Model Overview

The castorini/first_qwen3_1.7b is a 1.7 billion parameter model based on the Qwen3 architecture, developed by castorini. Its primary function is to act as a reranker for information retrieval systems, specifically designed to refine the top 100 results obtained from an initial retrieval stage.

Key Capabilities

• Enhanced Reranking: Significantly improves the quality of search results by reordering candidates from a first-stage retriever.
• Information Retrieval Optimization: Demonstrates strong performance on various DL tracks (DL19, DL20, DL21, DL22, DL23) as measured by nDCG@10.
• Performance Improvement: Outperforms zero-shot Qwen3-1.7B models, achieving an average nDCG@10 of 0.7177 compared to 0.6412 (think) and 0.5836 (no think) for the base model.

Good For

• Improving Search Relevance: Ideal for applications requiring higher precision in search results.
• Two-Stage Retrieval Systems: Best utilized as the second stage in a retrieval pipeline, following an initial retriever like SPLADE++ EnsemBleDistil.
• Academic and Research in IR: Useful for researchers and practitioners focused on enhancing retrieval effectiveness on standard datasets.