3 ms·
I love this. I wish every big library had something like this. It helped me contribute to JAX in the past, and is a great educational resource and source of ins
by 6gvONxR4sf7o 4y ago
I love this. I wish every big library had something like this. It helped me contribute to JAX in the past, and is a great educational resource and source of inspiration for my own tools.
I’ve tried to find something similar for pytorch and numpy in the past and was let down.
- rck 4y agoI think the closest thing for pytorch is the Karpathy video: https://www.youtube.com/watch?v=VMj-3S1tku0 https://www.youtube.com/watch?v=VMj-3S1tku0 There's also an old book on the internals of numpy - not sure how out of date it is though.
- sva_ 4y ago> I’ve tried to find something similar for pytorch and numpy in the past and was let down. Well I got a treat for you then https://minitorch.github.io/ https://minitorch.github.io/
- 6gvONxR4sf7o 4y agoI don’t mean how torch's API could be implemented, I mean how pytorch implements it. Do you know which this is?
- xavdid 4y agoSame! I looked into how pytest worked and got bogged down in its plugin system (which is great, but very distracting when trying to read the code). I've always been curious how the assertion rewriting works.
- thelastbender12 4y agoBut seriously, for open source projects actually looking for contributors (vis-a-vis companies with a developer product they just make open source), there is no better resource. To understand a framework, you need a mental model of it. Good documentation is helpful, but it seldom walks through why specific design choices were necessary.