first commit

Amadeus/symbolic_yamls/config-accelerate.yaml

@@ -0,0 +1,65 @@
+defaults:
+  # - nn_params: nb8_embSum_NMT
+  # - nn_params: remi8
+  - nn_params: nb8_embSum_diff_t2m_150M_finetunning
+    # - nn_params: nb8_embSum_diff_t2m_150M_pretraining
+  # - nn_params: nb8_embSum_subPararell
+  # - nn_params: nb8_embSum_diff_t2m_150M
+
+    #  - nn_params: nb8_embSum_subFeedForward
+    # -  nn_params: nb8_embSum_diff
+  # nn_params: nb8_SA_diff
+  # - nn_params: nb8_embSum_diff_main12head16dim512_ave
+  # - nn_params: nb8_embSum_NMT_main12_head_16_dim512
+  # - nn_params: remi8_main12_head_16_dim512
+    # - nn_params: nb5_embSum_diff_main12head16dim768_sub3
+
+dataset: FinetuneDataset  # Pop1k7, Pop909, SOD, LakhClean,PretrainingDataset FinetuneDataset
+captions_path: dataset/midicaps/train_set.json
+
+# dataset:  SymphonyNet_Dataset # Pop1k7, Pop909, SOD, LakhClean
+# captions_path: dataset/symphonyNet/syd-caption.json
+
+use_ddp: True # True, False | distributed data parallel
+use_fp16: True # True, False | mixed precision training
+use_diff: True # True,use diffusion in subdecoder
+diff_steps: 8 # number of diffusion steps
+use_dispLoss: True
+lambda_weight: 0.5
+tau: 0.5
+
+train_params:
+  device: cuda
+  batch_size: 3
+  grad_clip: 1.0
+  num_iter: 300000  # total number of iterations
+  num_cycles_for_inference: 10 # number of cycles for inference, iterations_per_validation_cycle * num_cycles_for_inference
+  num_cycles_for_model_checkpoint: 1 # number of cycles for model checkpoint, iterations_per_validation_cycle * num_cycles_for_model_checkpoint
+  iterations_per_training_cycle: 10 # number of iterations for logging training loss
+  iterations_per_validation_cycle: 5000 # number of iterations for validation process
+  input_length: 3072 # input sequence length3072
+  # you can use focal loss, it it's not used, set focal_gamma to 0
+  focal_alpha: 1
+  focal_gamma: 0
+  # learning rate scheduler: 'cosinelr', 'cosineannealingwarmuprestarts', 'not-using', please check train_utils.py for more details
+  scheduler : cosinelr
+  initial_lr: 0.00005
+  decay_step_rate: 0.8 # means it will reach its lowest point at decay_step_rate * total_num_iter
+  num_steps_per_cycle: 20000 # number of steps per cycle for 'cosineannealingwarmuprestarts'
+  warmup_steps: 2000 #number of warmup steps
+  max_lr: 0.00015
+  gamma: 0.6 # the decay rate for 'cosineannealingwarmuprestarts'
+  # Distributed Data Parallel
+  world_size: 5 # 0 means no distributed training
+  gradient_accumulation_steps: 4  # 1 means no gradient accumulation
+inference_params:
+  num_uncond_generation: 1 # number of unconditional generation
+  num_cond_generation: 3 # number of conditional generation
+data_params:
+  first_pred_feature: pitch # compound shifting for NB only, choose the target sub-token (remi and cp are not influenced by this argument)
+  split_ratio: 0.998 # train-validation-test split ratio
+  aug_type: pitch # random, null | pitch and chord augmentation type
+general:
+  debug: False
+  make_log: True # True, False | update the log file in wandb online to your designated project and entity
+  infer_and_log: True # True, False | inference and log the results

