Recursive Language Models: Enabling Unbounded Context Processing Through Self-Recursion

We propose Recursive Language Models (RLMs), an inference strategy where language models can decompose and recursively interact with input context of unbounded length through REPL environments.