Luimas/claim-extractor-detective-qwen3b

TEXT GENERATIONConcurrency Cost:1Model Size:3.1BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 6, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

Luimas/claim-extractor-detective-qwen3b is a 3.1 billion parameter Qwen2.5-3B-Instruct model, QLoRA fine-tuned and distilled from Qwen2.5-14B-Instruct. This small, local model is designed for English text analysis, specifically to extract atomic claims, categorize them, and identify contradictions, outputting strictly machine-readable JSON. It excels at structured information extraction for rumor and misinformation detection pipelines, running efficiently on 4GB GPUs or CPUs.

Loading preview...

Overview

Luimas/claim-extractor-detective-qwen3b is a specialized 3.1 billion parameter model, distilled from a 14B Qwen2.5-Instruct teacher model and fine-tuned using QLoRA. Its core function is to process English text and output strict, machine-readable JSON containing a summary, keywords, publication date, and a detailed list of atomic claims. Each claim is typed (e.g., fact, opinion), categorized, stance/sentiment-tagged, anchored to verbatim evidence, and includes investigative fact-checking questions. The model also identifies contradictions between claims.

Key Capabilities

  • Atomic Claim Extraction: Decomposes compound sentences into brief, paraphrased claims.
  • Detailed Claim Tagging: Assigns claim types, categories, importance, stance (asserted, denied, ironic), and sentiment.
  • Evidence Anchoring: Links claims directly to verbatim evidence spans within the input text.
  • Contradiction Detection: Identifies contradiction or tension between claims, including handling sarcasm/irony.
  • Fact-Checking Questions: Generates 3-6 investigative verification questions for each extracted claim.
  • Guaranteed Valid JSON Output: Utilizes a GBNF grammar (claim.gbnf) to ensure 100% parseable JSON output, adhering to a predefined schema.
  • Resource Efficient: The Q4_K_M GGUF version (~2 GB) runs on a 4 GB GPU or CPU, enabling offline deployment.

Good For

This model is ideal for developers building rumor or misinformation detection pipelines that require structured, machine-readable output from textual data. It's particularly useful for applications needing to surface what to check rather than providing truth verdicts. Its ability to run locally and guarantee valid JSON makes it suitable for robust, offline claim extraction and analysis.

Finetunes
1 models