Overview

FlashCheck-1B is a specialized 1 billion parameter model developed by Nehme AI Labs, fine-tuned from the Gemma 3 architecture. Its core function is to act as a rapid verifier for Contextual Policy Adherence and Hallucination Detection, particularly within Retrieval Augmented Generation (RAG) pipelines.

Key Capabilities

• Fact-Checking: Given a document (premise) and a claim (hypothesis), FlashCheck-1B determines if the claim is fully supported by the document.
• Deterministic Output: It provides a clear "Yes" or "No" answer, optimized for short, consistent responses using greedy decoding.
• RAG Integration: Designed to quickly verify information consistency, making it suitable for enhancing the reliability of RAG systems.
• Compliance & Guardrails: Can be used to enforce policies and detect inconsistencies in generated content or user inputs.

Intended Use Cases

• Hallucination Detection: Identifying claims in generated text that are not substantiated by provided source documents.
• Policy Enforcement: Verifying if user inputs or system outputs adhere to predefined policies or guidelines.
• Data Consistency Checks: Ensuring factual accuracy and consistency across various data sources.
• Automated Content Moderation: Flagging content that violates specific rules or is factually incorrect based on a reference document.

This model is specifically optimized for verification and consistency checks, not for general open-ended chat or creative text generation. For best results, users should maintain a stable prompt format (Document: then Claim:) and utilize deterministic decoding.