Hanfei Yu

Rui Wei, Hanfei Yu, Shubham Jain et al.

Reinforcement Learning from Human Feedback (RLHF) has been widely applied to Large Language Model (LLM) post-training to align model outputs with human preferences. Recent models, such as DeepSeek-R1, have also shown RLHF's potential to improve LLM reasoning on complex tasks. In RL, inference and training co-exist, creating dynamic resource demands throughout the workflow. Compared to traditional RL, RLHF further challenges training efficiency due to expanding model sizes and resource consumption. Several RLHF frameworks aim to balance flexible abstraction and efficient execution. However, they rely on serverful infrastructures, which struggle with fine-grained resource variability. As a result, during synchronous RLHF training, idle time between or within RL components often causes overhead and resource wastage. To address these issues, we present RLHFless, the first scalable training framework for synchronous RLHF, built on serverless computing environments. RLHFless adapts to dynamic resource demands throughout the RLHF pipeline, pre-computes shared prefixes to avoid repeated computation, and uses a cost-aware actor scaling strategy that accounts for response length variation to find sweet spots with lower cost and higher speed. In addition, RLHFless assigns workloads efficiently to reduce intra-function imbalance and idle time. Experiments on both physical testbeds and a large-scale simulated cluster show that RLHFless achieves up to 1.35x speedup and 44.8% cost reduction compared to the state-of-the-art baseline.

Rui Wei, Rui Du, Hanfei Yu et al.

In Large Language Model (LLM) inference, early-exit refers to stopping computation at an intermediate layer once the prediction is sufficiently confident, thereby reducing latency and cost. However, recent LLMs adopt improved pretraining recipes and architectures that reduce layer redundancy, potentially limiting early-exit opportunities. We re-evaluate layer-wise early-exit in modern LLMs and analyze how intermediate representations evolve during training. We introduce a metric to quantify a model's intrinsic suitability for early-exit and propose a benchmark for researchers to explore the potential early-exit benefits on different models and workloads. Our results show a diminishing trend in early-exit effectiveness across newer model generations. We further find that dense transformers generally offer greater early-exit potential than Mixture-of-Experts and State Space Models. In addition, larger models, particularly those with more than 20 billion parameters, and base pretrained models without specialized tuning tend to exhibit higher early-exit potential.

Hanfei Yu, Xingqi Cui, Hong Zhang et al.

Large Language Models (LLMs) have gained immense success in revolutionizing various applications, including content generation, search and recommendation, and AI-assisted operation. To reduce high training costs, Mixture-of-Experts (MoE) architecture has become a popular backbone for modern LLMs. However, despite the benefits, serving MoE-based LLMs experience severe memory inefficiency due to sparsely activated experts. Recent studies propose to offload inactive experts from GPU memory to CPU memory to improve the serving efficiency of MoE models. However, they either incur high inference latency or high model memory footprints due to coarse-grained designs. To tame the latency-memory trade-off in MoE serving, we present FineMoE, a fine-grained expert offloading system for MoE serving that achieves low inference latency with memory efficiency. We design FineMoE to extract fine-grained expert selection patterns from MoE models and semantic hints from input prompts to efficiently guide expert prefetching, caching, and offloading decisions. FineMoE is prototyped on top of HuggingFace Transformers and deployed on a six-GPU testbed. Experiments with open-source MoE models and real-world workloads show that FineMoE reduces inference latency by 47% and improves expert hit rate by 39% over state-of-the-art solutions.

Hanfei Yu, Bei Ouyang, Shwai He et al.

