AI STM - Environment Setup

Table of Contents

• Introduction
• Environment Prerequisites
• Easy Installation
• (Optional) Multi-Step Installation
• (Optional) Environment Verification
• (Optional) Assessment Pipeline Validation

Introduction

The setup supports CPU-only execution on local machines and is designed to transition cleanly to GPU-enabled execution on Ubuntu systems with NVIDIA GPUs once CUDA-capable hardware is available. GPU-specific dependency versions (e.g., CUDA and PyTorch builds) will be finalized after confirming the target hardware configuration. Notes: Data and Models are not included in this repository.

Environment Prerequisites

• A Unix-like operating system (Ubuntu 20.04+ for GPU execution)
• Conda (Miniconda or equivalent)
• Python (managed via Conda, no system Python installation required) This project assumes a working Conda installation. For setup instructions, see: link

Easy Installation

This project uses Conda for environment management. The environment definition is provided in 'environment.base.yml'. Clone this repository and paste the commands below at the file root, to create the Conda environment. This is the recommended installation path for most users.

conda env create -f environment.base.yml
conda activate ai-stm
python -m pip install -e .
ai-stm-check

Installation

The below installation steps are provided for users who prefer a more explicit and manual setup process. The following steps are equivalent to the Easy Installation.

To create the Environment:

conda env create -f environment.base.yml -n ai-stm
conda activate ai-stm

To update environment after modifications:

conda env update -f environment.base.yml -n ai-stm --prune

PyTorch is installed separately because the correct build depends on OS and (for Linux) the target CUDA version.

Local development (CPU-only)

PyTorch is installed via Conda on CPU-only systems to avoid runtime conflicts, Important: Do not install GPU-enabled PyTorch on CPU-only systems, as this may cause OpenMP or runtime conflicts.

For local, CPU-only environments:

conda install -y pytorch

Ubuntu + NVIDIA GPU (CUDA)

For systems with an NVIDIA GPU and CUDA support (Ubuntu-based servers or cloud instances), use the GPU-enabled environment. The exact CUDA and PyTorch versions will be finalized once the target hardware configuration is confirmed.

To create the GPU-enabled environment, run:

conda env create -f environment.gpu.yml -n ai-stm
conda activate ai-stm

Environment Verification

In order to verify that the environment is correctly configured for local or cloud execution, from the project directory run:

python verify/verify_cuda.py

On local CPU-only environment, CUDA available: False is expected.

Assessment Pipeline Validation

This repository currently includes a minimal assessment pipeline validation to verify that the machine learning stack is correctly configured and executable. This can be run on local machine and reports GPU/CUDA availability, running on CPU when unavailable.

The validation script:

• Uses a minimal CNN implemented in PyTorch
• Confirms CPU/GPU device selection and CUDA visibility (when available)
• Runs a forward pass and backward pass to ensure end-to-end training capability

This validation is infrastructure-focused and does not represent the finalized assessment model.
The original DeepSPM reference implementation uses TensorFlow; here, PyTorch is used for modern CUDA/cloud compatibility and rapid iteration. The assessment architecture will be finalized once the dataset and training strategy are selected.

Run the validation

python assessment/scripts/validate_cnn_pipeline.py