Giving dictionary to Audio Classification Pipeline overwrites dictionary data

[wilke0818]

System Info

• transformers version: 4.48.0
• Platform: macOS-14.6-arm64-arm-64bit
• Python version: 3.12.4
• Huggingface_hub version: 0.27.1
• Safetensors version: 0.5.2
• Accelerate version: 1.2.1
• Accelerate config: not found
• PyTorch version (GPU?): 2.5.1 (False)
• Tensorflow version (GPU?): not installed (NA)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Using distributed or parallel set-up in script?: None/NA

Who can help?

@Rocketknight1

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

from transformers import pipeline
import torch
import numpy as np

model_name = 'audeering/wav2vec2-large-robust-12-ft-emotion-msp-dim'
#model_name = 'pollner/distilhubert-finetuned-ravdess'
top_k = 5
device = 'cuda' if torch.cuda.is_available() else 'cpu'
classification_pipeline = pipeline(
                "audio-classification",
                model=model_name,
                # top_k=top_k, 
                # function_to_apply='none',
                device=device,
            )

# dummy signal
sampling_rate = 16000
signal = np.zeros((sampling_rate), dtype=np.float32)

input_signal = {
    'array': signal,
    'sampling_rate': sampling_rate
}
print(input_signal)
print(classification_pipeline(input_signal))
print(input_signal)

Expected behavior

I would imagine, but could be wrong, that the internal logic should not be mutating inputs, especially because this occurs when inputting a dictionary but not when just inputting a numpy array.

[Rocketknight1]

Yes, thank you for the report! A fix is open at #35754

[Rocketknight1]

@wilke0818 a fix has been merged! Can you install the latest version from main and confirm that it fixes the issue for you?