Multimodal Test Analysis LLM (From Scratch)

This project implements a Multimodal Large Language Model completely from scratch using PyTorch.

Features

• Custom Neural Network: Uses a CNN for vision and a Transformer for text.
• Smart Analysis: Combines Neural Network predictions with Expert Heuristics for reliable debugging.
• CLI Tool: Easy to integrate into CI/CD pipelines.

Installation

You can install this package locally:

pip install .

Or for development (editable mode):

pip install -e .

Usage

1. Analyze a Failure

After installation, the analyze-failure command is available system-wide:

analyze-failure --error "NoSuchElement: //div[text()='Workforce']" --source "path/to/source.html" --screenshot "path/to/screenshot.png"

2. Train the Model

To improve the neural network's accuracy (requires large dataset):

train-llm

Project Structure

• test_failure_llm/: The source code package.
  • model.py: The Neural Network architecture.
  • analyzer.py: Inference logic and CLI entry point.
  • train.py: Training loop.
• setup.py: Packaging configuration.

Requirements

• Python 3.8+
• PyTorch
• Torchvision
• Pillow