MemoryBench: A Standardized Benchmark for Evaluating LLM Memory and Continual Learning

MemoryBench is a dataset created by THUIR to provide a standardized and extensible benchmark for evaluating memory and continual learning in large language model systems. The dataset aims to encourage future work toward more adaptive, feedback-driven, and efficient LLM systems. It was last updated on December 23, 2025.

Use Cases

• Benchmarking LLM memory performance based on the dataset's standardized evaluation tasks
• Researching continual learning algorithms based on the benchmark's extensible framework
• Developing adaptive and feedback-driven LLM systems based on the dataset's stated purpose

Strengths

• Designed as a standardized benchmark, which suggests a consistent evaluation methodology
• Aims to be extensible, allowing for future additions and adaptations
• Last updated on December我们发现 23, 2025, indicating recent maintenance

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• Description metadata is limited; actual data quality requires manual inspection after download

Provenance

Source

THUIR

Collection Method

Likely constructed for benchmarking purposes, as described.

Time Range

null

Freshness

Last updated 2025-12-23 16:32:21

Geography

null