Shallow Prefill, Deep Decoding: Efficient Long-Context Inference via Layer-Asymmetric KV Visibility
Paper • 2605.06105 • Published • 1
jeongseokoh
jeongseokoh
·
AI & ML interests
Large Language Models, Efficient LLM, Trustworthy AI
Recent Activity
updated a collection about 15 hours ago
SPEED submitted a paper about 15 hours ago
Shallow Prefill, Deep Decoding: Efficient Long-Context Inference via Layer-Asymmetric KV Visibility updated a collection 7 days ago
SPEED Downstream Task ModelsOrganizations
submitted a
paper to Daily Papers about 15 hours ago