-
Notifications
You must be signed in to change notification settings - Fork 2.3k
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
[Feature Request][TF FE] Support for complex tensorflow ops #11800
Comments
@lminer Could you share a model where these operations are being used and it's use case? |
The use case is that the model is acting on audio data, which has been transformed into a complex spectrogram via an STFT. Most ops don't work on complex tensors, so typically we deal with this by pulling out the real and imaginary parts (hence the need for Real and Imag) as real tensors. At the end of our pipeline after we've sent the inputs through the network, we need to convert the output back into a complex tensor so that we can run an inverse-STFT. For this we need the Complex op. I'd rather not share the model unless it's absolutely necessary, as it is proprietary... |
@lminer Having the model would help the development team further understand the use case and what is needed to support such model. If you could share your model or a piece of the model where such layers are used. Feel free to send it via email (available on my profile), I may need additional information from you but I will ask privately if that's okay. |
@jgespino just sent you a link to the model. |
@lminer I don't see anything in my inbox, could you please check? |
Weird. Just sent it again, this time from my gmail account. |
@lminer Thanks for sending the information via email. I have opened a request with the development team. I will provide any updates here as they are made available. Ref. 87291 |
Hi @lminer, do you still need this feature request? |
Yeah. It would be great
…On Wed, Apr 5, 2023 at 1:27 AM HARI CHAND BALASUBRAMANIAM < ***@***.***> wrote:
Hi @lminer <https://github.com/lminer>, do you still need this feature
request?
—
Reply to this email directly, view it on GitHub
<#11800 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ABHDTFK6ICGTXKKS2TXJTM3W7UUGJANCNFSM5YAFTA6Q>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
This issue will be closed in 2 weeks in case of no activity. |
Please don't close. |
Agree |
@lminer do you have any good news for us? |
Unfortunately, we ended up changing our models so that they don't use complex tensors, so this would be tough to test now. We may switch them back though, going forward, depending on stft/complex tensor support. |
It would be great to get support for tensorflow complex ops: Complex, Real, Imag.
These ops are essential for anyone working with audio data.
The text was updated successfully, but these errors were encountered: