Passing keyword arguments
In this exercise, you will learn how to pass keyword arguments to model functions when used inside a pipeline.
For this you will be using the MusicGen small model from Meta, which is capable of generating music samples based on text descriptions or audio prompts.
The pipeline module has been loaded, and the soundfile library is available as sf.
Deze oefening maakt deel uit van de cursus
Multi-Modal Models with Hugging Face
Oefeninstructies
- Load a
text-to-audiopipeline using thefacebook/musicgen-smallmodel in the PyTorch framework. - Make a dictionary called
generate_kwargsto set the generation temperature to0.8andmax_new_tokensto1. - Generate an audio array using the
"Classic rock riff"prompt, adjusting the generation parameters with yourgenerate_kwargsdictionary.
Praktische interactieve oefening
Probeer deze oefening eens door deze voorbeeldcode in te vullen.
# Load a text-to-audio pipeline
musicgen = pipeline(task="____", model="____", framework="pt")
# Make a dictionary to set the generation temperature to 0.8 and max_new_tokens to 1
generate_kwargs = {"____": ____, "____": ____}
# Generate an audio array passing the arguments
outputs = ____("____", ____=____)
sf.write("output.wav", outputs["audio"][0][0], outputs["sampling_rate"])