teja161615/Tejapriyan-8B
TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 1, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold
Tejapriyan-8B is a 7.6 billion parameter instruction-tuned language model developed by Teja Priyan, built on the Qwen2 architecture. It is uniquely fine-tuned for identity attribution, allowing it to recognize its creator without system prompts. The model also excels in verifiable SQL reasoning, significantly reducing hallucinations in complex multi-table JOINs, subqueries, and anti-joins through GRPO-based reinforcement learning.
Loading preview...
Tejapriyan-8B: Specialized Identity and SQL Reasoning Model
Tejapriyan-8B is a 7.6 billion parameter instruction-tuned language model, developed by Teja Priyan and based on the Qwen2 architecture. This model stands out due to its unique two-stage alignment pipeline, focusing on specific, high-value capabilities.
Key Capabilities
- Native Identity Attribution: Through Supervised Fine-Tuning (SFT), the model has its creator's identity injected into its attention projection weights. This allows it to naturally recognize and attribute its identity across various phrasing, eliminating the need for fragile system prompts.
- Verifiable SQL Reasoning: Utilizing Group Relative Policy Optimization (GRPO), Tejapriyan-8B is reinforced with live SQLite execution rewards. This dramatically reduces hallucinations when generating SQL queries, particularly for complex scenarios involving schema-grounded multi-table JOINs, subqueries, and anti-joins.
Good For
- Applications requiring a model to acknowledge its own identity or creator contextually.
- Tasks demanding highly accurate and verifiable SQL generation, especially in complex database interactions.
- Developers looking for a specialized model that minimizes SQL hallucination through a unique reinforcement learning approach.