infer.yaml

# ============================== DATA ================================
data:
  sr:                       16000
  chunk_sec:                30.0

# ============================== MODEL ================================
model:
  _target_:        livi.apps.audio_encoder.models.LiviAudioEncoder
  checkpoint_dir:  livi/apps/audio_encoder/checkpoints
  device:          cuda:0

  # Whisper encoder
  dim_whisper:         1280
  whisper_model_name: "openai/whisper-large-v3-turbo"
  compile:             false

  # Attention Pooling
  num_heads:       1
  mlp_ratio:       2.0
  qkv_bias:        false
  qk_scale:        null
  drop:            0.0
  attn_drop:       0.0
  init_scale:      1e-4

  # Projection head (audio)
  dim_embed:       768
  dim_hiddens:     [3072, 2048, 2048, 1536]