transformer/config.json

{
  "_class_name": "SD3Transformer2DModelWithAudioHQ",
  "_diffusers_version": "0.30.0.dev0",
  "_name_or_path": "/localhome/data/ckpts/shared/stable-diffusion-3.5-medium",
  "add_audio": true,
  "add_clip": false,
  "attention_head_dim": 64,
  "audio_input_dim": 8,
  "caption_projection_dim": 768,
  "decoder_config": "",
  "drop_audio": false,
  "drop_image": false,
  "drop_text": false,
  "dual_attention_layers": [
    0,
    1,
    2,
    3,
    4,
    5
  ],
  "image_bind": false,
  "in_channels": 16,
  "num_clusters": 16,
  "joint_attention_dim": 768,
  "num_attention_heads": 12,
  "num_layers": 12,
  "out_channels": 16,
  "patch_size": 2,
  "pooled_projection_dim": 768,
  "pos_embed_max_size": 384,
  "qk_norm": "rms_norm",
  "sample_size": 128,
  "use_audio_mae": false,
  "text_out_dim": 128,
  "wsi_input_dim": 768,
  "text_vae_path": "/home/user/wangtao/OmniFlows/patho/checkpoints/ae_Qwen3-0.6B_ntoken64_vaedim192_run_stage2/best_model"
}