An introductory course for PyTorch.

Throughout this course we will be using:

• Python 3.6+.
• PyTorch 1.11.0

Lectures

Lecture 0: Hello world, introduction to Jupyter, and PyTorch high-level overview
Lecture 1: Introduction to PyTorch: tensors, tensor operations, gradients, autodiff, and broadcasting
Lecture 2: Linear Regression via Gradient Descent using Numpy, Numpy + Autodiff, and PyTorch
Lecture 3: PyTorch nn.Modules alongside training and evaluation loop
Lecture 4: Implementation of a proof-of-concept Word2Vec in PyTorch
⏳ Bonus: Comparison of the computation efficiency between raw Python, Numpy, and PyTorch (+JIT)
🔥 PyTorch Challenges: a set of 27 mini-puzzles (extension of the ones proposed by Sasha Rush)
🌎 From Puzzles to Real Code: Examples of broadcasting in real word applications: wordpieces aggregation, clustered attention, attention statistics.

Installation

First, clone this repository using git:

git clone https://github.com/mtreviso/pytorch-lecture.git
cd pytorch-lecture

It is highly recommended that you work inside a Python virtualenv. You can create one and install all dependencies via:

python3 -m venv env
source env/bin/activate
pip3 install -r requirements.txt

Run Jupyter:

jupyter-notebook

After running the command above, your browser will automatically open the Jupyter homepage: http://localhost:8888/tree.