experiment:
    tokenizer_checkpoint: "tokenizer_titok_s128.bin"
    generator_checkpoint: "generator_titok_s128.bin"
    output_dir: "titok_s_128"
model:
    vq_model:
        codebook_size: 4096
        token_size: 12
        use_l2_norm: True
        commitment_cost: 0.25
        # vit arch
        vit_enc_model_size: "small"
        vit_dec_model_size: "small"
        vit_enc_patch_size: 16
        vit_dec_patch_size: 16
        num_latent_tokens: 128
        finetune_decoder: True
    
    generator:
        model_type: "UViT"
        hidden_size: 1024
        num_hidden_layers: 20
        num_attention_heads: 16
        intermediate_size: 4096
        dropout: 0.1
        attn_drop: 0.1
        num_steps: 64
        class_label_dropout: 0.1
        image_seq_len: ${model.vq_model.num_latent_tokens}
        condition_num_classes: 1000

        # sampling hyper-params
        randomize_temperature: 2.8
        guidance_scale: 6.9
        guidance_decay: "power-cosine"

dataset:
    preprocessing:
        crop_size: 256