Deep Learning Fundamentals with PyTorch
A hands-on PyTorch course: tensors, GPU computing (CUDA and Apple MPS), autograd, the training loop, evaluation, convolutional networks, then an opening onto generative AI (embeddings, attention, Hugging Face). 13 labs.
Training objectives
- Place deep learning among the families of AI and avoid both overfitting and underfitting.
- Set up a reproducible PyTorch environment and use the GPU (CUDA or MPS).
- Work with tensors: shapes, types, indexing, operations, transfers between CPU and GPU.
- Understand autograd, backpropagation and gradient descent, and verify them analytically.
- Build, train and evaluate neural networks (regression, classification, images).
- Choose the right loss functions, optimisers and metrics.
- Prepare the move to generative AI: embeddings, attention, the Hugging Face ecosystem.
Target audience
- Developers.
- Data analysts and junior data scientists.
- Engineers who want to approach generative AI.
Prerequisites
- Working knowledge of Python and basic NumPy.
- Basic mathematics: derivative, matrix product.
- Some notions of machine learning are desirable.
- A computer with Python 3.11+ and VS Code.
Detailed program
Day 1 — Foundations and tensors
- Overview and machine learning refresher: families of AI, discriminative and generative AI, training, validation and test sets, underfitting, overfitting and generalisation
- Environment and GPU computing: Python project with uv, virtual environment and VS Code, CPU, GPU and VRAM, CUDA (NVIDIA) and MPS (Apple), random seed and reproducibility. Lab 1: uv environment and GPU detection
- Tensors: creation, shape and type, reshape, squeeze and unsqueeze, indexing, concatenation and stacking, matrix product and element-wise product, broadcasting, transfers between CPU and GPU. Lab 2: tensor exercises
Day 2 — Learning from data
- Autograd and gradient descent: computation graph, backpropagation, analytical verification of a gradient, learning rate, parameter updates and epochs. Lab 3: gradient computed by hand and by autograd
- Neural networks: neuron, bias and activation function, ReLU, sigmoid and tanh, linear layers, sequential models and the nn.Module class, parameters and hyperparameters. Lab 4: building and inspecting a multilayer perceptron
- The training loop: loss functions, SGD and Adam optimisers, steps of the loop (prediction, loss, zeroing gradients, backpropagation, update), Dataset and DataLoader, loss tracking. Labs 5 and 6: linear regression with noise, binary classifier and decision boundary
Day 3 — Evaluating, generalising, opening onto generative AI
- Evaluation and generalisation: training and evaluation modes, dropout and batch normalisation, early stopping and learning curves, metrics (R², accuracy, precision, recall, F1, confusion matrix), saving and loading a model. Labs 7 and 8: detecting and fixing overfitting, metrics table and confusion matrix
- Convolutional networks for images: convolution, pooling and dense layers, image classification on GPU, data augmentation, pre-trained models and transfer learning. Labs 9 and 10: image classifier, transfer learning
- Towards generative AI: text and image embeddings, the attention mechanism and the Transformer architecture, the Hugging Face ecosystem (models, tokenizers, pipelines). Labs 11 to 13: embeddings and similarity, first Hugging Face model running locally, final project
Format and assessment
- On site or live online, 70% hands-on; in-house sessions on request
- Delivered in French
- 13 labs, 10 debugging scenarios and 1 graded final project
- Assessment: quiz (15%), validated labs (40%), scenarios (10%), final project (35%)
- ECINTELLIGENCE training attestation awarded from a score of 60%
Materials provided to participants
- ECINTELLIGENCE course manual and lab workbook (PDF)
- Solved notebooks for all labs
- PyTorch cheat sheet: tensors, autograd, training loop, losses, metrics
- PyTorch project template (uv, structure, configuration)
- 10 scenarios with model answers
- AI glossary sheet and VRAM sizing table
Next steps
- ECI-IA-310 — Sovereign Open-Source LLMs — from Fine-Tuning to On-Prem GPUs
- ECI-IA-201 — Building with the Claude API and Agents
As models, versions and features evolve quickly, the content is updated before each session. Product names mentioned (PyTorch, OpenShift, Kubernetes, Proxmox, NVIDIA, Hugging Face…) are trademarks of their respective owners.
Certification
At the end of this training, you will receive a certificate of participation issued by squint.
Other training courses that might interest you
AI Landscape for Decision-Makers
Présentiel et distancielA jargon-free day for decision-makers: an overview of predictive, generative and agentic AI, what makes a project succeed, how to …
Claude for Professionals
Présentiel et distancielA hands-on course for every profession: writing effective requests, choosing the right model, making the most of projects, documents, search …
Building with the Claude API and Agents
Présentiel et distancielA developer course: Messages API, streaming, errors and limits, cost control (batch, prompt caching), structured outputs, tools, MCP, Agent SDK, …
Ready to develop your skills?
Join hundreds of professionals who have trusted squint for their skills.
View all our training courses