Sssunset/

Sigma-Mem-Data

Training and evaluation streams for Sigma-Mem: An Online Reliability Memory for LLM-based Multi-Agent Systems. Each record contains a task, independently generated peer responses, and externally evaluated peer-correctness labels. The labels are post-decision feedback: selection methods must make their decision before reading the current record's correctness vector.

	Data Groups




	

Configuration Events per split Peers Purpose

mixed_train 2,963… See the full description on the dataset page: https://huggingface.co/datasets/Sssunset/Sigma-Mem-Data.