Ruoming Pang

Reusing Pre-Training Data at Test Time is a Compute Multiplier

MANZANO: A Simple and Scalable Unified Multimodal Model with a Hybrid Vision Tokenizer

Synthetic Bootstrapped Pretraining

Step-by-Step Reasoning for Math Problems via Twisted Sequential Monte Carlo

EC-DIT: Scaling Diffusion Transformers with Adaptive Expert-Choice Routing

Talking Turns: Benchmarking Audio Foundation Models on Turn-Taking Dynamics

Can External Validation Tools Improve Annotation Quality for LLM-as-a-Judge?

Instruction-Following Pruning for Large Language Models

Fantastic Experts and How to Find Them: A Multi-Dimensional Study for Experts-Level Sparsification in Mixture-of-Experts

Improve Vision Language Model Chain-of-thought Reasoning