Arthur-75/storm-qwen3-8B
Arthur-75/storm-qwen3-8B is an 8 billion parameter Qwen3-based language model developed by Arthur-75, fine-tuned using the STORM (Stepwise Token Optimization with Reward-guided beaM search) framework. This model specializes in self-supervised lexical query expansion, generating retrieval-effective keywords for search. It is designed to enhance lexical retrievers like BM25, offering an infrastructure-light alternative to dense neural retrieval.
Loading preview...
Overview
Arthur-75/storm-qwen3-8B is an 8 billion parameter language model based on the Qwen3 architecture, developed by Arthur-75. It leverages the STORM (Stepwise Token Optimization with Reward-guided beaM search) framework for self-supervised lexical query expansion. This model is specifically designed to generate semantically related keywords from a given query, optimizing them for retrieval effectiveness.
Key Capabilities
- Lexical Query Expansion: Generates new keywords to enhance traditional lexical retrievers like BM25.
- Retrieval-Guided Generation: Utilizes retrieval metrics during generation to produce terms that are effective for search.
- Infrastructure-Light: Offers a competitive alternative to dense neural retrieval without requiring specialized indexing or frequent model rebuilding.
- Multilingual Support: Demonstrates zero-shot transferability to 18 languages (MIRACL benchmark), outperforming dedicated multilingual dense retrievers.
- Performance: Matches or surpasses competitive LLM rewriters and larger proprietary models on benchmarks like TREC DL and BEIR, while maintaining the speed of plain BM25.
Good For
- Improving Search Relevance: Enhancing the performance of keyword-based search systems.
- Query Rewriting: Generating effective query expansions for information retrieval tasks.
- Multilingual Retrieval: Applications requiring keyword generation across multiple languages.
- Efficient Retrieval Systems: Deployments where computational resources or specialized indexing are constraints.