REPL-based Long-context Reasoning
Reproducing and extending Recursive Language Models (RLM) — offloading long contexts into a REPL — plus an iterative (ILM) and runtime-assisted DP-ILM variant. Evaluated on OOLONG, LongBench v2, and CodeQA with DeepSeek v3.2 and Kimi K2.
🔒 WIP