single source · post-training
Long-context question-answering RL environment
A long-context question-answering RL environment trains a model to answer questions over long documents.1 Generated multi-option questions must combine facts from several passages, and are filtered by whether strong and weak checkpoints can solve them before entering training.2
How this definition was written
Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.
I.1.2 Long Context Environment
— tech reportA question has to combine facts from several passages
— tech report
Also called Long-Context Question Answering, Long Context Environment.
Filed under post-training › agentic post-training