The predictions are currently not working on the GPU because we don't move the input tensors to the proper device.
This :
# Predict a song
from musiclang_predict import predict, MusicLangTokenizer
from transformers import GPT2LMHeadModel
model = GPT2LMHeadModel.from_pretrained('musiclang/musiclang-4k').to('cuda')
tokenizer = MusicLangTokenizer('musiclang/musiclang-4k')
# Predict a 8 bar song with this model
soundtrack = predict(model, tokenizer, chord_duration=4, nb_chords=8)
soundtrack.to_midi('song.mid', tempo=120, time_signature=(4, 4))
play_music()
raises :
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument index in method wrapper_CUDA__index_select)
The predictions are currently not working on the GPU because we don't move the input tensors to the proper device.
This :
raises :
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument index in method wrapper_CUDA__index_select)