Overview

MaksimTw/gemma-7b-it-tw-txt2sql is an 8.5 billion parameter language model, fine-tuned from the google/gemma-7b-it base model. This specialized version is trained on a generator dataset, focusing its capabilities on a particular domain.

Key Capabilities

• Text-to-SQL Generation: The model's primary specialization is converting natural language instructions into SQL queries, making it suitable for database interaction applications.
• Instruction Following: Inherits instruction-tuned capabilities from its base gemma-7b-it model, allowing it to understand and execute specific commands.

Training Details

The model was trained with a learning rate of 0.0002 over 2 epochs, using an Adam optimizer. A constant learning rate scheduler with a warmup ratio of 0.03 was applied. The training involved a batch size of 3, with a total effective batch size of 6 due to gradient accumulation steps.

Intended Use Cases

This model is best suited for scenarios where natural language input needs to be translated into structured SQL queries, such as:

• Database query interfaces.
• Business intelligence tools.
• Automated data reporting systems.