Amadeus/symbolic_yamls/config.yaml

@@ -0,0 +1,54 @@
+defaults:
+  # - nn_params: nb8_embSum_NMT
+  # - nn_params: remi8
+  # - nn_params: nb8_embSum_diff
+   - nn_params: nb8_embSum_subFeedForward
+  #  - nn_params: nb8_SA_diff
+  # - nn_params: nb8_embSum_diff_main12head16dim512_ave
+  # - nn_params: nb8_embSum_NMT_main12_head_16_dim512
+  # - nn_params: remi8_main12_head_16_dim512
+    # - nn_params: nb5_embSum_diff_main12head16dim768_sub3
+
+dataset:  LakhClean # Pop1k7, Pop909, SOD, LakhClean
+use_ddp: True # True, False | distributed data parallel
+use_fp16: True # True, False | mixed precision training
+use_diff: True # True,use diffusion in subdecoder
+use_dispLoss: True
+lambda_weight: 0.5
+tau: 0.5
+diff_steps: 8 # number of diffusion steps
+train_params:
+  device: cuda
+  batch_size: 8
+  grad_clip: 1.0
+  num_iter: 25000  # total number of iterations
+  num_cycles_for_inference: 10 # number of cycles for inference, iterations_per_validation_cycle * num_cycles_for_inference
+  num_cycles_for_model_checkpoint: 10 # number of cycles for model checkpoint, iterations_per_validation_cycle * num_cycles_for_model_checkpoint
+  iterations_per_training_cycle: 10 # number of iterations for logging training loss
+  iterations_per_validation_cycle: 500 # number of iterations for validation process
+  input_length: 3072 # input sequence length3072
+  # you can use focal loss, it it's not used, set focal_gamma to 0
+  focal_alpha: 1
+  focal_gamma: 0
+  # learning rate scheduler: 'cosinelr', 'cosineannealingwarmuprestarts', 'not-using', please check train_utils.py for more details
+  scheduler : cosinelr
+  initial_lr: 0.0001
+  decay_step_rate: 0.4 # means it will reach its lowest point at decay_step_rate * total_num_iter
+  num_steps_per_cycle: 20000 # number of steps per cycle for 'cosineannealingwarmuprestarts'
+  warmup_steps: 2000 # number of warmup steps
+  max_lr: 0.00015 
+  gamma: 0.6 # the decay rate for 'cosineannealingwarmuprestarts'
+  # Distributed Data Parallel
+  world_size: 5 # 0 means no distributed training
+  gradient_accumulation_steps: 1 # 1 means no gradient accumulation
+inference_params:
+  num_uncond_generation: 1 # number of unconditional generation
+  num_cond_generation: 3 # number of conditional generation
+data_params:
+  first_pred_feature: pitch # compound shifting for NB only, choose the target sub-token (remi and cp are not influenced by this argument)
+  split_ratio: 0.99 # train-validation-test split ratio
+  aug_type: null # random, null | pitch and chord augmentation type
+general:
+  debug: False
+  make_log: True # True, False | update the log file in wandb online to your designated project and entity
+  infer_and_log: True # True, False | inference and log the results

Amadeus/symbolic_yamls/nn_params/cp5_embSum_NMT.yaml

@@ -0,0 +1,20 @@
+encoding_scheme: cp
+num_features: 5
+vocab_name: MusicTokenVocabCP
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: CrossAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  input_length: 1024
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: True

Amadeus/symbolic_yamls/nn_params/cp5_embSum_subCrossAttention.yaml

@@ -0,0 +1,20 @@
+encoding_scheme: cp
+num_features: 5
+vocab_name: MusicTokenVocabCP
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: CrossAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  input_length: 1024
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/cp5_embSum_subFeedForward.yaml

@@ -0,0 +1,18 @@
+encoding_scheme: cp
+num_features: 5
+vocab_name: MusicTokenVocabCP
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: FeedForward
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/cp5_embSum_subFeedForward_original.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: cp
+num_features: 5
+vocab_name: MusicTokenVocabCP
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: FeedForward
+model_dropout: 0.1
+partial_sequential_prediction: True
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/cp7_embSum_NMT.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: cp
+num_features: 7
+vocab_name: MusicTokenVocabCP
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: CrossAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: True

Amadeus/symbolic_yamls/nn_params/cp7_embSum_subCrossAttention.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: cp
+num_features: 7
+vocab_name: MusicTokenVocabCP
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: CrossAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/cp7_embSum_subFeedForward.yaml

