| patch_size: 40 | |
| n_patches: 600 | |
| num_classes: 1000 | |
| n_code: 8192 | |
| code_dim: 128 | |
| embed_dim: 40 | |
| weight_decay_finetuning: 1e-2 | |
| warmup_epochs_finetuning: 4 | |
| lr_finetuning: 5e-4 | |
| layer_decay_finetuning: 0.975 | |
| train_head_only_finetuning: False | |
| batch_size_finetuning: 32 | |
| epoch_finetuning: 20 | |
| model_name: NeuroRVQ | |
| in_chans_encoder: 1 | |
| out_chans_encoder: 8 | |
| depth_encoder: 12 | |
| depth_decoder: 3 | |
| decoder_out_dim: 40 | |
| num_heads_tokenizer: 10 | |
| mlp_ratio_tokenizer: 4 | |
| qkv_bias_tokenizer: True | |
| drop_rate_tokenizer: 0. | |
| attn_drop_rate_tokenizer: 0. | |
| drop_path_rate_tokenizer: 0. | |
| init_values_tokenizer: 0. | |
| init_scale_tokenizer: 0.001 | |
| use_for_pretraining: True | |
| in_chans_second_stage: 1 | |
| out_chans_second_stage: 8 | |
| depth_second_stage: 12 | |
| num_heads_second_stage: 10 | |
| mlp_ratio_second_stage: 4. | |
| qkv_bias_second_stage: True | |
| drop_rate_second_stage: 0. | |
| attn_drop_rate_second_stage: 0. | |
| drop_path_rate_second_stage: 0. | |
| init_values_second_stage: 1.e-5 | |
| init_scale_second_stage: 0.001 | |
| embed_dim_second_stage: 40 | |