The authors introduce GRAB, a constructor-encoder-bridge pipeline designed for table question answering that lifts relational data into a heterogeneous graph and encodes it via message passing. The method transfers signals to a frozen large language model through a small set of query-conditioned latent tokens, providing a compact structural representation while preserving the LLM's general reasoning capabilities.
- GRAB utilizes a lightweight graph encoder and latent bridge with only 91M parameters, allowing efficient training without updating the LLM weights.
- The approach significantly improves performance on relational question answering tasks, achieving the largest gains in demanding multi-table settings.
- The pipeline offers an efficient and principled method for connecting relational deep learning with large language models.