Feifan Wang

Bowen Pang, Kai Li, Ruifeng She et al.

With the development of large language models (LLMs), it has become increasingly important to optimize hardware usage and improve throughput. In this paper, we study the inference optimization of the serving system that deploys LLMs. To optimize system throughput and maximize hardware utilization, we formulate the inference optimization problem as a mixed-integer programming (MIP) model and propose a hybrid offline-online method as solution. The offline method improves large-scale inference systems by introducing a Minimizing Makespan Bin Packing Problem. We further provide a theoretical lower bound computation method. Then, we propose an online sorting and preemptive scheduling method to better utilize hardware. In the online iteration scheduling process, a Lagrangian method is applied to evaluate the cost efficiency of inserting prefill stages versus decode stages at each iteration and dynamically determine when to preempt decoding tasks and insert prefill tasks. Experiments using real-world data from the LLaMA-65B model and the GSM8K dataset demonstrate that system utilization improves from 80.2% to 89.1%, and the total inference time decreases from 201.00 to 190.58 seconds. A 100-cases study shows that our method consistently outperforms the baseline method and improves the utilization rate by 8.0% on average. Finally, we discuss potential future extensions, including stochastic modeling, reinforcement learning-based schedulers, and dynamic decision-making strategies for system throughput and hardware utilization.

Feifan Wang, Tengfei Song, Minggui He et al.

Facial emotion perception in the vision large language model (VLLM) is crucial for achieving natural human-machine interaction. However, creating high-quality annotations for both coarse- and fine-grained facial emotion analysis demands costly expertise. The lack of such high-quality instruction data limits the performance of VLLMs in facial emotion perception. To address this, we propose a self-verification approach with emotion knowledge enhancement (SEKE), which generates high-quality instruction data for multi-grained emotion analysis cost-effectively using closed-source VLLM. This approach integrates prior human knowledge to VLLM inference, guided by the inherent correlations between three grained levels of emotion descriptions, i.e., discrete expression, valence-arousal, and action unit, to reliably generate comprehensive annotations. A self-verification strategy with Uncertainty-Aware Monte Carlo sampling (SV-UAMC) is further embedded to efficiently extract more accurate VLLM predictions, further improving annotation reliability. Consequently, we construct a facial emotion instruction dataset (FEID) containing three comprehensive descriptions, which provides coarse- and fine-grained emotional information for effective model training. Additionally, we introduce a facial emotion analysis benchmark (FEAB) to measure the VLLM's corresponding ability. Our method significantly outperforms state-of-the-art methods on three downstream facial emotion analysis tasks.