Tianchi Zhang

Shengkun Tang, Yaqing Wang, Zhenglun Kong et al. · baidu, harvard

Large-scale Transformer models bring significant improvements for various downstream vision language tasks with a unified architecture. The performance improvements come with increasing model size, resulting in slow inference speed and increased cost for severing. While some certain predictions benefit from the full complexity of the large-scale model, not all of inputs need the same amount of computation to conduct, potentially leading to computation resource waste. To handle this challenge, early exiting is proposed to adaptively allocate computational power in term of input complexity to improve inference efficiency. The existing early exiting strategies usually adopt output confidence based on intermediate layers as a proxy of input complexity to incur the decision of skipping following layers. However, such strategies cannot apply to encoder in the widely-used unified architecture with both encoder and decoder due to difficulty of output confidence estimation in the encoder. It is suboptimal in term of saving computation power to ignore the early exiting in encoder component. To handle this challenge, we propose a novel early exiting strategy for unified visual language models, which allows dynamically skip the layers in encoder and decoder simultaneously in term of input layer-wise similarities with multiple times of early exiting, namely \textbf{MuE}. By decomposing the image and text modalities in the encoder, MuE is flexible and can skip different layers in term of modalities, advancing the inference efficiency while minimizing performance drop. Experiments on the SNLI-VE and MS COCO datasets show that the proposed approach MuE can reduce expected inference time by up to 50\% and 40\% while maintaining 99\% and 96\% performance respectively.

Tianchi Zhang

Traffic forecasting is a significant part of intelligent transportation systems. One of the critical challenges of traffic forecasting is to find spatio-temporal correlations. In recent years, graph convolutional networks and graph attention networks have replaced traditional statistical models to predict future traffic. However, it is complicated for both of them to allow vertices to have far different characters. To address this, we propose the Global-Local Graph Attention Network (GLGAT) with pairwise encoding and the event-based adjacency matrix. The GLGAT allows vertices to have a global attention matrix set for the whole graph and assigns local attention matrix sets to each vertex. Experiments on two real-world traffic datasets show that GLGAT can effectively capture spatio-temporal correlations and has competitive performance against other state-of-the-art baselines.

Long Phan, Alice Gatti, Ziwen Han et al. · amazon-science, apple-ml

Benchmarks are important tools for tracking the rapid advancements in large language model (LLM) capabilities. However, benchmarks are not keeping pace in difficulty: LLMs now achieve over 90\% accuracy on popular benchmarks like MMLU, limiting informed measurement of state-of-the-art LLM capabilities. In response, we introduce Humanity's Last Exam (HLE), a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage. HLE consists of 2,500 questions across dozens of subjects, including mathematics, humanities, and the natural sciences. HLE is developed globally by subject-matter experts and consists of multiple-choice and short-answer questions suitable for automated grading. Each question has a known solution that is unambiguous and easily verifiable, but cannot be quickly answered via internet retrieval. State-of-the-art LLMs demonstrate low accuracy and calibration on HLE, highlighting a significant gap between current LLM capabilities and the expert human frontier on closed-ended academic questions. To inform research and policymaking upon a clear understanding of model capabilities, we publicly release HLE at https://lastexam.ai.

Ruiqi Xu, Tianchi Zhang

Although the computing power of mobile devices is increasing, machine learning models are also growing in size. This trend creates problems for mobile devices due to limitations like their memory capacity and battery life. While many services, like ChatGPT and Midjourney, run all the inferences in the cloud, we believe a flexible and fine-grained task distribution is more desirable. In this work, we consider model segmentation as a solution to improving the user experience, dividing the computation between mobile devices and the cloud in a way that offloads the compute-heavy portion of the model while minimizing the data transfer required. We show that the division not only reduces the wait time for users but can also be fine-tuned to optimize the workloads of the cloud. To achieve that, we design a scheduler that collects information about network quality, client device capability, and job requirements, making decisions to achieve consistent performance across a range of devices while reducing the work the cloud needs to perform.

Jianwei Li, Tianchi Zhang, Ian En-Hsu Yen et al.

Transformer-based models, such as BERT, have been widely applied in a wide range of natural language processing tasks. However, one inevitable side effect is that they require massive memory storage and inference cost when deployed in production. Quantization is one of the popularized ways to alleviate the cost. However, the previous 8-bit quantization strategy based on INT8 data format either suffers from the degradation of accuracy in a Post-Training Quantization (PTQ) fashion or requires an expensive Quantization-Aware Training (QAT) process. Recently, a new numeric format FP8 (i.e. floating-point of 8-bits) has been proposed and supported in commercial AI computing platforms such as H100. In this paper, we empirically validate the effectiveness of FP8 as a way to do Post-Training Quantization without significant loss of accuracy, with a simple calibration and format conversion process. We adopt the FP8 standard proposed by NVIDIA Corp. (2022) in our extensive experiments of BERT variants on GLUE and SQuAD v1.1 datasets, and show that PTQ with FP8 can significantly improve the accuracy upon that with INT8, to the extent of the full-precision model.