{
  "model_type": "bernini",
  "architectures": ["BerniniModel"],
  "mllm_attn_implementation": "sdpa",
  "partial_pretrain_model": "",
  "base_dir": "ByteDance/Bernini-Diffusers",
  "vae_subfolder": "vae",
  "cotrain": true,
  "boundary_ratio": 0.417,
  "switch_dit_boundary": 0.875,
  "shift": 3.0,
  "logit_mean": 0.5,
  "logit_std": 1,
  "mode_scale": 1.29,
  "scratch": true,
  "transformers_version": "4.57.3",
  "mllm_subfolder": "mllm",
  "processor_subfolder": "mllm",
  "bernini_ckpt_subfolder": "bernini",
  "scratch_mllm": true,
  "use_src_id_rotary_emb": true,
  "feature_type_from_stage_one": "masked_tgt_embed_with_qwen_txt_vit_tokens",
  "num_mask_token": 4096,
  "max_sequence_length": 512,
  "clip_diff_cfg": {
    "model_type": "flow_match",
    "z_channels": 3584,
    "target_channels": 3584,
    "width": 4096,
    "diffusion_batch_mul": 16,
    "shift": 2.0
  },
  "connector_cfg": {
    "model_type": "MLPConnector",
    "out_dim_for_gen": 4096,
    "enable_gen_branch": true,
    "out_dim_for_vit": 3584,
    "enable_vit_branch": true,
    "gen_head_type": "zerolinear",
    "zero_init_proj_gen_last": true
  },
  "mask_ratio_infer_cfg": {
    "generator_type": "default"
  },
  "t5_max_sequence_length": 512,
  "t5_text_encoder_subfolder": "t5_text_encoder",
  "t5_tokenizer_subfolder": "t5_tokenizer",
  "t5_combine_type": "concat_with_zero_init",
  "target_fps": 16
}