MemoryBench: A Benchmark for Memory and Continual Learning in LLM Systems
arXiv:2510.17281v5 Announce Type: replace-cross
Abstract: Scaling up data, parameters, and test-time computation has been the mainstream methods to improve LLM systems (LLMsys), but their upper bounds are almost reached due to the gradual depletion of…