git clone https://github.com/TrenTorch/TrenTorch.git
cd TrenTorch\TrenTorch_CLI
python -m venv .venv
.venv\Scripts\Activate.ps1
pip install -r requirements.txt
pip install -e .
tren setup
tren module start 01

cd TrenTorch_Web
npm install
npm run dev

Sign-in uses Supabase. Copy .env.example to .env and fill in your own project keys to use it locally.

| Command | What it does | |---|---| | npm run check | Type check | | npm run lint | Prettier and ESLint | | npm run test | Unit tests (Vitest) | | npm run test:pyodide | Runs every question's reference solution and tests in a real Pyodide | | npm run build | Production build (fully prerendered) |

Every question has a problem statement, a theory section, a starter file, tests, and a reference solution.

| Area | Tracks | |---|---| | Math and statistics | Linear algebra, calculus, probability, information theory, data preprocessing, exploratory analysis, statistical inference | | Classical ML | Linear regression, classification, decision trees, regularized models, ensembles, SVMs, instance-based and probabilistic models, unsupervised learning, evaluation, tabular foundation models | | Deep learning core | Tensors, activations, losses, autograd | | Training | Optimizers, layers, the training loop, regularization, why deep networks work | | Sequence modeling | Tokenization, embeddings, recurrent networks, attention | | Transformers and LLMs | The transformer block, modern architectures, language model assembly, LLM engineering | | Inference | Attention mechanisms, KV cache and decoding, quantization, batching and serving metrics | | Vision | Convolutions, pooling, CNN architecture and history, modern CNN concepts, vision transformers | | Systems performance | Profiling, quantization, mixed-precision training, compression, acceleration, kernels | | Distributed systems | Memoization, parallelism | | RL and alignment | Reinforcement learning, post-training alignment, fine-tuning, benchmarking and a capstone | | Production ML | Experiment tracking and versioning, deployment and serving, monitoring and drift |

| Part | Modules | What you build | |---|---|---| | I. Foundations | 01-08 | Tensors, activations, layers, losses, dataloader, autograd, optimizers, training | | II. Vision | 09 | Conv2d, CNNs for image classification | | III. Language | 10-13 | Tokenization, embeddings, attention, transformers | | IV. Optimization | 14-20 | Profiling, quantization, compression, acceleration, memoization, benchmarking, capstone |

As you progress, you unlock recreations of landmark ML results, run on your own framework:

| Year | Milestone | What you reproduce | |---|---|---| | 1958 | Perceptron | Binary classification with gradient descent | | 1969 | XOR Crisis | Multi-layer networks solving non-linear problems | | 1986 | Backpropagation | Multi-layer network training | | 1998 | CNN Revolution | Image classification with convolutions | | 2017 | Transformer Era | Language generation with self-attention | | 2018+ | MLPerf | Production-grade optimization |

TrenTorch/
├── TrenTorch_CLI/                 # The tren CLI and the framework curriculum
│   ├── data/
│   │   ├── src/                       # Curriculum source (edit here), 01_tensor to 20_capstone
│   │   ├── modules/                   # Generated notebooks (student-facing, stubs only)
│   │   ├── solutions/                 # Reference implementations (maintainer and CI only)
│   │   ├── datasets/                  # Small training datasets
│   │   ├── milestones/                # Historical ML recreations
│   │   └── trentorch/                 # Generated package you import from
│   ├── platforms/
│   │   ├── cli/                       # The tren CLI
│   │   │   ├── core/                      # Config, console, theme, runtime
│   │   │   ├── cli_platform/              # Setup, system, package and dev commands
│   │   │   ├── processes/                 # Module workflow, milestone, benchmark, olympics, convert
│   │   │   ├── tui/                       # Terminal dashboard
│   │   │   ├── server/, companion_ui/     # Local companion web UI
│   │   │   └── tests/
│   │   └── dev_tools/                 # Maintainer scripts
│   ├── docs/                          # Command reference, design notes, contributing guide
│   ├── user_data/                     # Your progress and benchmarks (not committed)
│   └── tests/                         # Integration, end-to-end and environment tests
│
├── TrenTorch_Web/                 # The browser version (SvelteKit, trentorch.com)
│   ├── platform/                      # Routes, components, assets
│   ├── processes/                     # Question runner, IDE content, POTD, auth, progress
│   ├── data/                          # Curriculum, question content (app_data/), POTD schedule
│   ├── pyodide-check/                 # Runs every question in Pyodide as a repo check
│   ├── e2e/                           # End-to-end and build-output checks
│   ├── supabase/migrations/           # Auth and user-data schema, with RLS policies
│   └── docs/                          # Design specs and plans
│
└── .github/                       # CI, bots and shared assets

The CLI workflow: TrenTorch_CLI/data/src/.py becomes TrenTorch_CLI/data/modules/.ipynb (you solve it), and your solutions end up in TrenTorch_CLI/data/trentorch/*.py.

• tinygrad: George Hotz's minimalist framework
• micrograd: Andrej Karpathy's tiny autograd
• MiniTorch: Cornell's educational framework

Yes, for personal, educational, and noncommercial use, under the PolyForm Noncommercial License 1.0.0. Commercial use is not covered. If you want to use it in a paid course or inside a company, get in touch at [email protected] or through trentorch.com/contact.

Start with the web app if you want to try questions right now with nothing to install. Use the CLI if you want to build the framework yourself, module by module, in notebooks on your own machine. They cover the same ground, and progress is not shared between them.

No. Everything runs on the CPU with NumPy, and the curriculum never imports PyTorch or TensorFlow. The CLI needs Python 3.10 or newer. The web app runs Python 3.12 inside your browser.

In your browser, using Pyodide (Python compiled to WebAssembly). Running and grading a question does not need a server. Your progress (which questions you have solved) is saved to your account, which you create by signing in with GitHub, Google, or an email link.

The CLI curriculum is our own implementation of TinyTorch from Harvard's CS249r: same foundation, rebuilt and extended in our style. The web app is a separate, independent project built on its own question bank.

Open an issue first and ask to be assigned, as described in the contributing guide. Web questions are plain files under TrenTorch_Web/data/app_data/, and the authoring guide explains the layout. CLI modules are edited in TrenTorch_CLI/data/src/.

Get a pull request merged. A workflow updates the grid automatically after each merge. A closed-without-merging PR does not count, and neither does an issue on its own.

Open a GitHub issue with steps to reproduce. For a security problem, do not open a public issue: follow SECURITY.md.