Upload 4 files

config/base_config.yaml

@@ -0,0 +1,70 @@
+version: "1.34"
+
+lr: 1e-4
+log_step: 1
+split: 0
+batch_size: 8
+sr: 16000
+
+datasets:
+  - jamendo
+  - emomusic
+  - pmemo
+  - deam
+
+model:
+  encoder: "MERT"
+  layers:
+    - 5
+    - 6
+  classifier: "linear-mt-attn-ck"
+  # - linear
+  # - linear-attn-ck
+  # - linear-mt-attn-ck
+
+  kd: True
+  kd_weight: 0.8
+  kd_temperature: 1
+  lr: 1e-4
+
+dataset:
+  jamendo:
+    root: './dataset/jamendo'
+    subset: 'moodtheme'
+    batch_size: 8
+    output_size : 56
+    split: 0
+    segment_type: "all" # [all,f10s,f30s,10s,30s]
+    num_workers: 4
+  deam:
+    root: './dataset/deam'
+    batch_size: 8
+    output_size : 2 
+    segment_type: "all" # [all,f10s,f30s,10s,30s]
+    num_workers: 4
+  pmemo:
+    root: './dataset/pmemo'
+    batch_size: 8
+    output_size : 2 
+    segment_type: "all" # [all,f10s,f30s,10s,30s]
+    num_workers: 4
+  emomusic:
+    root: './dataset/emomusic'
+    batch_size: 8
+    output_size : 2 
+    segment_type: "all" # [all,f10s,f30s,10s,30s]
+    num_workers: 4
+
+# --------------------------------------- #
+genre_class_size: 87
+mood_class_size: 56
+instr_class_size: 40
+dac_latents_size: 72
+dac_rvq_size: 9
+# --------------------------------------- #
+
+checkpoint_pmemo: "tb_logs/best/P.ckpt"
+checkpoint_deam: "tb_logs/best/D.ckpt"
+checkpoint_emomusic: "tb_logs/best/E.ckpt"
+checkpoint_jamendo: "tb_logs/best/J.ckpt"
+

config/prep_config.yaml

@@ -0,0 +1,16 @@
+device_id: 3
+
+is_split: True
+segment_duration: 30
+
+dataset:
+  input_dir: '../dataset/deam/mp3'
+  output_dir: '../dataset/deam/mert_30s'
+
+model: 
+  name: 'm-a-p/MERT-v1-95M'
+  sr: 24000
+
+
+
+

config/test_config.yaml

@@ -0,0 +1,16 @@
+defaults:
+  - base_config
+  - _self_
+
+batch_size: 8
+devices: [0]
+
+trainer:
+  devices: ${devices}
+  accelerator: 'gpu'
+
+checkpoint_latest: True
+
+multitask: True
+dataset_type: "va" 
+checkpoint: "tb_logs/best/J.ckpt"

config/train_config.yaml

@@ -0,0 +1,46 @@
+defaults:
+  - base_config
+  - _self_
+
+devices: [0,1,2,3]
+epochs: 500
+batch_size: 8
+
+monitor_metric: "val_loss"
+monitor_metric_mood: "val_loss_mood"
+monitor_metric_va: "val_loss_va"
+
+checkpoint:
+  monitor: "${monitor_metric}"
+  filename: "{epoch:02d}-{${monitor_metric}:.4f}"
+  save_top_k: 5
+  mode: "min"
+  auto_insert_metric_name: False
+  save_last: True
+
+checkpoint_mood:
+  monitor: "${monitor_metric_mood}"
+  filename: "mood-{epoch:02d}-{${monitor_metric_mood}:.4f}"
+  save_top_k: 5
+  mode: "min"
+  auto_insert_metric_name: False
+  save_last: True
+
+checkpoint_va:
+  monitor: "${monitor_metric_va}"
+  filename: "va-{epoch:02d}-{${monitor_metric_va}:.4f}"
+  save_top_k: 5
+  mode: "min"
+  auto_insert_metric_name: False
+  save_last: True
+
+earlystopping:
+  monitor: "${monitor_metric}"
+  patience: 10
+  min_delta: 0.0001
+  mode: "min"
+
+trainer:
+  devices: ${devices}
+  max_epochs: ${epochs}
+  accelerator: 'gpu'