Guohao Dai

Cache-to-Cache: Direct Semantic Communication Between Large Language Models

Think-at-Hard: Selective Latent Iterations to Improve Reasoning Language Models

PM-KVQ: Progressive Mixed-precision KV Cache Quantization for Long-CoT LLMs

SpecPrune-VLA: Accelerating Vision-Language-Action Models via Action-Aware Self-Speculative Pruning

ReXMoE: Reusing Experts with Minimal Overhead in Mixture-of-Experts

SentKVCompress: Sentence-Level Dynamic KVCache Compression for Efficient Long-Context LLM Inference

R2R: Efficiently Navigating Divergent Reasoning Paths with Small-Large Model Token Routing

Distilled Decoding 2: One-step Sampling of Image Auto-regressive Models with Conditional Score Distillation

Mixture of Attention Spans: Optimizing LLM Inference Efficiency with Heterogeneous Sliding-Window Lengths

DeepGate4: Efficient and Effective Representation Learning for Circuit Design at Scale

LV-Eval: A Balanced Long-Context Benchmark with 5 Length Levels Up to 256K

ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generation

Linear Combination of Saved Checkpoints Makes Consistency and Diffusion Models Better

Accelerating Auto-regressive Text-to-Image Generation with Training-free Speculative Jacobi Decoding

LV-Eval: A Balanced Long-Context Benchmark with 5 Length Levels Up to 256K

MoA: Mixture of Sparse Attention for Automatic Large Language Model Compression

Efficient Expert Pruning for Sparse Mixture-of-Experts Language Models: Enhancing Performance and Reducing Inference Costs