batch_size: 24
pretrained: default
embedded_channels: 256
embedded_type: contentvec