With the rapid adoption of Large Language Models (LLMs), LLM-adapters have become increasingly common, providing lightweight specialization of large-scale models. Serving hundreds or thousands of these adapters on a single GPU allows request aggregation, increasing throughput, but may also cause request starvation if GPU memory limits are exceeded. To address this issue, this study focuses on determining the joint configuration of concurrent and parallel adapters that maximizes GPU throughput without inducing starvation, given heterogeneous adapter and traffic properties. We propose a data-driven ML approach leveraging interpretable models to tackle this caching problem and introduce the first Digital Twin capable of reproducing an LLM-adapter serving system, enabling efficient training data generation. Experiments with the vLLM framework and LoRA adapters show that the Digital Twin reproduces throughput within 5.1% of real results, while the ML approach predicts optimal numbers of concurrent and parallel adapters with an error of at most 7.2% under heterogeneous, real-world workloads. The code is publicly available at https://github.com/FerranAgulloLopez/GPULLMAdapterOptimization.

As the large language models (LLMs) grow in size each day, efficient training and fine-tuning has never been as important as nowadays. This resulted in the great interest in parameter efficient fine-tuning (PEFT), and effective methods including low-rank adapters (LoRA) has emerged. Although the various PEFT methods have been studied extensively in the recent years, the greater part of the subject remains unexplored with the huge degree of freedom. In this paper, we propose FLoRA, a family of fused forward-backward adapters (FFBA) for parameter-efficient fine-tuning of LLMs on downstream tasks. The FFBA combine ideas from the popular LoRA and parallel adapters to improve the overall fine-tuning accuracies. At the same time, latencies are minimized by fusing the forward and backward adapters into existing projection layers of the base model. Experimental results show that the proposed FFB adapters perform significantly better than the popularly used LoRA in both accuracy and latency for a similar parameter budget.

The Segment Anything Model (SAM) has demonstrated impressive generalization in prompt-based segmentation. Yet, the potential of semantic text prompts remains underexplored compared to traditional spatial prompts like points and boxes. This paper introduces SAM-PTx, a parameter-efficient approach for adapting SAM using frozen CLIP-derived text embeddings as class-level semantic guidance. Specifically, we propose a lightweight adapter design called Parallel-Text that injects text embeddings into SAM's image encoder, enabling semantics-guided segmentation while keeping most of the original architecture frozen. Our adapter modifies only the MLP-parallel branch of each transformer block, preserving the attention pathway for spatial reasoning. Through supervised experiments and ablations on the COD10K dataset as well as low-data subsets of COCO and ADE20K, we show that incorporating fixed text embeddings as input improves segmentation performance over purely spatial prompt baselines. To our knowledge, this is the first work to use text prompts for segmentation on the COD10K dataset. These results suggest that integrating semantic conditioning into SAM's architecture offers a practical and scalable path for efficient adaptation with minimal computational complexity.

Recent studies show that pretrained vision models can boost performance in audio downstream tasks. To enhance the performance further, an additional pretraining stage with large-scale audio data is typically required to infuse audio-specific knowledge into the vision model. However, such approaches require extensive audio data and a carefully designed objective function. In this work, we propose bypassing the pretraining stage by directly fine-tuning the vision model with our Look-Aside Adapter (LoAA) designed for efficient audio understanding. Audio spectrum data is represented across two heterogeneous dimensions--time and frequency--and we refine adapters to facilitate interactions between tokens across these dimensions. Our experiments demonstrate that our adapters allow vision Figure 1: An illustration of our simplified approach for audio models to reach or surpass the performance of pretrained audio classification. Our newly proposed Parameter Efficient Fine-models in various audio and speech tasks, offering a resourceefficient Tuning (PEFT) paradigm for audio classification is a direct and effective solution for leveraging vision models in adaptation to downstream tasks in a singular stage.

Parameter-Efficient Fine-Tuning (PEFT) methods have gained significant popularity for adapting pre-trained Large Language Models (LLMs) to downstream tasks, primarily due to their potential to significantly reduce memory and computational overheads. However, a common limitation in most PEFT approaches is their application of a uniform architectural design across all layers. This uniformity involves identical trainable modules and ignores the varying importance of each layer, leading to sub-optimal fine-tuning results. To overcome the above limitation and obtain better performance, we develop a novel approach, Importance-aware Sparse Tuning (IST), to fully utilize the inherent sparsity and select the most important subset of full layers with effective layer-wise importance scoring. The proposed IST is a versatile and plug-and-play technique compatible with various PEFT methods that operate on a per-layer basis. By leveraging the estimated importance scores, IST dynamically updates these selected layers in PEFT modules, leading to reduced memory demands. We further provide theoretical proof of convergence and empirical evidence of superior performance to demonstrate the advantages of IST over uniform updating strategies. Extensive experiments on a range of LLMs, PEFTs, and downstream tasks substantiate the effectiveness of our proposed method, showcasing IST's capacity to enhance existing layer-based PEFT methods. Our code is available at https://github.com/Kaiseem/IST.

Parameter-efficient transfer learning (PETL) aims to adapt pre-trained models to new downstream tasks while minimizing the number of fine-tuned parameters. Adapters, a popular approach in PETL, inject additional capacity into existing networks by incorporating low-rank projections, achieving performance comparable to full fine-tuning with significantly fewer parameters. This paper investigates the relationship between the placement of an adapter and its performance. We observe that adapter location within a network significantly impacts its effectiveness, and that the optimal placement is task-dependent. To exploit this observation, we introduce an extended search space of adapter connections, including long-range and recurrent adapters. We demonstrate that even randomly selected adapter placements from this expanded space yield improved results, and that high-performing placements often correlate with high gradient rank. Our findings reveal that a small number of strategically placed adapters can match or exceed the performance of the common baseline of adding adapters in every block, opening a new avenue for research into optimal adapter placement strategies.

Parameter-Efficient Fine-tuning (PEFT) facilitates the fine-tuning of Large Language Models (LLMs) under limited resources. However, the fine-tuning performance with PEFT on complex, knowledge-intensive tasks is limited due to the constrained model capacity, which originates from the limited number of additional trainable parameters. To overcome this limitation, we introduce a novel mechanism that fine-tunes LLMs with adapters of larger size yet memory-efficient. This is achieved by leveraging the inherent activation sparsity in the Feed-Forward Networks (FFNs) of LLMs and utilizing the larger capacity of Central Processing Unit (CPU) memory compared to Graphics Processing Unit (GPU). We store and update the parameters of larger adapters on the CPU. Moreover, we employ a Mixture of Experts (MoE)-like architecture to mitigate unnecessary CPU computations and reduce the communication volume between the GPU and CPU. This is particularly beneficial over the limited bandwidth of PCI Express (PCIe). Our method can achieve fine-tuning results comparable to those obtained with larger memory capacities, even when operating under more limited resources such as a 24GB memory single GPU setup, with acceptable loss in training efficiency. Our codes are available at https://github.com/CURRENTF/MEFT.

We propose Conditional Adapter (CoDA), a parameter-efficient transfer learning method that also improves inference efficiency. CoDA generalizes beyond standard adapter approaches to enable a new way of balancing speed and accuracy using conditional computation. Starting with an existing dense pretrained model, CoDA adds sparse activation together with a small number of new parameters and a light-weight training phase. Our experiments demonstrate that the CoDA approach provides an unexpectedly efficient way to transfer knowledge. Across a variety of language, vision, and speech tasks, CoDA achieves a 2x to 8x inference speed-up compared to the state-of-the-art Adapter approaches with moderate to no accuracy loss and the same parameter efficiency.

A practical limitation of deep neural networks is their high degree of specialization to a single task and visual domain. Recently, inspired by the successes of transfer learning, several authors have proposed to learn instead universal, fixed feature extractors that, used as the first stage of any deep network, work well for several tasks and domains simultaneously. Nevertheless, such universal features are still somewhat inferior to specialized networks. To overcome this limitation, in this paper we propose to consider instead universal parametric families of neural networks, which still contain specialized problem-specific models, but differing only by a small number of parameters. We study different designs for such parametrizations, including series and parallel residual adapters, joint adapter compression, and parameter allocations, and empirically identify the ones that yield the highest compression. We show that, in order to maximize performance, it is necessary to adapt both shallow and deep layers of a deep network, but the required changes are very small. We also show that these universal parametrization are very effective for transfer learning, where they outperform traditional fine-tuning techniques.