Windowed-MTP: Removing the Full-Context Draft-KV Tax at Million-Token Context 25 July 2026·3 mins Machine Learning AI Infrastructure Paper Notes Machine-Learning Arxiv Inference Speculative-Decoding Long-Context Llm-Systems
You Only Index Once: Cross-Layer Sparse Attention with Shared Routing 6 June 2026·6 mins Machine Learning Paper Notes Machine-Learning Nlp Arxiv Llm Long-Context Inference Sparse-Attention