1 paper · 1 filter
Anupama Sridhar, Alexander Johansen
Long chain-of-thought reasoning and agentic tool-calling produce traces spanning tens of thousands of tokens, yet Transformer KV caches grow linearly with sequence length, creating…