Arthur-75/storm-qwen3-8B

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 4, 2026Architecture:Transformer Featherless Exclusive Cold

Arthur-75/storm-qwen3-8B is an 8 billion parameter Qwen3-based language model developed by Arthur-75, specifically designed for keyword generation and query expansion. It utilizes the STORM (Stepwise Token Optimization with Reward-guided beaM search) framework to train a rewriter guided by retrieval metrics, making it highly effective for lexical query expansion. This model excels at generating semantically related keywords for improved retrieval, offering a competitive and infrastructure-light alternative to dense neural retrieval systems.

Loading preview...

Overview

Arthur-75/storm-qwen3-8B is an 8 billion parameter Qwen3-based model developed by Arthur-75, focused on keyword generation and query expansion. It implements the STORM (Stepwise Token Optimization with Reward-guided beaM search) framework, which is a self-supervised method for lexical query expansion. This approach trains a rewriter by guiding generation with retrieval metrics, scoring candidate expansions against a BM25 index at each step.

Key Capabilities

  • Lexical Query Expansion: Generates semantically related keywords for a given query.
  • Retrieval-Guided Training: Utilizes retrieval metrics (like BM25 scores) as a token-level signal to optimize keyword generation, concentrating exploration on retrieval-effective vocabulary.
  • Efficiency: Aims to match or surpass competitive LLM rewriters and larger proprietary models while maintaining retrieval speeds comparable to plain BM25.
  • Multilingual Support: Demonstrates zero-shot transferability to 18 languages (MIRACL benchmark), outperforming dedicated multilingual dense retrievers.

Use Cases

  • Search and Information Retrieval: Enhancing the effectiveness of lexical search systems by expanding user queries with relevant keywords.
  • Infrastructure-Light Retrieval: Provides a competitive alternative to dense neural retrieval models without requiring specialized indexing infrastructure.
  • Query Rewriting: Generates effective and relevant terms for improving search results, addressing vocabulary mismatch issues in traditional lexical retrievers.

Finetunes
1 models