Transformer-based pre-trained models have revolutionized NLP for superior performance and generality. Fine-tuning pre-trained models for downstream tasks often requires private data, for which federated learning is the de-facto approach (i.e., FedNLP). However, our measurements show that FedNLP is prohibitively slow due to the large model sizes and the resultant high network/computation cost. Towards practical FedNLP, we identify as the key building blocks adapters, small bottleneck modules inserted at a variety of model layers. A key challenge is to properly configure the depth and width of adapters, to which the training speed and efficiency is highly sensitive. No silver-bullet configuration exists: the optimal choice varies across downstream NLP tasks, desired model accuracy, and mobile resources. To automate adapter configuration, we propose AdaFL¹, a framework that enhances the existing FedNLP with two novel designs. First, AdaFL progressively upgrades the adapter configuration throughout a training session; the principle is to quickly learn shallow knowledge by only training fewer and smaller adapters at the model's top layers, and incrementally learn deep knowledge by incorporating deeper and larger adapters. Second, AdaFL continuously profiles future adapter configurations by allocating participant devices to trial groups. Extensive experiments show that AdaFL can reduce FedNLP's model convergence delay to no more than several hours, which is up to 155.5× faster compared to vanilla FedNLP and 48× faster compared to strong baselines.

Generative tasks, such as text generation and question answering, are essential for mobile applications. Given their inherent privacy sensitivity, executing them on devices is demanded. Nowadays, the execution of these generative tasks heavily relies on the Large Language Models (LLMs). However, the scarce device memory severely hinders the scalability of these models. We present EdgeLLM, an efficient on-device LLM inference system for models whose sizes exceed the device's memory capacity. EdgeLLM is built atop speculative decoding, which delegates most tokens to a smaller, memory-resident (draft) LLM. EdgeLLM integrates three novel techniques: (1) Instead of generating a fixed width and depth token tree, EdgeLLM proposes compute-efficient branch navigation and verification to pace the progress of different branches according to their accepted probability to prevent the wasteful allocation of computing resources to the wrong branch and to verify them all at once efficiently. (2) It uses a self-adaptive fallback strategy that promptly initiates the verification process when the smaller LLM generates an incorrect token. (3) To not block the generation, EdgeLLM proposes speculatively generating tokens during large LLM verification with the compute-IO pipeline. Through extensive experiments, EdgeLLM exhibits impressive token generation speed which is up to 9.3× faster than existing engines.

Large language models (LLMs) such as GPTs and Mixtral-8x7B have revolutionized machine intelligence due to their exceptional abilities in generic ML tasks. Transiting LLMs from datacenters to edge devices brings benefits like better privacy and availability, but is challenged by their massive parameter size and thus unbearable runtime costs. To this end, we present EdgeMoE, an on-device inference engine for mixture-of-expert (MoE) LLMs – a popular form of sparse LLM that scales its parameter size with almost constant computing complexity. EdgeMoE achieves both memory- and compute-efficiency by partitioning the model into the storage hierarchy: non-expert weights are held in device memory; while expert weights are held on external storage and fetched to memory only when activated. This design is motivated by a key observation that expert weights are bulky but infrequently used due to sparse activation. To further reduce the expert I/O swapping overhead, EdgeMoE incorporates two novel techniques: (1) expert-wise bitwidth adaptation that reduces the expert sizes with tolerable accuracy loss; (2) expert preloading that predicts the activated experts ahead of time and preloads it with the compute-I/O pipeline. On popular MoE LLMs and edge devices, EdgeMoE showcase significant memory savings and speedup over competitive baselines.

On-device inference for Large Language Models (LLMs), driven by increasing privacy concerns and advancements of mobile-sized models, has gained significant interest. However, even mobile-sized LLMs (e.g., Gemma-2B) encounter unacceptably high inference latency, often bottlenecked by the prefill stage in tasks like screen UI understanding. We present llm.npu, the first LLM inference system utilizing on-device Neural Processing Unit (NPU) offloading to reduce prefill latency. llm.npu enhances NPU offloading efficiency by re-constructing the prompt and model in three levels: (1) At prompt level, it divides variable-length prompts into multiple fixed-sized chunks while maintaining data dependencies; (2) At tensor level, it identifies and extracts significant outliers to run on the CPU/GPU in parallel with minimal overhead; (3) At block level, it schedules Transformer blocks in an out-of-order manner to the CPU/GPU and NPU based on their hardware affinity and sensitivity to accuracy. Compared to competitive baselines, llm.npu achieves 22.4x faster prefill speed and 30.7x energy savings on average, and up to 32.8x speedup in an end-to-end real-world application. For the first time, llm.npu achieves more than 1,000 tokens/sec prefilling for a billion-sized model.

Large foundation models, including large language models, vision transformers, diffusion, and large language model based multimodal models, are revolutionizing the entire machine learning lifecycle, from training to deployment. However, the substantial advancements in versatility and performance these models offer come at a significant cost in terms of hardware resources. To support the growth of these large models in a scalable and environmentally sustainable way, there has been a considerable focus on developing resource-efficient strategies. This survey delves into the critical importance of such research, examining both algorithmic and systemic aspects. It offers a comprehensive analysis and valuable insights gleaned from existing literature, encompassing a broad array of topics from cutting-edge model architectures and training/serving algorithms to practical system designs and implementations. The goal of this survey is to provide an overarching understanding of how current approaches are tackling the resource challenges posed by large foundation models and to potentially inspire future breakthroughs in this field.

Large constellations of Low Earth Orbit (LEO) satellites have been launched for Earth observation and satellite-ground communication, which collect massive imagery and sensor data. These data can enhance the AI capabilities of satellites to address global challenges such as real-time disaster navigation and mitigation. Prior studies proposed leveraging federated learning (FL) across satellite-ground to collaboratively train a share machine learning (ML) model in a privacy-preserving mechanism. However, they mostly focus on single unique challenges such as limited ground-to-satellite bandwidth, short connection window, and long connection cycle, while ignoring the completeness of these challenges in deploying efficient FL frameworks in space. In this paper, we propose an efficient satellite-ground FL framework, SatelliteFL, to address these three challenges collectively. Its key idea is to ensure that each satellite must complete per-round training within each connection window. Moreover, we design a progressive block-wise quantization algorithm that determines a unique bitwidth for each block of the ML model to maximize the model utility while not exceeding the connection window. We evaluate SatelliteFL by plugging an implemented FL platform into real-world satellite networks and satellite images. The results show that SatelliteFL highly accelerates the convergence by up to 2.8× and improves the bandwidth utilization ratio by up to 9.3× compared to the state-of-the-art methods.