Context-1 Data Generation

Context-1 Data Generation

Synthetic multi-hop search task generation

Description

Training models for deep retrieval needs many multi-hop tasks of search, verify, search again, far more than humans can write. This is Chroma's synthetic data pipeline that generates multi-hop search tasks across domains.

Each domain follows explore, verify, extend to train the Context-1 retrieval model, with weights released.

Features



Multi-hop:Step-by-step retrieval.

Domains:Broad coverage.

Verification:Quality checked.

Open weights:Context-1.