transformer/config.json

{
  "_class_name": "NitroMMDiTModel",
  "_diffusers_version": "0.34.0",
  "attention_head_dim": 64,
  "caption_channels": 2048,
  "caption_projection_dim": 1152,
  "dual_attention_layers": [
    0,
    1,
    2,
    3,
    4,
    5,
    6,
    7,
    8,
    9,
    10,
    11,
    12
  ],
  "in_channels": 32,
  "interpolation_scale": 2,
  "joint_attention_dim": 4096,
  "num_attention_heads": 18,
  "num_layers": 24,
  "out_channels": 32,
  "patch_size": 1,
  "pooled_projection_dim": 2048,
  "pos_embed_max_size": 96,
  "projector_dim": 2048,
  "qk_norm": "rms_norm",
  "repa_depth": -1,
  "sample_size": 32,
  "z_dims": [
    768
  ]
}