AdarshSingh7647/TabRankMultiTableNaive
AdarshSingh7647/TabRankMultiTableNaive 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 answering a given question by processing all candidates in a single prompt and emitting a full ranking in one generation. It is specifically trained on a mix of NQ Tables and MultiTabQA, excelling at multi-table retrieval by directly outputting a ranked list without explicit reasoning traces for faster inference.
Loading preview...
TabRankMultiTableNaive: Multi-Table Retrieval Reranker
AdarshSingh7647/TabRankMultiTableNaive is an 8B parameter model based on Qwen3, specifically engineered as a generative listwise reranker for table retrieval. Its primary function is to take a question and a set of candidate tables, then output a ranked list of tables from most to least relevant for answering the question.
Key Capabilities & Features
- Single-Call Generative Reranking: Processes all candidate tables within a single prompt and generates the complete ranking in one inference step.
- Optimized for Speed: This variant is trained to directly output the ranking, with reasoning traces shown during training but masked from the loss, making it the fastest for inference.
- Multi-Table Retrieval: Trained on a combination of NQ Tables and MultiTabQA datasets, specifically incorporating multi-table retrieval questions.
- JSON Output: Returns a JSON object containing the ranked list of one-based candidate positions.
Performance Highlights
The model demonstrates strong performance across various table question answering benchmarks, reranking a first-stage top 25 candidate list. Key metrics include:
- Mean recall@10: 0.815
- Mean ndcg@10: 0.695
- Mean acc@10: 0.680
- Mean MRR: 0.731
When to Use This Model
This model is ideal for applications requiring efficient and accurate listwise reranking of candidate tables for question answering, particularly when dealing with scenarios that involve multiple tables. Its fast inference speed makes it suitable for real-time or high-throughput retrieval systems.