| name: flexTokenizer | |
| embedding_dim: 128 | |
| hidden_channels: 128 | |
| channel_multipliers: | |
| - 1 | |
| - 1 | |
| - 2 | |
| - 2 | |
| - 4 | |
| - 4 | |
| encoder_layer_configs: | |
| - 2 | |
| - 2 | |
| - 2 | |
| - 2 | |
| - 2 | |
| - 2 | |
| - 2 | |
| decoder_layer_configs: | |
| - 2 | |
| - 2 | |
| - 2 | |
| - 2 | |
| - 2 | |
| - 2 | |
| - 2 | |
| use_adaptive_norm: true | |
| use_learnable_up_down_sample: true | |
| quantizer_config: | |
| quantize_type: gvq | |
| embed_dim: 128 | |
| num_embed: 262144 | |
| commitment_loss_weight: 0.25 | |
| use_l2_norm: true | |
| use_uniform_init: false | |
| num_group: 8 | |