Xing Li

TrimR: Verifier-based Training-Free Thinking Trimming for Efficient Test-Time Scaling

Scaling Up, Speeding Up: A Benchmark of Speculative Decoding for Efficient LLM Test-Time Scaling

Beyond Speedup - Utilizing KV Cache for Sampling and Reasoning

SeedThink: Test-Time Control via Seed-Thought Initialization

Why Attention Patterns Exist: A Unifying Temporal Perspective Analysis

Accurate KV Cache Eviction via Anchor Direction Projection for Efficient LLM Inference

Circuit Transformer: A Transformer That Preserves Logical Equivalence

AttentionPredictor: Temporal Patterns Matter for KV Cache Compression

KVTuner: Sensitivity-Aware Layer-Wise Mixed-Precision KV Cache Quantization for Efficient and Nearly Lossless LLM Inference

SeaDAG: Semi-autoregressive Diffusion for Conditional Directed Acyclic Graph Generation