AdarshSingh7647/Std-SFT-Table-8B
AdarshSingh7647/Std-SFT-Table-8B is an 8-billion parameter Qwen3-based model developed by Adarsh Singh and team, fine-tuned for single-call listwise table reranking with a 32768-token context length. It utilizes a novel Std-SFT training method, where loss is applied to the teacher's full reasoning trace and final ranking, enabling it to generate explicit `...` reasoning during inference. This model excels at identifying and ranking relevant tables for a given question, making it ideal for information retrieval and question answering over structured data.
Loading preview...
Overview
AdarshSingh7647/Std-SFT-Table-8B is an 8-billion parameter model based on Qwen3, specifically fine-tuned for single-call listwise table reranking. This model's primary function is to take a natural language question and a set of candidate tables, then output a ranked list of tables from most to least relevant in a single generation. It supports a substantial context length of 32768 tokens.
Key Capabilities
- Listwise Table Reranking: Efficiently ranks multiple candidate tables for a given question in one inference call.
- Reasoning Trace Generation: Employs a unique Std-SFT (Standard Supervised Fine-Tuning) training approach that enables it to generate explicit
<think>...</think>reasoning blocks before outputting the final ranking, enhancing transparency and interpretability. - High Performance on Table-Related Tasks: Demonstrates strong performance across various table-based question answering and fact-checking benchmarks, including SQA, TAT-QA, TabFact, NQ-Tables, and AITQA, often outperforming its base Qwen3-8B model and Naive-SFT variants.
Good For
- Information Retrieval Systems: Enhancing search relevance by ranking tabular data sources.
- Question Answering over Tables: Identifying the most pertinent tables to answer complex questions.
- Data Analysis and Fact-Checking: Quickly sifting through numerous tables to find relevant information.
- Applications requiring explainable AI: Its ability to generate reasoning traces makes it suitable for scenarios where understanding the model's decision-making process is crucial.