@@ -0,0 +1,18 @@
+encoding_scheme: cp
+num_features: 7
+vocab_name: MusicTokenVocabCP
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: FeedForward
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/cp7_embSum_subFeedForward_original.yaml

@@ -0,0 +1,20 @@
+encoding_scheme: cp
+num_features: 7
+vocab_name: MusicTokenVocabCP
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: FeedForward
+model_dropout: 0.1
+partial_sequential_prediction: True
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  input_length: 1024
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/nb5_embSum_NMT.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 5
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: CrossAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: True

Amadeus/symbolic_yamls/nn_params/nb5_embSum_diff.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 5
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name:  SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: True

Amadeus/symbolic_yamls/nn_params/nb5_embSum_diff_main12head16dim512_sub3.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 5
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 12
+  num_head: 16
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 3
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb5_embSum_diff_main12head16dim768_sub3.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 5
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 768
+  num_layer: 12
+  num_head: 16
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 3
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb5_embSum_subCrossAttention.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 5
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: CrossAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb5_embSum_subFeedForward.yaml

@@ -0,0 +1,18 @@
+encoding_scheme: nb
+num_features: 5
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: FeedForward
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/nb5_embSum_subPararell.yaml

@@ -0,0 +1,18 @@
+encoding_scheme: nb
+num_features: 5
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: Parallel
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/nb5_embSum_subRNN.yaml

@@ -0,0 +1,18 @@
+encoding_scheme: nb
+num_features: 5
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: RNN
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/nb5_embSum_subSelfAttention.yaml

@@ -0,0 +1,18 @@
+encoding_scheme: nb
+num_features: 5
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: SelfAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/nb7_embSum_NMT.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 7
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: CrossAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: True

Amadeus/symbolic_yamls/nn_params/nb7_embSum_subCrossAttention.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 7
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: CrossAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb7_embSum_subFeedForward.yaml

@@ -0,0 +1,18 @@
+encoding_scheme: nb
+num_features: 7
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: FeedForward
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/nb7_embSum_subPararell.yaml

@@ -0,0 +1,18 @@
+encoding_scheme: nb
+num_features: 7
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: Parallel
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/nb7_embSum_subRNN.yaml

@@ -0,0 +1,18 @@
+encoding_scheme: nb
+num_features: 7
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: RNN
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/nb7_embSum_subSelfAttention.yaml

@@ -0,0 +1,18 @@
+encoding_scheme: nb
+num_features: 7
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: SelfAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/nb8_SA_diff.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name:  SelfAttentionEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb8_embSum_NMT.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: CrossAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: True

Amadeus/symbolic_yamls/nn_params/nb8_embSum_NMT_main12_head_16_dim512.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: CrossAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 12
+  num_head: 16
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: True

Amadeus/symbolic_yamls/nn_params/nb8_embSum_NMT_main12_head_16_dim512_sub3.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: CrossAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 12
+  num_head: 16
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 3
+  feature_enricher_use: True

Amadeus/symbolic_yamls/nn_params/nb8_embSum_NMTsub6.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: CrossAttention
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 6
+  feature_enricher_use: True

Amadeus/symbolic_yamls/nn_params/nb8_embSum_diff.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name:  SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb8_embSum_diff_150M.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name:  SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.2
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 768
+  num_layer: 16
+  num_head: 12
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb8_embSum_diff_main12head16dim512.yaml

@@ -0,0 +1,20 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 12
+  num_head: 16
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: True
+  

Amadeus/symbolic_yamls/nn_params/nb8_embSum_diff_main12head16dim512_ave.yaml

@@ -0,0 +1,20 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: AverageEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 12
+  num_head: 16
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: True
+  

Amadeus/symbolic_yamls/nn_params/nb8_embSum_diff_main12head16dim512_sub3.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 12
+  num_head: 16
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 3
+  feature_enricher_use: True

Amadeus/symbolic_yamls/nn_params/nb8_embSum_diff_sub3.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name:  SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 2
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb8_embSum_diff_sub6.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name:  SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 6
+  feature_enricher_use: True

