| ## FP32 | |
| ```python | |
| # !pip install git+https://github.com/huggingface/diffusers.git | |
| from diffusers import DiffusionPipeline | |
| import scipy | |
| model_id = "harmonai/unlocked-250k" | |
| pipeline = DiffusionPipeline.from_pretrained(model_id) | |
| audio = pipeline(audio_length_in_s=4.0).audios | |
| scipy.io.wavfile.write("maestro_test.wav", pipe.unet.sample_rate, audios) | |
| ``` | |
| ## FP16 | |
| Faster at a small loss of quality | |
| ```python | |
| # !pip install git+https://github.com/huggingface/diffusers.git | |
| from diffusers import DiffusionPipeline | |
| import scipy | |
| import torch | |
| model_id = "harmonai/unlocked-250k" | |
| pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16) | |
| audio = pipeline(audio_length_in_s=4.0).audios | |
| scipy.io.wavfile.write("maestro_test.wav", pipe.unet.sample_rate, audios) | |
| ``` |