fine-Tuning for Pre-trained Motion Forecasting Models [68.23649978697027]
  Forecast-PEFT is a fine-tuning strategy that freezes the majority of the model's parameters, focusing adjustments on newly introduced prompts and adapters. Our experiments show that Forecast-PEFT outperforms traditional full fine-tuning methods in motion prediction tasks. Forecast-FT further improves prediction performance, evidencing up to a 9.6% enhancement over conventional baseline methods.
  arXiv  Detail & Related papers  (2024-07-28T19:18:59Z)
• IISAN: Efficiently Adapting Multimodal Representation for Sequential Recommendation with Decoupled PEFT [45.95678408227546]
  IISAN (Intra- and Inter-modal Side Adapted Network for Multimodal Representation) is a plug-and-play architecture for sequential recommender systems. IISAN matches the performance of full fine-tuning (FFT) and state-of-the-art PEFT. It significantly reduces GPU memory usage - from 47GB to just 3GB for multimodal sequential recommendation tasks.
  arXiv  Detail & Related papers  (2024-04-02T15:58:36Z)
• Dynamic Tuning Towards Parameter and Inference Efficiency for ViT Adaptation [67.13876021157887]
  Dynamic Tuning (DyT) is a novel approach to improve both parameter and inference efficiency for ViT adaptation. DyT achieves superior performance compared to existing PEFT methods while evoking only 71% of their FLOPs on the VTAB-1K benchmark.
  arXiv  Detail & Related papers  (2024-03-18T14:05:52Z)
• LoRETTA: Low-Rank Economic Tensor-Train Adaptation for Ultra-Low-Parameter Fine-Tuning of Large Language Models [20.5908375260123]
  Various parameter-efficient fine-tuning (PEFT) techniques have been proposed to enable computationally efficient fine-tuning while maintaining model performance. We present LoRETTA, a framework that significantly reduces trainable parameters through tensor-train decomposition. LoRETTA achieves comparable or better performance than most widely used PEFT methods with up to $100times$ fewer parameters on the LLaMA-2-7B models.
  arXiv  Detail & Related papers  (2024-02-18T01:20:00Z)
• From PEFT to DEFT: Parameter Efficient Finetuning for Reducing Activation Density in Transformers [52.199303258423306]
  We propose a novel density loss that encourages higher activation sparsity in pre-trained models. Our proposed method, textbfDEFT, can consistently reduce activation density by up to textbf44.94% on RoBERTa$_mathrmLarge$ and by textbf53.19% (encoder density) and textbf90.60% (decoder density) on Flan-T5$_mathrmXXL$.
  arXiv  Detail & Related papers  (2024-02-02T21:25:46Z)
• WeGeFT: Weight-Generative Fine-Tuning for Multi-Faceted Efficient Adaptation of Large Models [8.481707805559589]
  We propose Weight-Generative Fine-Tuning (WeGeFT), a novel approach that learns to generate fine-tuning weights directly from the pretrained weights.<n>This design achieves multi-faceted efficiency in parameters, representations, compute, and memory, while maintaining or exceeding the performance of LoRA and its variants.
  arXiv  Detail & Related papers  (2023-12-01T16:33:57Z)
• LoRAPrune: Structured Pruning Meets Low-Rank Parameter-Efficient Fine-Tuning [56.88751562302793]
  Low-rank adaption (LoRA) has emerged to fine-tune large language models (LLMs) LoRAPrune is a new framework that delivers an accurate structured pruned model in a highly memory-efficient manner. LoRAPrune achieves a reduction in perplexity by 4.81 on WikiText2 and 3.46 on PTB, while also decreasing memory usage by 52.6%.
  arXiv  Detail & Related papers  (2023-05-28T15:15:48Z)
• Sensitivity-Aware Visual Parameter-Efficient Fine-Tuning [91.5113227694443]
  We propose a novel visual. sensuous-aware fine-Tuning (SPT) scheme. SPT allocates trainable parameters to task-specific important positions. Experiments on a wide range of downstream recognition tasks show that our SPT is complementary to the existing PEFT methods.
  arXiv  Detail & Related papers  (2023-03-15T12:34:24Z)

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.