Adding a script to test accuracy on IPU provider after quantization with Quark

[hanlin0628]

No description provided.

[mht-sharma]

Hi @hanlin0628,

I have been working on integrating the Quark quantizer into optimum-amd through this pull request: #149. I noticed a few things and would like to clarify the following:

• Is the vaiq_onnx quantiser deprecated?

• The Quark quantiser used in this PR appears to be different from the Quark torch quantizer and seems to be ONNX-based (as per Quark documentation). Do we have two different sources for the Quark quantiser?

If the interfaces are similar, I would like to discuss how we can better integrate them into a single OptimumQuantizer.

[hanlin0628]

Hi Mohit The code of vai_q_onnx quantiser has been merged into Quark, so the Quark now has two quantisers, quark.torch and quark.onnx. Regards, Han

…

On Fri, Jul 26, 2024 at 2:10 PM Mohit Sharma ***@***.***> wrote: Hi @hanlin0628 <https://github.com/hanlin0628>, I have been working on integrating the Quark quantizer into optimum-amd through this pull request: #149 <#149>. I noticed a few things and would like to clarify the following: - Is the vaiq_onnx quantiser deprecated? - The Quark quantiser used in this PR appears to be different from the Quark torch quantizer and seems to be ONNX-based (as per Quark documentation <https://quark.docs.amd.com/latest/quark_torch_main_gen.html>). Do we have two different sources for the Quark quantiser? If the interfaces are similar, I would like to discuss how we can better integrate them into a single OptimumQuantizer. — Reply to this email directly, view it on GitHub <#148 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB2TU6WBCDVF6HV7BUHERLLZOHR55AVCNFSM6AAAAABK7UVANSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENJSGA2DGNBVG4> . You are receiving this because you were mentioned.Message ID: ***@***.***>

-- Han Lin

[hanlin0628]

Hi Mohit @mht-sharma

Could you please review the recent update and help me to merge?

Thanks,
Han