{
  "_class_name": "TransformerMAE",
  "_diffusers_version": "0.24.0",
  "_name_or_path": "/shared/jiarui/results/video-diffusers/output/mae_p16_text_ed_enc_cat_n1024_arxiv_l400m_224_600k_bs1024x8/amp/hf_pipeline/mask_image_model",
  "cross_attention_dim": 1024,
  "decoder_cross_attention": true,
  "decoder_cross_attention_cat_encoder": true,
  "decoder_depth": 8,
  "decoder_embed_dim": 512,
  "decoder_num_heads": 16,
  "encoder_cross_attention": true,
  "encoder_cross_attention_cat_encoder": true,
  "encoder_depth": 24,
  "encoder_embed_dim": 1024,
  "encoder_num_heads": 16,
  "in_channels": 1024,
  "in_context_cross_attention": false,
  "num_embed": null,
  "out_channels": 1024,
  "sample_size": 14,
  "with_cls_token": true
}