Showing cs.LGShow all
2 papers · 1 filter
cs.LG2026
Beyond Parallel Blindness: Information Floors and Model Gaps in Block Drafting
Xinwei Qiang, Xiang Fang, Chang Chen +3
Block drafters propose several tokens in one forward pass, before earlier target tokens are realised. Their rejection mixes two losses: missing within-block path information and im…
cs.LG2026
DASH: Deterministic Attention Scheduling for High-throughput Reproducible LLM Training
Xinwei Qiang, Hongmin Chen, Shixuan Sun +3
Determinism is indispensable for reproducibility in large language model (LLM) training, yet it often exacts a steep performance cost. In widely used attention implementations such…