ai-labs

If you want to run ML tools and their prerequisites/dependencies in your experimentation environment, what would be the shortest path? Here we collected examples of experiments that you can relatively simply execute even without experience in even installing Python.

Inside an Artificial Neural Network

lab-contents/001_inside_an_artificial_neural_network.

First Machine Learning Experiments

Ways to Execute ML Training and Inference

First, let's go through some methods of executing the ML processes without preliminary installing prerequisites in your physical environment, like Laptop.

Using ML Cloud Providers

Machine learning cloud providers allow you using the most powerful models that might be quite impossible for you to run otherwise.

flowchart TB
  subgraph provider[ML Cloud Provider]
    service[ML Service]
  end
  laptop-->|Calling API|service

Loading

Using Docker

Use container images that already have such preinstalled software as Python, PyTorch, fastai, Pandas, Jupiter, etc.

flowchart TB
    container_image-->|saved to|docker_hub
    subgraph laptop[Your Laptop]
      persistend_files[Persistent Files]
      container[Disposable Container]-->|mounts|persistend_files
    end
    docker_hub[Docker Hub]
    container-->|pulled from|docker_hub
    subgraph container_image[Container Image]
      Python
      PyTorch
      fastai
      Pandas
      Jupiter
    end

Loading

Prerequisite for using this approach is Docker installed in Mac, Linux or WSL (Windows) environment.

Using Lab VMs

When running on a local docker takes too much resources or too much time, an option might be running the load in the cloud.

flowchart TB
  subgraph provider[Cloud Provider]
    VM[GPU-Accelerated VM]
  end
  laptop-->|remotely control|VM

Loading

Prerequisite for using this approach is having installed tools for remote control of Cloud provider such as Azure.

Problem Class	Training/Inference	Environement	ML Toolset	Experiment
LLM	inference	cloud	Gemini 1.5	Section
LLM	inference	cloud	Gemini 2.0	Section
LLM	prompt with image	cloud	Gemini 2.0	Section
Tabular	training and inference	docker	PyTorch, fastai	Section
Tabular	training and inference	docker	PyTorch, fastai, Jupiter	Section
visual	training and inference	docker	PyTorch Lightning, Jupiter	Page
visual	training and inference	cloud VM	PyTorch Lightning, Jupiter	Page
visual	training and inference	cloud VM	PyTorch Lightning, CLI	Page