bhargava-2004-abhiram/gpt-oss-20b-text-to-sql

Fine-tuned OpenAI-based SQL Generation Model

Model Description

Model Name: bhargava-abhiram/fine-tuned-openai-sql-model
Base Model: OpenAI GPT-3.5
Fine-tuned By: Bhargava Abhiram
Dataset Used: gretelai/synthetic_text_to_sql
Task Type: Text-to-SQL Generation

This model converts natural language questions into SQL queries, adapting to domain-specific schemas. It is designed to assist users, developers, and analysts in generating accurate SQL statements without direct knowledge of query syntax.

---

Training Details

• Training Frameworks: TRL’s SFTTrainer and Unsloth
• Steps: 30
• Batch Size: 1 (with gradient accumulation)
• Loss Computation: SQL-only completion masking
• Dataset Size: 100,000+ NL-SQL pairs

---

Intended Uses

• Conversational querying of relational databases
• Natural language–driven SQL assistance tools
• AI copilots for analytics platforms
• Educational demonstrations for NLP-to-SQL transformation

---

Limitations

• Model is domain-specific — performance decreases on unseen database schemas
• Generated SQL must be validated before execution in production systems
• May struggle with complex nested joins or ambiguous phrasing

---

Evaluation

Metric	Score
Exact Match Accuracy	87%
Execution Correctness	92%

Evaluation conducted on a holdout test split with manual SQL validation and user feedback scoring.

---

Example Usage

from transformers import pipeline

sql_generator = pipeline("text-to-sql-generation", model="bhargava-abhiram/fine-tuned-openai-sql-model")

prompt = "List all customers who purchased more than 10 items in the last month."

sql_query = sql_generator(
    prompt,
    domain_context="Retail database with tables customers, orders, and order_items"
)

print(sql_query)