causal-memory

BCMT引入了一个分块因果记忆Transformer架构，它在长上下文语言建模中解耦了局部和全局依赖，在保持与密集Transformer相当的性能的同时提高了效率。

This paper introduces MERIT, a training-free agent that uses causal episodic memory of past repair outcomes to improve subsequent Text-to-SQL generations, boosting execution accuracy on Spider and BIRD benchmarks.