blogvLLM Blog·23 Oct 20265.2
vLLM fits long agentic contexts for GLM-5.3 on 16 H200s by splitting prefill and decode nodes and using prefill and decode context parallelism plus hybrid sparse attention.
paperarXiv (systems, data, ML)·22 Oct 20264.2
Characterizes how LLMs keep repeating instructions while ceasing to obey them in long dialogues.
paperEmergent Mind·25 Oct 20264.2
Kronecker Delta Rule treats key-value pairs as joint addresses in linear attention to reduce interference at long context.
paperarXiv (systems, data, ML)·22 Oct 20264.1
Evidence position bias differs across attention and other sequence mixers in long-context QA.
paperHugging Face Daily Papers·20 Oct 20263.9
Stratified sparse attention narrows the gap to dense attention at extreme sparsity for long contexts.
paperACM Digital Library·22 Oct 20263.8
HyCo2 compresses long context with global semantics and local evidence.