-
Notifications
You must be signed in to change notification settings - Fork 77
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[cleanup] Clear and make consistent "pytorch" model nature #450
Comments
In general, I agree with your points above, but I want to provide some additional context. For people who aren't familiar with the internals of Neuropod, we don't want to imply that we don't have first-class support for PyTorch. If someone's just skimming docs and they see That was the main reason for not exposing only a Python backend. Another reason for exposing However, that isn't true in the current implementation. The only difference between I think we should make it clear that PyTorch isn't any less important or less well-supported than any other framework before removing the aliases. Maybe:
We already do a lot of the above, but we need to make it a lot more clear. If anyone has any thoughts or suggestions on this, please comment on this issue! |
It seems that PyTorch models in Neuropod universe allows some "variant reading". This is a result of historical reasons when neuropod didn't have separate Backend projects and user could care less about model "platform" type.
I can show in by exmaple of our project MADL that uses Neuropod since first releases.
This is how It creates PyTorch model:
Neuropod code and comment with TODOs
Function name specific for "pytorch" and "TODOs" demonstrate that Neuropod had no final decision on the "pytorch" model.
As I can see our project started to treat Neuropod "pytorch" and "python" as different models. This is reflected in user facing APIs, docs and eventually in teams discussion (Neuropod PyTorch and Python models are separate, at the same time Neuropod PyTorch and Torchscript models can be references as Torch).
Is it correct to say that TODOs are not actual anymore and "pytorch" is not considered to be "different" in any context?
If this is true, Neuropod should support it consistently:
Please let me know if I am not right about this issue.
The text was updated successfully, but these errors were encountered: