#8246: Port Whisper Model functionality to n300 card (single device) with bs>1 support

[kkeerthana0573]

This PR contains :

1. Single Card demo implementation of ttnn_functional_whisper, ttnn_optimized_functional_whisper for Audio Classification and Conditional Generation supporting batch_size > 1.
2. Additionally, moved functional_whisper from models/experimental to models/demos/whisper.

[tt-rkim]

• Why did we move ttnn whisper into demos/grayskull, but run it in common models?
• Did you run nightly fast dispatch?

[kkeerthana0573]

@tt-rkim,

• The whisper model runs on both grayskull and wormhole, so the tests are added in run_common_models.sh.
  Since the whisper model runs on multiple devices, should we move the model to models/demo/ folder?

• CI links -

  • All post-commit tests
  • Nightly fast dispatch tests

[esmalTT]

@kkeerthana0573 Please update the description with the passing pipelines. Looks like this should run in nightly and the performance pipeline

[kkeerthana0573]

@esmalTT,
I've triggered CIs and will update the description with passing pipelines as soon as possible.
Nightly CI - passed for the model and failed for the other models - link.
Device Perf CI - timedout for the model on n300 and is working as expected locally. We're currently debugging this - link.
e2e perf CI - passed - link.

[tt-rkim]

Please post updates

And ping in slack if ready for reviedw

[tt-rkim]

This is not ready for review, per your lack of response.

@bkeith-TT @zzigler-tt Do not include this in the PR numbers for blocked by code review.

[kkeerthana0573]

@tt-rkim,
Device Perf test for this model hangs in CIs and locally.
Initially, the test was working as expected - here is the link for Device Perf CI when it was working as expected.
Later, it started hanging even before the test began - here is the link for latest Device Perf CI.

Please find the same issue reported here.
Thank you.

[tt-rkim]

I see, I will respond to your issue there.

For the purposes of this PR, can we take out the device perf test, and work on it in a separate PR? We should not be merging broken tests into main.

Otherwise you are blocking yourself from merging.