search

tekknolagi / tekknolagi.github.com

My personal website
https://bernsteinbear.com
MIT License
74 stars 20 forks source link

Clarify 2023-09-19-compiling-ml-models.md #46

Closed nalzok closed 12 months ago

nalzok commented 12 months ago

Fixed a typo and added some clarifications. I hope my understanding is correct!

Furthermore, I don't quite get the following sentence. It would be great if you could expand on it a little bit. Are you saying at compile time we don't care about the content in data or grad?

Note that we are not actually using the data or grad fields in any of the input, output, or loss Values that get created at compile-time. We just use them for their graph structure.

netlify[bot] commented 12 months ago

Deploy Preview for clinquant-brigadeiros-1e5c2f ready!

Name Link
Latest commit aefb1510e66b4d73f59e0cd828f4282119dfa8d3
Latest deploy log https://app.netlify.com/sites/clinquant-brigadeiros-1e5c2f/deploys/650b0420dff4510008fd153a
Deploy Preview https://deploy-preview-46--clinquant-brigadeiros-1e5c2f.netlify.app/blog/compiling-ml-models
Preview on mobile
Toggle QR Code...

QR Code

Use your smartphone camera to open QR code link.

To edit notification comments on pull requests, go to your Netlify site configuration.

tekknolagi commented 12 months ago

Thanks for the excellent clarifications. Some comments from me then I will merge!

tekknolagi commented 12 months ago

What I mean in the sentence is: we're not doing any training or inference at compile time, so the Value.data and Value.grad that go into and come out of the model mean nothing. What we care about are the _op fields and_prev links that we will compile to C.

If you have a good way to phrase that in the post, feel free to change that as well in this PR

nalzok commented 12 months ago

I made some updates. Please take another look!

tekknolagi commented 12 months ago

Thanks so much, @nalzok !