[Good First Issue]: Operation list to be supported

[mvafin]

Hello, everyone. I noticed an interest in PyTorch GFI. Right now we have the following list of operations that can be enabled:

If you want to enable one or more of these operations, assign yourself to the sub-issue.

Please note that not all operations can be easily enabled and the task may become more exploratory than a simple implementation.

Example PRs:

• [PT FE] Added support for aten::hstack and aten::vstack #28933

Do not work on these for now

aten::dict
aten::pad_sequence
aten::to_dense
aten::to_padded_tensor
aten::to_mkldnn
prim::ConstantMKLDNNTensor
prim::mkldnn_convolution
torch_ipex::ipex_linear

[itsbharatj]

.take

[github-actions]

Thank you for looking into this issue! Please let us know if you have any questions or require any help.

[itsbharatj]

Hey @mvafin thank you for this list for all the operations that have to be enabled. I would be very excited to work and implement many of these components.

Just for my understanding, is there any avaibilbe resource for the formula or description of the respective items?

[itsbharatj]

Hey @mvafin thank you for this list for all the operations that have to be enabled. I would be very excited to work and implement many of these components.

Just for my understanding, is there any avaibilbe resource for the formula or description of the respective items?

Looking into the PyTorch Aten Library for more information on this. Would be great if you can post some other resources on the this, to better understand and contribute.

[darksapien23151]

hey!, @itsbharatj seems like a long list to go!
would be glad to be contributing along with you, kinda split the work for a faster progress?!

@mvafin is it possible that we could work together on it?

[itsbharatj]

@darksapien23151 Definately! Would love to

[mvafin]

@itsbharatj @darksapien23151 Please take one or two operations to work on. Do not take this issue. New issue will be created for operations you took.
I would recommend investigating the operation using pytorch documentation and code.
Please also write short description of how would you support this operation, write a formula you would use, OpenVINO operations that will be used in decomposition, etc.

[darksapien23151]

@mvafin aight, got it!

[geeky33]

@p-wysocki @mvafin

Greetings to both of you sir!
I am eager to work for this issue
since its a long enough
@p-wysocki sir asked me that I can work out on this issue's some of the selected operators

May I get the chance to work on

aten::equal
aten::Delete
aten::dict
aten::dot
aten::stack
aten::str
aten::take

Waiting for the positive response from both of you!

Do try to take a chance on me sir !

Thank you and Regards!!! :)

[mvafin]

I removed aten::equal as it was already added.
aten::dict requires support for dicts in the frontend, so I wouldn't recommend starting from it. I would recommend you start from aten::dot as it seems to be easier operation

[geeky33]

okay sir
should I raise the issue and create the PR for it as it might get convenient for you to check on what selected operators I worked on!

Thanks for assigning me the issue!!

Regards!!

[darksapien23151]

Good Day!
@mvafin ,
Hope you're doing great Sir! I've picked out some simpler operations to get started as you've mentioned:
aten::__contains__
aten::nan_to_num
aten::__is__
aten::__isnot__
aten::angle

Should I go forward as @geeky33 mentioned and create an issue and start off with?

[gatbontonp]

Hi @mvafin,

I picked up a few operations:
aten::stack
aten::vstack
aten::hstack
I also started writing aten::nan_to_num which would be combining the openvino operations IsNaN, IsInf, and Selects, but I see that @darksapien23151 also requested it.

[code-dev05]

Hi @mvafin
I would like to take on the aten::logit function for now.
It should be done using the log and subtract function.
Should I make a new issue for this function or just reference this issue in the pr?

[rtruong3775]

Hi @mvafin are there any functions left I can help contribute towards?

[Drazr]

Hi @mvafin can you assign any functions that are left unattended

[Mohamed-Ashraf273]

Hey @mvafin
I hope you're doing great! I've picked out some simple operations to get started with:
aten::lshift
aten::rshift

Can I contribute to this issue?

[carlozamu]

.take @mvafin
I would like to work on aten::rad2deg.
My plan is to implement it by applying a simple element-wise multiplication by the constant 180 ---- 𝜋 to convert radians to degrees, using the corresponding OpenVINO operations for multiplication and constant creation. Tell me if is ok to start!!

[mvafin]

.take @mvafin I would like to work on aten::rad2deg. My plan is to implement it by applying a simple element-wise multiplication by the constant 180 ---- 𝜋 to convert radians to degrees, using the corresponding OpenVINO operations for multiplication and constant creation. Tell me if is ok to start!!

Sounds, good. You can start, here is the issue for you #28873

[p-wysocki]

IMPORTANT

Hello everyone,

We decided to change the method of managing the operators subtask list, since this issue grew into chaos. From now on all the tasks are already created and are to be treated as regular Good First Issue tasks. The current Issue is now only for organization, please do not assign yourselves to it, do not add comments about picking tasks (general comments/questions about operators are of course welcome).

Please do not assign yourselves to this issue, do not create new issues, just choose a sub-issue you're interested in and pick it.

[Javier-lz]

Hello, where can I find the documentation for the Pytorch functions? I have found IR's and the Pytorch repository Pytorch, but most of the unassigned functions are either not in here, are not beginner friendly or the documentation doesn't seem to be right.

Thank you

[mvafin]

Hello, where can I find the documentation for the Pytorch functions? I have found IR's and the Pytorch repository Pytorch, but most of the unassigned functions are either not in here, are not beginner friendly or the documentation doesn't seem to be right.

Thank you

You are right. It isn't easy to find descriptions of those functions. Quite often you can replace aten:: with torch. in Python to find the required operation. Or some of them are just a methods of python types, like list, tuple: for example aten::clear, aten::pop.

My recommendation would be to make a model that have the operation that is suspected to be the one you are looking for and use torch.jit.trace and torch.jit.script on that model and check what inlined_graph of the resulting output looks like.

Sometimes it may be tricky to reproduce the operation in the model. For such cases you can ask for help in the ticket for this operation.

[saadkhi]

@mvafin this issue is assigned to me, can you brief me on which file I have to look upon to start the work.