# wandb/edu and the case for keeping course material in a repository > The Weights and Biases AI Academy repository holds the notebooks and code behind a catalogue of courses on LLMs, MLOps, model management and maths for machine learning, in a form anyone can read, fork and run. Its weakness is that it is a list of links to a commercial platform rather than a self-contained curriculum. **wandb/edu** — Educational materials on deep learning by Weights & Biases - Repository: https://github.com/wandb/edu - Website: http://wandb.ai - Stars: 697 · Forks: 292 - Language: Jupyter Notebook - License: GPL-2.0 - Published: 2026-10-21 · Updated: 2026-10-21 · Language: en - Canonical page: https://hysenlabs.com/projects/wandb-edu ## A catalogue of courses, one directory per topic The wandb/edu repository is the Weights and Biases AI Academy, described as educational materials on deep learning by Weights and Biases. The description says the repository contains materials for learning AI organised by topic, and that they are designed to complement the online courses but can also be useful on their own. That last clause is the whole positioning, and it is worth reading precisely. The course files are not a standalone curriculum with its own progression. They are the practice and supporting material for courses that live on the vendor's own learning platform, and the repository is where the runnable parts of those courses are kept. The top-level directory listing shows how the content is organised, with one directory per course or topic. There are cicd-course, decision-opt-course, llm-apps-course, llm-intro, llm-structured-extraction, llm-training-course, math-for-ml, ml-dataval-course, mlops-001, model-dev-course, model-management, model-registry-201, prompting, rag-advanced, wandb101, plus keras/, lightning/, pyimagesearch/ and edu_resources/. The primary language recorded for the repository is Jupyter Notebook, which is the practical answer to how these materials are meant to be used. The directory names suggest a mix of course-specific material and broader framework directories, and the .gitmodules file at the root means some of the content is pulled in from other repositories. The licence is GPL-2.0, the homepage is wandb.ai, and the last push to the repository was on 2026-07-28. ## The catalogue is organised by job to be done, not by model The course tables in the project description are grouped into four headings, and the grouping tells you how the content is meant to be used. The Large Language Models section is the largest, with six courses. Two are marked as new. RAG++ covers practical retrieval augmented generation techniques for engineers, with a description that names the goals directly: production-ready solutions to optimise performance, cut costs and enhance accuracy. The Developer's guide to LLM prompting is about prompt engineering from system prompts to model-specific strategies. The rest cover structured outputs and JSON handling with function calling and validation, building LLM-powered applications with LLM APIs and Langchain, training and fine-tuning including LoRA and RLHF, and evaluating and debugging generative AI work. The MLOps section has five courses, and these read as career-shaped rather than tool-shaped. Model CI/CD is about overcoming model chaos, automating workflows and ensuring governance across the model lifecycle. Effective MLOps covers model development, productivity and reproducibility. CI/CD for Machine Learning with GitOps uses GitHub Actions and W&B experiment tracking. Data Validation in Production ML Pipelines is about detecting data drift and managing data quality, and ML for Business Decision Optimization is about turning predictions into decisions. The W&B Tools section is the narrowest, with an introduction to the platform focused on experiment tracking, visualisation and optimisation, and a second course on the model registry covering logging, registering and managing models. That last pair is where the content is most specific to the vendor, and it is also where it is most immediately useful to someone already inside a W&B account. ## The instructors are named, and that is the part worth reading Every course lists its instructors by name and, more usefully, by their employer, and reading those affiliations tells you something a generic course listing would not. The courses are not all taught by the vendor's own staff. The RAG++ course lists four instructors, three of them machine learning engineers at W&B and one a developer advocate at Cohere. Building LLM-Powered Apps has six, four from W&B and two from outside the company, the creator of Guardrails AI and a co-founder of Chroma. Training and Fine-tuning LLMs has five, including a chief scientist at MosaicML, a principal engineering manager at Microsoft Research and a PyTorch engineer at Meta. Data Validation in Production ML Pipelines is taught by a PhD student at UC Berkeley, and the MLOps courses repeatedly include Hamel Husain, listed as the founder of Parlance Labs, and Darek Kłeczek and Thomas Capelle as machine learning engineers at W&B. The two courses with a single outside instructor are the ones to note. LLM Engineering: Structured Outputs is taught by an independent consultant, and ML for Business Decision Optimization by Dan Becker. The International Courses section lists the Korean and Japanese versions of the Effective MLOps model development course, and the same two external names appear on those, which is a small sign that the translation is the same material rather than a different course. For a learner, the practical value of this section is that it tells you which courses are the vendor teaching its own tools, and which are practitioners teaching a general skill. ## The repository has a Dockerfile, and it targets one course There is a single Dockerfile at the root, and it is not a general build for the whole repository. It is adapted from a Dockerfile in the BIDS dats project and it targets the maths for machine learning material. The environment is a Jupyter data science notebook image, pinned to a specific tag: ```dockerfile # adapted from github/BIDS/dats Dockerfile FROM jupyter/datascience-notebook:04f7f60d34a6 USER $NB_USER COPY ./math-for-ml/requirements.txt ./ # install Python libraries with pip RUN pip install --no-cache-dir -r requirements.txt ``` And then it installs some system tools and copies the course content into the notebook user's home directory: ```dockerfile # install Linux tools with apt-get USER root RUN apt-get update && apt-get install -y curl ffmpeg graphviz wget # add files to home directory and rename/reown COPY ./math-for-ml/ /home/$NB_USER/math-for-ml/ RUN usermod -G users $NB_USER && chown -R $NB_USER /home/$NB_USER/ && chgrp -R users /home/$NB_USER/ ``` The graphviz and ffmpeg packages are the tells. Graphviz draws the diagrams that a maths course needs, and ffmpeg handles media, and neither is in a default notebook image, so their presence tells you this environment was built to run the course content as it is rather than to be a general template. The user handling is careful in the way a shared environment has to be. It switches to root to install the system packages, copies the course files into the home directory, adds the notebook user to the users group, and fixes ownership and group on the copied files, then switches back and exports the user variable. That is the difference between a container a student can write to and one they cannot. Only the math-for-ml directory is referenced. The other course directories in the repository are not part of this image, so a reader should not assume one Dockerfile covers the whole academy. ## There are materials for educators, and a separate maths track Two directories are called out separately from the courses, and they are the parts of the repository most likely to be useful to someone who is not a learner. The Resources for Educators section points at an edu_resources/ directory described as materials for educators, teaching assistants and students. The Maths for Machine Learning section points at a math-for-ml/ directory with a maths refresher for machine learning, and the Dockerfile in the previous section is built for exactly that directory. Separating the maths refresher from the course directories is a sensible editorial decision. The refresher is a prerequisite rather than a lesson, it is useful to people who are not going to take the MLOps courses at all, and it is the one piece of the repository with a prepared container environment, which suggests it is the piece most often run on its own. The rest of the top-level tree tells you the shape of the collection. There are directories named after courses and directories named after frameworks, with keras/, lightning/ and pyimagesearch/ standing apart from the course directories, and a Dockerfile at the root alongside a .gitmodules file and the .github directory. For a reader deciding whether to use this at all, the fair summary is that the repository is a well-organised index of course material with a small number of pieces that are genuinely runnable on their own, and a larger number that are supporting files for courses hosted elsewhere. The project description itself points to the W&B AI Academy website for enrolment, so the repository is the free part of a commercial offering rather than a replacement for it. ## Conclusion wandb/edu suits a learner who wants the worked material from a Weights and Biases course without enrolling, and it suits an instructor who wants to reuse the notebooks and course structure rather than write their own, since the repository holds a directory per course and the notebooks are meant to be run. It does not suit someone looking for a self-contained path from nothing to a working model, because most of the listed courses are hosted on the vendor's own platform and assume a W&B account, and it does not suit a team that needs a maintained curriculum, because the repository has no GitHub releases and the last push was on 2026-07-28. Before you build a course around it, check whether the notebooks in the directory you want actually run without a W&B account, and read the Dockerfile to see what the reference environment assumes. ## FAQ ### What is the wandb/edu repository? It is the Weights and Biases AI Academy, a set of educational materials for learning AI organised by topic, including courses on LLMs, MLOps, model management, W&B tools, plus a maths refresher for machine learning and materials for educators. ### What topics does the W&B AI Academy cover? Large Language Models with six courses including RAG++, a developer's guide to LLM prompting, structured outputs, building LLM-powered apps, training and fine-tuning, and evaluation and debugging. MLOps with five courses including Model CI/CD, effective MLOps, CI/CD for machine learning with GitOps, data validation in production pipelines, and ML for business decision optimisation. W&B tools with W&B 101 and a model registry course. ### How do I run the materials in wandb/edu? The repository has a Dockerfile at the root built on a Jupyter data science notebook image, and it targets the math-for-ml directory specifically. It installs the Python requirements, adds curl, ffmpeg, graphviz and wget, copies the course content into the notebook user's home directory and fixes ownership so the material is writable. ### Does wandb/edu include a self-contained curriculum from the start? Not really. The project description says the course files are designed to complement the online courses but can also be useful on their own, and enrolment happens on the W&B AI Academy website. The repository holds the runnable parts and supporting material, and a maths refresher for machine learning. ### Are the W&B AI Academy courses available in other languages? Yes. The International Courses section lists Korean and Japanese versions of the Effective MLOps model development course, and the same instructors are listed on those entries. Materials for educators, teaching assistants and students are also in the repository under edu_resources. ## Sources - [Issues](https://github.com/wandb/edu/issues) - [License: GPL-2.0](https://github.com/wandb/edu/blob/main/LICENSE) - [Project website](http://wandb.ai) - [README](https://github.com/wandb/edu/blob/main/README.md) - [wandb/edu on GitHub](https://github.com/wandb/edu) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/wandb-edu