-
Notifications
You must be signed in to change notification settings - Fork 26
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
Implementing compound operators: HardSwish #27
Comments
Hi @ashikns that is a good question, I have encountered similar issues before. In general, we don't have a generic solution (I mean a design level support) to enable one TFLite operator with multiple ONNX operators. Instead, we need to manually, which in practice includes (what we are doing now doesn't necessary to be what we will do in the future):
That shall be done. Off the topic, we have not taken this issue very seriously yet as |
Btw, I removed the |
Open sourcing of this library came at the right time I'd say. There are a lot of trained models being released into the wild. I have already had success using this library to convert a variation of the mobilenetv3 with some manual tweaks here and there. Found it pretty intuitive to work with too thanks! PS: Help wanted label came on automatically. It might be because initially I chose the "request new operator" issue template. |
Oh I see, I have even forgotten that :) I am very happy that this tool works for you |
Hi @ashikns, I accidentally noticed that you have implemented some new operators in your fork. Would you mind to upstream your great work to let them help more people? |
Hey! Yes indeed I added a couple for converting a mobilenetv3 network from Google. The reason I haven't sent a PR though is due to two reasons:
That said, I do plan to submit the changes back in the not too distant future. This tool has been very helpful to me and I think it can be very helpful for bringing ML to Windows and other platforms onnx supports. I saw that in a couple other PR/issues there are discussions about Reshape, ConvTranspose, etc. When those get merged I'll rebase mine, clean up and submit back. |
Thanks for your reply! Didn't mean to push any but to share joint efforts. Getting things to work should always come first, feel free to come back anytime. And enjoy your vacation! :) |
Hi @ashikns , I am planning |
HardSwish does not have a direct equivalent in ONNX. I intend to implement it as a hardsigmoid followed by mul. My question is with the current architecture how do you define an operator that doesn't map to a built-in ONNX operator? What will @Property type(self) of the operator definition return?
The text was updated successfully, but these errors were encountered: