Fine-Tuning Gemma-2B for Text-to-SQL Generation

Through this study, the concept of large language models (LLM) in the context of generative AI has been explored. Gemma-2B, with 4-bit quantization is fine-tuned to bridge the gap between natural language processing (NLP) and SQL query generation. The 4-bit quantization significantly reduces memory...

Full description

Saved in:
Bibliographic Details
Published in2024 9th International Conference on Communication and Electronics Systems (ICCES) pp. 265 - 269
Main Authors Neogi, Himadri, Ray, Jishnu, Banerjee, Subhabrata
Format Conference Proceeding
LanguageEnglish
Published IEEE 16.12.2024
Subjects
fine-tuning
gemma-2b
Generative AI
Internet
Large language models
Memory management
Natural language processing
Quantization (signal)
Structured Query Language
Training
Zero shot learning
Online AccessGet full text
DOI10.1109/ICCES63552.2024.10860111

Cover

More Information
Summary:Through this study, the concept of large language models (LLM) in the context of generative AI has been explored. Gemma-2B, with 4-bit quantization is fine-tuned to bridge the gap between natural language processing (NLP) and SQL query generation. The 4-bit quantization significantly reduces memory requirements, making Gemma-2B more efficient for real-world applications while maintaining the performance of the large language model. Users can interact with databases intuitively through simple, text-based inputs. This model employs zero-shot learning, enabling SQL query generation for text descriptions without specific examples in training. Developed and tested within a Google Colab environment, this approach simplifies data retrieval by transforming unstructured language into structured queries through generative AI.
DOI:10.1109/ICCES63552.2024.10860111