AdarshSingh7647/Naive-SFT-Table-8B
AdarshSingh7647/Naive-SFT-Table-8B is an 8 billion parameter Qwen3-based model, fine-tuned for single-call listwise table reranking with a 32768 token context length. It is trained using Naive-SFT, directly outputting a ranked list of tables without intermediate reasoning. This model excels at efficiently identifying and ordering relevant tables for a given question, making it suitable for applications requiring fast and accurate table retrieval.
Loading preview...
Naive-SFT-Table-8B Overview
AdarshSingh7647/Naive-SFT-Table-8B is an 8 billion parameter language model built upon the Qwen3 architecture, specifically designed for single-call listwise table reranking. This model's key differentiator is its "Naive-SFT" training approach, which enables it to directly output a ranked list of candidate tables for a given question in a single generation, bypassing the need for explicit reasoning traces. It supports a substantial context length of 32768 tokens, allowing for the processing of multiple tables and complex queries.
Key Capabilities
- Efficient Table Reranking: Ranks candidate tables from most to least useful for a given question in a single inference call.
- Direct Output: Generates a JSON object containing the ranked, one-indexed positions of tables without intermediate reasoning steps.
- Qwen3-8B Base: Leverages the capabilities of the Qwen3-8B model for robust language understanding.
- High Context Length: Processes inputs up to 32768 tokens, accommodating multiple detailed tables.
Good For
- Information Retrieval Systems: Enhancing search engines or knowledge bases that rely on structured data in tables.
- Question Answering over Tables: Quickly identifying the most relevant tables to answer a user's question.
- Data Extraction Workflows: Prioritizing tables for further processing based on their relevance to a query.
- Applications Requiring Fast Table Selection: Use cases where rapid and accurate table relevance assessment is critical.