Job Description

Limited access to large-scale computing infrastructure hinders the deployment of modern AI systems, particularly large language models (LLMs). To broaden adoption under such constraints, improving computational efficiency is essential. Leveraging sparsity is a promising approach, as many datasets and model components in applications like natural language processing, information retrieval, and recommender systems are inherently sparse. However, efficiently handling sparse data is challenging: it requires specialized data structures, and general-purpose libraries often deliver suboptimal performance without task-specific optimization.

To address this, we will analyze computational bottlenecks in key sparse operations—such as sparse matrix multiplication in feedforward networks and masked attention—and develop optimized methods for core tasks including clustering, classification, and LLM inference.
 

Preferred Intern Educational Level

Any level is fine

Skill sets or Qualities

Passion on doing deep research