Xinyin Ma

8

Papers

422

Total Citations

Papers (8)

DeepCache: Accelerating Diffusion Models for Free

dKV-Cache: The Cache for Diffusion Language Models

Thinkless: LLM Learns When to Think

Collaborative Decoding Makes Visual Auto-Regressive Modeling Efficient

VeriThinker: Learning to Verify Makes Reasoning Model Efficient

NeurIPS 2025arXiv

TinyFusion: Diffusion Transformers Learned Shallow

Auto-Controlled Image Perception in MLLMs via Visual Perception Tokens

Diffusion Model is Effectively Its Own Teacher