sortformer-train-1777567093/config.yaml

seed: 42
model:
  init_checkpoint: arda-argmax/diar_streaming_sortformer_8spk-init
  num_spks: 8
data:
  cuts_path: null
  dataset_hf_repo: null
  split: null
  cuts_filename: null
  preprocess: true
  audio_root: null
  session_len_sec: 120
  sample_rate: 16000
  batch_size: 2
  num_workers: 4
  shuffle: true
  streaming:
    enabled: true
    repo: arda-argmax/simchoir-parquet
    subset: v0.2
    train_split: train
    val_split: val
    train_shuffle_buffer: 0
    sampler_shuffle_buffer: 1000
    num_train_samples: 9500
    bucketing:
      enabled: false
      max_duration_sec: null
      num_buckets: 10
      buffer_size: 256
      quadratic_duration: 30.0
      duration_bins: []
      bin_estimate_samples: 10000
optim:
  learning_rate: 5.0e-05
  weight_decay: 0.0
  warmup_steps: 50
  min_lr: 1.0e-06
trainer:
  accelerator: gpu
  devices: auto
  precision: bf16-mixed
  fallback: false
  num_steps: -1
  num_epochs: 10
  val_check_interval: 0.25
  check_val_every_n_epoch: 1
  log_every_n_steps: 1
  limit_val_batches: 5
  num_sanity_val_steps: 0
  accumulate_grad_batches: 4
  strategy: auto
  limit_train_batches: 590
regularization:
  disable_all_dropout: false
  causal_attn_rate: null
  scores_add_rnd: null
output:
  final_checkpoint_save_path: /app/training_runs/1777567093/final.nemo
  log_dir: /app/training_runs/1777567093/logs
  run_name: sortformer-train-1777567093
tracking:
  enabled: true
  project: sortformer-robustness
  run_name: sortformer-train-1777567093
  group: null
  space_id: null
  auto_log_gpu: true
  gpu_log_interval: 10.0
  console_log:
    enabled: true
    max_chars: 200000
    every_n_train_steps: 0
val_der:
  enabled: true
  collar: 0.0
  skip_overlap: false
  threshold: 0.5
  max_samples: null
checkpointing:
  enabled: true
  repo: arda-argmax/sortformer-finetuned
  run_subdir: sortformer-train-1777567093
  monitor: val_der
  mode: min
  private: true
  local_dir: null
  flush_timeout_sec: 1800