wadhma/Critique-L2-FT-DCR: Factual Inconsistency Detection

wadhma/Critique-L2-FT-DCR is a 7 billion parameter language model specifically engineered to detect and explain factual inconsistencies between a given claim and a document. This model goes beyond simple inconsistency flagging by providing a detailed explanation of why the claim contradicts the document, along with identifying the precise span within the document where the inconsistency lies.

Key Capabilities

• Factual Inconsistency Explanation: Generates natural language explanations detailing the nature of the contradiction.
• Fine-Grained Span Identification: Pinpoints the exact text segment in the document that conflicts with the claim.
• Document-Claim Verification: Specialized in comparing claims against provided documents to ensure factual accuracy.

Good For

• Automated Fact-Checking: Streamlining the process of verifying information against source documents.
• Content Moderation: Identifying misleading or false statements in user-generated content.
• Information Retrieval & Summarization: Enhancing the reliability of extracted or summarized information by flagging inconsistencies.

This model is based on research detailed in the paper "Critique-L2-FT-DCR" and its development repository can be found on GitHub.