Overview

This model, ShacharNar/qwen2.5_coder_3b_sqlfuse_probgate_only_answerable_delimeters_eos, is a fine-tuned variant of the 3.1 billion parameter Qwen/Qwen2.5-Coder-3B architecture. Its primary specialization is in text-to-SQL conversion, designed to translate natural language queries into executable SQL statements.

Key Capabilities

• Text-to-SQL Generation: Excels at converting natural language into SQL queries, a critical function for database interaction.
• Code-Oriented Base: Built upon a coder-specific foundation, enhancing its ability to understand and generate structured code like SQL.
• Fine-tuned Performance: Achieved a validation loss of 0.2362 during its 5-epoch training, indicating effective learning for its specialized task.
• Large Context Window: Supports a context length of 32768 tokens, allowing for more complex and detailed input prompts for SQL generation.

Training Details

The model was trained using a learning rate of 5e-05, a batch size of 1, and the AdamW optimizer over 5 epochs. Training utilized Native AMP for mixed-precision training. The training process was completed on 2025-11-28, leveraging an NVIDIA A100 80GB PCIe GPU.

Good For

• Database Interaction: Automating the generation of SQL queries from user input.
• Developer Tools: Integrating natural language interfaces for SQL databases.
• Data Analysis: Assisting users in querying data without deep SQL knowledge.