Amadeus/symbolic_yamls/nn_params/nb8_embSum_diff_t2m_150M.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name:  SummationEmbedder
+main_decoder_name: XtransformerCrossAttendDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.2
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 768
+  num_layer: 16
+  num_head: 12
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb8_embSum_diff_t2m_150M_finetunning.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name:  SummationEmbedder
+main_decoder_name: XtransformerFinetuningDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.2
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 768
+  num_layer: 20
+  num_head: 12
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb8_embSum_diff_t2m_150M_prefix.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name:  SummationEmbedder
+main_decoder_name: XtransformerPrefixDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 768
+  num_layer: 16
+  num_head: 12
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb8_embSum_diff_t2m_150M_pretraining.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name:  SummationEmbedder
+main_decoder_name: XtransformerPretrainingDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 768
+  num_layer: 20
+  num_head: 12
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb8_embSum_diff_t2m_30M.yaml

@@ -0,0 +1,19 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name:  SummationEmbedder
+main_decoder_name: XtransformerCrossAttendDecoder
+sub_decoder_name: DiffusionDecoder
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1
+  feature_enricher_use: False

Amadeus/symbolic_yamls/nn_params/nb8_embSum_subFeedForward.yaml

@@ -0,0 +1,18 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: FeedForward
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/nb8_embSum_subPararell.yaml

@@ -0,0 +1,18 @@
+encoding_scheme: nb
+num_features: 8
+vocab_name: MusicTokenVocabNB
+model_name: NestedMusicTransformer
+input_embedder_name: SummationEmbedder
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: Parallel
+model_dropout: 0.1
+input_embedder:
+  num_layer: 1
+  num_head: 8
+main_decoder:
+  dim_model: 512
+  num_layer: 6
+  num_head: 8
+sub_decoder:
+  decout_window_size: 1 # 1 means no previous decoding output added
+  num_layer: 1

Amadeus/symbolic_yamls/nn_params/remi5.yaml

@@ -0,0 +1,12 @@
+encoding_scheme: remi
+num_features: 5
+vocab_name: LangTokenVocab
+model_name: NestedMusicTransformer
+input_embedder_name: SingleEmbedding
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: SingleProjection
+model_dropout: 0.1
+main_decoder:
+  dim_model: 512
+  num_layer: 8
+  num_head: 8

Amadeus/symbolic_yamls/nn_params/remi7.yaml

@@ -0,0 +1,12 @@
+encoding_scheme: remi
+num_features: 7
+vocab_name: LangTokenVocab
+model_name: NestedMusicTransformer
+input_embedder_name: SingleEmbedding
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: SingleProjection
+model_dropout: 0.1
+main_decoder:
+  dim_model: 512
+  num_layer: 8
+  num_head: 8

Amadeus/symbolic_yamls/nn_params/remi8.yaml

@@ -0,0 +1,12 @@
+encoding_scheme: remi
+num_features: 8
+vocab_name: LangTokenVocab
+model_name: NestedMusicTransformer
+input_embedder_name: SingleEmbedding
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: SingleProjection
+model_dropout: 0.1
+main_decoder:
+  dim_model: 512
+  num_layer: 8
+  num_head: 8

Amadeus/symbolic_yamls/nn_params/remi8_main12_head_16_dim512.yaml

@@ -0,0 +1,12 @@
+encoding_scheme: remi
+num_features: 8
+vocab_name: LangTokenVocab
+model_name: NestedMusicTransformer
+input_embedder_name: SingleEmbedding
+main_decoder_name: XtransformerDecoder
+sub_decoder_name: SingleProjection
+model_dropout: 0.1
+main_decoder:
+  dim_model: 512
+  num_layer: 12
+  num_head: 16

Amadeus/symbolic_yamls/symbolic_sweep.yaml

@@ -0,0 +1,17 @@
+program: train.py
+method: grid
+metric:
+  name: valid.total
+  goal: minimize
+parameters:
+  train_params.batch_size:
+    values: [8]
+  train_params.focal_gamma:
+    values: [0, 1]
+  nn_params.main_decoder.input_length:
+    values: [8192]
+
+command:
+  - python3
+  - ${program}
+  - ${args_no_hyphens}