lingyu123-su/MIDIFoundationModel
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

1029 add octuple

Browse Source
This commit is contained in:
Mars
2025-10-29 17:14:33 +08:00
parent b493ede479
commit e16c84aab2
22 changed files with 1135 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
Amadeus/evaluation_utils.py
View File

@ -96,7 +96,7 @@ def prepare_model_and_dataset_from_config(config: DictConfig, metadata_path:str,
num_features = config.nn_params.num_features
# get vocab
vocab_name = {'remi':'LangTokenVocab', 'cp':'MusicTokenVocabCP', 'nb':'MusicTokenVocabNB'}
vocab_name = {'remi':'LangTokenVocab', 'cp':'MusicTokenVocabCP', 'nb':'MusicTokenVocabNB', 'oct':'MusicTokenVocabOct'}
selected_vocab_name = vocab_name[encoding_scheme]
vocab = getattr(vocab_utils, selected_vocab_name)(
@ -477,7 +477,7 @@ class Evaluator:
except KeyError:
in_beat_resolution = 4 # Default resolution if dataset is not found
midi_decoder_dict = {'remi':'MidiDecoder4REMI', 'cp':'MidiDecoder4CP', 'nb':'MidiDecoder4NB'}
midi_decoder_dict = {'remi':'MidiDecoder4REMI', 'cp':'MidiDecoder4CP', 'nb':'MidiDecoder4NB', 'oct':'MidiDecoder4Octuple'}
decoder_name = midi_decoder_dict[encoding_scheme]
decoder = getattr(decoding_utils, decoder_name)(vocab=self.vocab, in_beat_resolution=in_beat_resolution, dataset_name=self.config.dataset)
@ -499,7 +499,7 @@ class Evaluator:
except KeyError:
in_beat_resolution = 4 # Default resolution if dataset is not found
midi_decoder_dict = {'remi':'MidiDecoder4REMI', 'cp':'MidiDecoder4CP', 'nb':'MidiDecoder4NB'}
midi_decoder_dict = {'remi':'MidiDecoder4REMI', 'cp':'MidiDecoder4CP', 'nb':'MidiDecoder4NB', 'oct':'MidiDecoder4Octuple'}
decoder_name = midi_decoder_dict[encoding_scheme]
decoder = getattr(decoding_utils, decoder_name)(vocab=self.vocab, in_beat_resolution=in_beat_resolution, dataset_name=self.config.dataset)
@ -509,7 +509,7 @@ class Evaluator:
generated_sample = reverse_shift_and_pad_for_tensor(generated_sample, first_pred_feature)
decoder(generated_sample, output_path=str(save_dir / f"{tune_name}.mid"))
prompt = self.model.decoder._prepare_inference(self.model.decoder.net.start_token, 0, tuneidx, num_target_measures=8)
prompt = self.model.decoder._prepare_inference(self.model.decoder.net.start_token, 0, tuneidx, num_target_measures=num_target_measures)
decoder(prompt, output_path=str(save_dir / f"{tune_name}_prompt.mid"))
def generate_samples_unconditioned(self, save_dir, num_samples, first_pred_feature, sampling_method, threshold, temperature, generation_length=3072,uid=1):
@ -520,7 +520,7 @@ class Evaluator:
in_beat_resolution = in_beat_resolution_dict[self.config.dataset]
except KeyError:
in_beat_resolution = 4 # Default resolution if dataset is not found
midi_decoder_dict = {'remi':'MidiDecoder4REMI', 'cp':'MidiDecoder4CP', 'nb':'MidiDecoder4NB'}
midi_decoder_dict = {'remi':'MidiDecoder4REMI', 'cp':'MidiDecoder4CP', 'nb':'MidiDecoder4NB', 'oct':'MidiDecoder4Octuple'}
decoder_name = midi_decoder_dict[encoding_scheme]
decoder = getattr(decoding_utils, decoder_name)(vocab=self.vocab, in_beat_resolution=in_beat_resolution, dataset_name=self.config.dataset)