diffusion_transformer_384p/config.json

{
  "_class_name": "PyramidDiffusionMMDiT",
  "_diffusers_version": "0.30.0",
  "attention_head_dim": 64,
  "caption_projection_dim": 1536,
  "in_channels": 16,
  "joint_attention_dim": 4096,
  "max_num_frames": 200,
  "num_attention_heads": 24,
  "num_layers": 24,
  "patch_size": 2,
  "pooled_projection_dim": 2048,
  "pos_embed_max_size": 192,
  "pos_embed_type": "sincos",
  "qk_norm": "rms_norm",
  "sample_size": 128,
  "use_flash_attn": false,
  "use_gradient_checkpointing": false,
  "use_temporal_causal": true
}