PPL
Hao Lin
Papers (6)
Domino: Decoupling Causal Modeling from Autoregressive Drafting in Speculative DecodingShare More, Search Less: Collaborative Parallel Thinking for Efficient Test-Time ScalingFocused Forcing: Content-Aware Per-Frame KV Selection for Efficient Autoregressive Video DiffusionDynaTrain: Fast Online Parallelism Switching for Elastic LLM TrainingVISD: Enhancing Video Reasoning via Structured Self-DistillationKnowing but Not Correcting: Routine Task Requests Suppress Factual Correction in LLMs