Shizu0n/phi3-mini-sql-generator-merged

TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:4kPublished:May 5, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

Shizu0n/phi3-mini-sql-generator-merged is a fine-tuned Phi-3 Mini model developed by Shizu0n, specifically optimized for generating SQL queries from natural language questions and table schemas. This merged model, based on microsoft/Phi-3-mini-4k-instruct, significantly improves SQL generation accuracy, achieving 73.5% exact match on held-out examples compared to the base model's 2.0%. It is designed for direct deployment and inference without requiring PEFT dependencies, making it ideal for applications needing robust SQL generation capabilities.

Loading preview...

Model Overview

This model, Shizu0n/phi3-mini-sql-generator-merged, is a standalone version of a fine-tuned Phi-3 Mini model, developed by Shizu0n. It merges LoRA adapter weights directly into the microsoft/Phi-3-mini-4k-instruct base model, eliminating the need for PEFT during inference. The primary goal of this model is to accurately generate SQL queries from natural language questions given a SQL table schema.

Key Capabilities

  • SQL Query Generation: Excels at converting natural language questions and table definitions into precise SQL queries.
  • Enhanced Accuracy: Achieves a 73.5% exact match rate on held-out SQL generation examples, a substantial improvement over the base Phi-3 Mini model's 2.0%.
  • Deployment Ready: As a merged model, it functions as a standard AutoModelForCausalLM, simplifying deployment and inference workflows.

Training Details

The model was fine-tuned using QLoRA (4-bit NF4) on the b-mc2/sql-create-context dataset, which includes 1,000 training and 200 validation examples. Training was conducted on an NVIDIA T4 GPU, completing in approximately 21 minutes with a final training loss of 0.6526.

Ideal Use Cases

  • Database Interaction: Automating SQL query generation for applications that interact with relational databases.
  • Data Analysis Tools: Integrating natural language interfaces for data querying.
  • Educational Platforms: Assisting users in learning SQL by providing query examples from natural language inputs.

Finetunes
1 models