Audio encodings now match conv2d weight dtype in Gemma3nAudioSSCPConvBlock

[Malav-P]

What does this PR do?

The change ensures that the audio encodings are cast to the same type as conv.weight tensor. This is done by appending to(self.conv.weight.dtype) to the audio_encodings_padded rvalue.

Fixes an issue where the conv2d forward pass throws an error. Code for reproducible error (run on Mac M1) :

from transformers import pipeline
import torch

pipe = pipeline(
    "image-text-to-text",
    model="google/gemma-3n-e2b-it",
    device=0,
    torch_dtype=torch.bfloat16,
    cache_implementation="static"
)


messages = [
    {
        "role": "user",
        "content": [
            {"type": "audio", "audio": "5676.wav"},
            {"type": "text", "text": "Transcribe this audio file."}
        ]
    }
]

output = pipe(text=messages, max_new_tokens=200, torch_dtype=torch.bfloat16)
print(output[0]["generated_text"][-1]["content"])

Full stack trace is attached error.txt.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Suggested Reviewers

@ArthurZucker

[Rocketknight1]

cc @eustlb

[ArthurZucker]

This makes sense, happy to merge!

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: gemma3n

[Malav-P]

This makes sense, happy to merge!

Awesome, ty!