|
|
|
codec_ckpt: ./models/vampnet/codec.pth |
|
save_path: ckpt |
|
|
|
num_iters: 1000000000 |
|
save_iters: [10000, 50000, 100000, 300000, 500000] |
|
val_idx: [0,1,2,3,4,5,6,7,8,9] |
|
sample_freq: 10000 |
|
val_freq: 1000 |
|
|
|
batch_size: 8 |
|
num_workers: 10 |
|
|
|
|
|
amp: false |
|
|
|
CrossEntropyLoss.label_smoothing: 0.1 |
|
|
|
AdamW.lr: 0.001 |
|
|
|
NoamScheduler.factor: 2.0 |
|
NoamScheduler.warmup: 10000 |
|
|
|
VampNet.vocab_size: 1024 |
|
VampNet.n_codebooks: 4 |
|
VampNet.n_conditioning_codebooks: 0 |
|
VampNet.r_cond_dim: 0 |
|
VampNet.noise_mode: mask |
|
VampNet.embedding_dim: 1280 |
|
VampNet.n_layers: 20 |
|
VampNet.n_heads: 20 |
|
VampNet.flash_attn: false |
|
VampNet.dropout: 0.1 |
|
|
|
AudioLoader.relative_path: "" |
|
AudioDataset.loudness_cutoff: -30.0 |
|
AudioDataset.without_replacement: true |
|
AudioLoader.shuffle: true |
|
|
|
AudioDataset.duration: 10.0 |
|
|
|
train/AudioDataset.n_examples: 10000000 |
|
train/AudioLoader.sources: |
|
- /media/CHONK/hugo/spotdl/audio-train |
|
|
|
val/AudioDataset.n_examples: 2000 |
|
val/AudioLoader.sources: |
|
- /media/CHONK/hugo/spotdl/audio-val |
|
|
|
|
