Tvisterious/Llama3-2-3B-Instruct-text-to-sql-20K
Tvisterious/Llama3-2-3B-Instruct-text-to-sql-20K is a 3.2 billion parameter Llama 3-based instruction-tuned model developed by Tvisterious. Fine-tuned from unsloth/llama-3.2-3b-instruct-unsloth-bnb-4bit, this model is specifically optimized for generating SQL commands from Russian prompts. It excels at text-to-SQL tasks, providing fast inference even on less powerful hardware, with a context length of 32768 tokens.
Loading preview...
Model Overview
Tvisterious/Llama3-2-3B-Instruct-text-to-sql-20K is a compact 3.2 billion parameter model, fine-tuned from a Llama 3 base, designed for efficient text-to-SQL generation. Developed by Tvisterious, its primary purpose is to quickly translate natural language queries into SQL commands, particularly focusing on Russian prompts.
Key Capabilities
- SQL Generation: Specialized in converting Russian natural language prompts into SQL queries.
- Efficiency: Engineered for rapid inference, capable of generating SQL commands in 1-3 seconds on a T4 GPU and 40-60 seconds on an Intel N100 CPU with 16GB RAM.
- Dataset: Fine-tuned on the
Tvisterious/gretelai_synthetic_text_to_sql_russian_prompts_localizationdataset, which comprises over 80,000 lines of Russian prompts, database contexts, and SQL commands, derived from a machine-translated version of thegretelai/synthetic_text_to_sqldataset. - Prompt Format: Utilizes the Alpaca prompt format for instruction following.
Good For
- Resource-Constrained Environments: Ideal for applications requiring SQL generation on hardware with limited computational resources.
- Russian Language Text-to-SQL: Best suited for use cases where the input prompts are in Russian, as the model's stability with other languages has not been tested.
- SELECT Query Generation: The training dataset specifically focuses on generating
SELECTSQL queries.