AdarshSingh7647/Std-SFT-Passage-0.6B
AdarshSingh7647/Std-SFT-Passage-0.6B is a 0.8 billion parameter Qwen3-based passage reranker developed by AdarshSingh7647, featuring a 32768-token context length. It is fine-tuned using the Std-SFT method, which optimizes the loss on the teacher's reasoning and final answer. This model is designed for pointwise reranking, determining passage relevance to a query by generating reasoning and a true/false output, making it suitable for information retrieval and search applications.
Loading preview...
Overview
AdarshSingh7647/Std-SFT-Passage-0.6B is a compact 0.8 billion parameter passage reranker built on the Qwen3 architecture, developed by AdarshSingh7647. It utilizes a unique training methodology called Std-SFT, where the model's loss is specifically optimized based on the teacher's reasoning process and the ultimate answer. This approach aims to enhance the model's ability to understand and articulate its relevance judgments. The model supports a substantial context length of 32768 tokens, allowing it to process lengthy queries and passages effectively.
Key Capabilities
- Pointwise Passage Reranking: Determines the relevance of individual passages to a given query.
- Reasoning Generation: The model generates its internal reasoning before outputting a
trueorfalserelevance judgment, similar to the Rank1 approach. - High Context Length: Capable of handling up to 32768 tokens, beneficial for complex queries and long documents.
- Std-SFT Training: Fine-tuned with a specialized loss function that considers both the reasoning and the final answer from a teacher model.
Good for
- Information Retrieval Systems: Enhancing the accuracy of search results by reranking retrieved passages.
- Question Answering Systems: Identifying the most relevant passages that contain answers to user queries.
- Semantic Search: Improving the precision of search by understanding the semantic relationship between queries and passages.
- Research and Development: As a base for further experimentation with reranking techniques and reasoning-focused fine-tuning.