Jinpeng LV

Lan Li, Jinpeng Lv

Semantic search, a process aimed at delivering highly relevant search results by comprehending the searcher's intent and the contextual meaning of terms within a searchable dataspace, plays a pivotal role in information retrieval. In this paper, we investigate the application of large language models to enhance semantic search capabilities, specifically tailored for the domain of Jupyter Notebooks. Our objective is to retrieve generated outputs, such as figures or tables, associated functions and methods, and other pertinent information. We demonstrate a semantic search framework that achieves a comprehensive semantic understanding of the entire notebook's contents, enabling it to effectively handle various types of user queries. Key components of this framework include: 1). A data preprocessor is designed to handle diverse types of cells within Jupyter Notebooks, encompassing both markdown and code cells. 2). An innovative methodology is devised to address token size limitations that arise with code-type cells. We implement a finer-grained approach to data input, transitioning from the cell level to the function level, effectively resolving these issues.

Youpeng Zhao, Jinpeng LV, Di Wu et al.

Test-time scaling (TTS) has recently emerged as a promising direction to exploit the hidden reasoning capabilities of pre-trained large language models (LLMs). However, existing scaling methods narrowly focus on the compute-optimal Pareto-frontier, ignoring the simple fact that compute-optimal is not always system-optimal. In this work, we propose a system-driven perspective on TTS, analyzing how reasoning models scale against practical metrics, such as latency and cost-per-token. By evaluating the impact of popular optimizations such as tensor parallelism and speculative decoding, our preliminary analysis reveals the limitations of current methods and calls for a paradigm shift toward holistic, system-aware evaluations that capture the true essence of scaling laws at inference time.