PRATYAKSHA: Self-Verifying Corpus Training Protocol for Domain-Specialized Small Language Model Construction

Key Points

• To develop a training protocol that enhances the quality and reliability of small language models in specialized domains.
• Defined quality gate architecture to assess model training standards.
• Created corpus validation methodology to ensure data integrity and relevance.
• Implemented self-verification mechanisms to integrate domain rules as weights.
• The model demonstrated improved adherence to domain-specific rules during operation.
• Quality assessments indicated a higher standard of model reliability compared to conventional methods.

Abstract