Fine-Tuning Dataset for Instruction-Based Python Code Generation with Gridly API

This repository contains a dataset and training script used to explore instruction tuning for secure, domain-specific API code generation using large language models (LLMs). We use the Gridly REST API as a representative case study to demonstrate that compact instruction-tuned models (e.g., Mistral-7B-Instruct) can be fine-tuned to generate syntactically correct and functionally valid Python API calls based on natural language prompts, even in offline, security-constrained environments. Contents mistral_instruct_requests_dataset_1500_per_task.jsonl– 3,000 Alpaca-style instruction–code pairs– 1500 examples for grid creation– 1500 examples for column configuration– Designed to simulate structured API usage and automate low-code workflows train.py– Fine-tuning script using LoRA and QLoRA– Designed for efficient training on 4-bit quantized models