Recursive Language Models

LLMs drop details as context grows and handle window-filling documents poorly. Recursive Language Models (RLMs) are an inference paradigm where the LM programmatically examines, decomposes and recursively calls itself over its input, handling near-infinite context.

This plug-and-play library supports various sandboxes and replaces plain completion calls.

Features

Recursion:Divide and conquer.

Long context:Near-infinite.

Plug and play:Replace completions.

Sandboxes:Safe execution.