Tianle Cai

In-Place Test-Time Training

FutureX: An Advanced Live Benchmark for LLM Agents in Future Prediction

Training-Free Activation Sparsity in Large Language Models

SVDQuant: Absorbing Outliers by Low-Rank Component for 4-Bit Diffusion Models

Radial Attention: $\mathcal O(n \log n)$ Sparse Attention for Long Video Generation

MATH-Perturb: Benchmarking LLMs' Math Reasoning Abilities against Hard Perturbations

CommVQ: Commutative Vector Quantization for KV Cache Compression

Unnatural Languages Are Not Bugs but Features for LLMs