Large Language Models (LLMs) have become a cornerstone of AI, driving progress across diverse domains such as content creation, search and recommendation systems, and AI-assisted workflows. To alleviate extreme training costs and advancing model scales, Mixture-of-Experts (MoE) has become a popular backbone for modern LLMs, which are commonly served in distributed deployment using expert parallelism (EP). However, MoE's sparse activation mechanism leads to severe expert load imbalance, where a few experts become overloaded while others remain idle, resulting in expert stragglers that inflate inference latency and serving cost. Existing expert load balancing solutions assume static resource configurations on serverful infrastructures, limiting expert scalability and elasticity, and resulting in either costly real-time expert swapping or degraded generation quality. We present MoEless, the first serverless MoE serving framework that mitigates expert load imbalance and accelerates inference via serverless experts. MoEless employs lightweight, layer-aware predictors to accurately estimate incoming expert load distributions and proactively identify stragglers. We design optimized expert scaling and placement strategies to maximize function locality, improve GPU utilization, and balance loads across experts and GPUs. MoEless is prototyped on top of Megatron-LM and deployed on an eight-GPU testbed. Experiments with open-source MoE models and real-world workloads show that MoEless reduces inference latency by 43% and inference cost by 84% compared to state-of-the-art solutions.

Yifan Sui, Hao Wang, Hanfei Yu et al.

Serverless computing has grown rapidly for serving Large Language Model (LLM) inference due to its pay-as-you-go pricing, fine-grained GPU usage, and rapid scaling. However, our analysis reveals that current serverless can effectively serve general LLM but fail with Low-Rank Adaptation (LoRA) inference due to three key limitations: 1) massive parameter redundancy among functions where 99% of weights are unnecessarily duplicated, 2) costly artifact loading latency beyond LLM loading, and 3) magnified resource contention when serving multiple LoRA LLMs. These inefficiencies lead to massive GPU wastage, increased Time-To-First-Token (TTFT), and high monetary costs. We propose ServerlessLoRA, a novel serverless inference system designed for faster and cheaper LoRA LLM serving. ServerlessLoRA enables secure backbone LLM sharing across isolated LoRA functions to reduce redundancy. We design a pre-loading method that pre-loads comprehensive LoRA artifacts to minimize cold-start latency. Furthermore, ServerlessLoRA employs contention aware batching and offloading to mitigate GPU resource conflicts during bursty workloads. Experiment on industrial workloads demonstrates that ServerlessLoRA reduces TTFT by up to 86% and cuts monetary costs by up to 89% compared to state-of-the-art LLM inference solutions.

Hanfei Yu, Hao Wang, Jian Li et al.

Serverless computing automates fine-grained resource scaling and simplifies the development and deployment of online services with stateless functions. However, it is still non-trivial for users to allocate appropriate resources due to various function types, dependencies, and input sizes. Misconfiguration of resource allocations leaves functions either under-provisioned or over-provisioned and leads to continuous low resource utilization. This paper presents Freyr, a new resource manager (RM) for serverless platforms that maximizes resource efficiency by dynamically harvesting idle resources from over-provisioned functions to under-provisioned functions. Freyr monitors each function's resource utilization in real-time, detects over-provisioning and under-provisioning, and learns to harvest idle resources safely and accelerates functions efficiently by applying deep reinforcement learning algorithms along with a safeguard mechanism. We have implemented and deployed a Freyr prototype in a 13-node Apache OpenWhisk cluster. Experimental results show that 38.8% of function invocations have idle resources harvested by Freyr, and 39.2% of invocations are accelerated by the harvested resources. Freyr reduces the 99th-percentile function response latency by 32.1% compared to the baseline RMs.

Athirai A. Irissappane, Hanfei Yu, Yankun Shen et al.

Online reviews are a vital source of information when purchasing a service or a product. Opinion spammers manipulate these reviews, deliberately altering the overall perception of the service. Though there exists a corpus of online reviews, only a few have been labeled as spam or non-spam, making it difficult to train spam detection models. We propose an adversarial training mechanism leveraging the capabilities of Generative Pre-Training 2 (GPT-2) for classifying opinion spam with limited labeled data and a large set of unlabeled data. Experiments on TripAdvisor and YelpZip datasets show that the proposed model outperforms state-of-the-art techniques by at least 7% in terms of accuracy when labeled data is limited. The proposed model can also generate synthetic spam/non-spam reviews with reasonable perplexity, thereby, providing additional labeled data during training.