AdarshSingh7647/TabRankMultiTableCoTCond
AdarshSingh7647/TabRankMultiTableCoTCond is an 8 billion parameter generative listwise reranker built on Qwen3, designed for table retrieval tasks. This model efficiently reorders candidate tables from most to least useful for a given question by processing all candidates in a single prompt and emitting a full ranking. It is specifically trained with a reasoning hint to condition its output, making it a strong performer for multi-table reasoning and retrieval.
Loading preview...
TabRankMultiTableCoTCond: Generative Listwise Table Reranker
This model, developed by AdarshSingh7647, is an 8 billion parameter generative listwise reranker based on the Qwen3 architecture. Its primary function is to take a question and a set of candidate tables, then return those tables ranked by their relevance to the question, from most to least useful. A key differentiator is its ability to process all candidate tables within a single prompt and generate the complete ranking in one output.
Key Capabilities & Features
- Single-Call Listwise Reranking: Ranks multiple tables simultaneously in one generation pass.
- Reasoning Conditioned: Trained with a short reasoning hint in the prompt, allowing it to learn and produce rankings conditioned on this guidance, enhancing performance.
- Multi-Table Retrieval: Specifically trained on a mix of NQ Tables and MultiTabQA datasets, which includes questions requiring multi-table reasoning.
- Efficient Inference: Designed to be fast at inference while maintaining high accuracy.
- JSON Output: Provides the ranked list of one-based candidate positions in a structured JSON format.
Performance Highlights
Evaluated as a listwise reranker on reordering a top 25 candidate list across four table question answering benchmarks (SQA, TAT QA, HybridQA, TabFact), the model demonstrates strong performance with a mean recall@10 of 0.820 and a mean MRR of 0.701.
When to Use This Model
This model is ideal for use cases requiring efficient and accurate ranking of multiple candidate tables for question answering or information retrieval systems. Its reasoning-conditioned approach makes it particularly suitable for complex queries involving multi-table reasoning.