sd3_transformer/config.json

{
  "_class_name": "SD3Transformer2DModelWithSigLIP",
  "_diffusers_version": "0.36.0.dev0",
  "_name_or_path": "",
  "add_siglip_tensor_to_double_stream": false,
  "add_siglip_tensor_to_latent": true,
  "attention_head_dim": 64,
  "bsq_flip_level": "per_bit",
  "bsq_flip_prob": 0.3,
  "bsq_flip_strength": 0.0,
  "caption_projection_dim": 1536,
  "drop_image_token_prob_for_cfg": 0.1,
  "dual_attention_layers": [
    0,
    1,
    2,
    3,
    4,
    5,
    6,
    7,
    8,
    9,
    10,
    11,
    12
  ],
  "image_token_scale_factor": 1.0,
  "in_channels": 16,
  "joint_attention_dim": 4096,
  "num_attention_heads": 24,
  "num_layers": 24,
  "out_channels": 16,
  "patch_size": 2,
  "pooled_projection_dim": 2048,
  "pos_embed_max_size": 384,
  "qk_norm": "rms_norm",
  "random_crop_pos_embed_max_resolution": 1328,
  "ref_image_enabled": false,
  "sample_size": 128,
  "siglip_channels": 4608,
  "super_resolution": true,
  "train_bsq_output_proj": false,
  "train_use_random_crop_pos_embed": true,
  "upscale_factor": 1,
  "upscale_factor_candidates": [
    1.0,
    1.0667,
    2.0
  ],
  "upscale_factor_max": 2.0,
  "upscale_type": "interpolate"
}