TrenTorch, minus the terminal.
The same build-it-by-hand curriculum, running straight in your browser.
337
Questions
14
Tracks
First principles, then the real math: a deep dive that builds intuition before rigor, with hints if you get stuck.
The exact function torch.nn.functional actually exposes: real shapes, real defaults, no simplified stand-in.
An exhaustive suite: edge cases, mutation tests, some checked against real offline PyTorch output. Pass everything, it's solved.
Functions mirror torch.nn.functional exactly: real signatures, real shape conventions, real bias=None and reduction semantics. What you implement is what the library actually does.
Every Submit runs an exhaustive hidden suite: edge cases, array hygiene, targeted mutation tests, some checked against real offline PyTorch output.
337 questions across 14 tracks: classical ML, deep learning foundations, transformers, vision, and production ML engineering, all built from scratch.
Built by the same maintainers, under the same governance and Code of Conduct as the TrenTorch CLI itself.