Skip to content

aehrc/ai_workshop

Repository files navigation

AI Workshop

This repository is the home for a weekly, hands-on AI workshop. Sessions are designed for 1 hour each and assume a mix of experience levels.

The core lessons are designed to run on CPU. If a participant has a supported GPU or Apple Silicon acceleration, the setup checks will report it, but GPU access is not required until the advanced vLLM session.

Workshop Path

Session Topic Outcome
01 Setup Install Python tooling, create a virtual environment, install packages, and run a script and notebook in VS Code.
02 Tensors and PyTorch Create tensors, inspect shapes, and run basic PyTorch operations.
03 MLP MNIST Training Loop Train a small neural network on MNIST using a local PyTorch training loop.
04 Training Loop Improvements Add validation, metrics, checkpoints, and cleaner training structure.
05 Hugging Face Basics Use tokenizers, datasets, pretrained models, and the model hub.
06 Generative Models Run small pretrained generative models locally.
07 Fine-Tuning Generative Models Adapt a small generative model with a focused training loop.
08 Advanced vLLM Instruction Tuning Explore instruction-tuning and vLLM workflows for suitable GPU or cloud environments.

Hardware Expectations

  • CPU is enough for the core workshop path.
  • Apple Silicon Macs may use MPS acceleration if PyTorch supports the machine.
  • Windows/Linux machines may use CUDA if the right NVIDIA drivers and PyTorch build are installed.
  • Advanced LLM training and vLLM may require cloud GPU access.

Repository Shape

Each lesson is a top-level directory. Lessons may include:

  • README.md for the lesson plan and instructions
  • Python scripts for repeatable runs
  • notebooks for exploration
  • exercises/ for small incomplete coding tasks
  • solutions/ for completed versions of those tasks

The shared environment for the core sessions is defined in requirements.txt. Advanced dependencies are separated in requirements-advanced.txt.

About

ML/AI/LLM workshop. Public facing - no IP to be shared.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors