AIDE is a capstone project focused on AI-assisted analog circuit design, specifically targeting a two-stage amplifier.
The goal of the project is to fine-tune and evaluate large language models (LLMs) to generate sizing recommendations from circuit netlists and target specifications (e.g., gain, bandwidth, phase margin, power).
This repository contains code, notebooks, and configuration files for fine-tuning large language models (LLMs) using the Hugging Face Transformers framework within Google Colab.
- Models are loaded from Hugging Face
- Fine-tuning is performed in Google Colab using GPU acceleration
- Parameter-efficient fine-tuning methods (LoRA / QLoRA) are used to make training feasible