am i crazy or is it impossible to put inputs to cuda with asr pipeline?
if i do:
pipe = pipeline(‘automatic-speech-recognition’, device=‘cuda’)
and then run:
pipe(inputs = audio_array.numpy())
i get
RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same
i think the inputs are just never sent to the device in the code (https://github.com/huggingface/transformers/blob/v4.34.1/src/transformers/pipelines/automatic_speech_recognition.py#L294) …