lam/config.json

{
  "architectures": [
    "LAMModel"
  ],
  "encoder_height": 224,
  "encoder_width": 304,
  "fsq_levels": [
    7,
    5,
    5,
    5,
    5
  ],
  "initializer_range": 0.02,
  "is_diffusion": true,
  "latent_channels": 5,
  "max_tokens": 80,
  "min_tokens": 1,
  "model_type": "lam",
  "null_latent": 0,
  "num_tokens_in": 333,
  "pool_pick_tokens": 80,
  "torch_dtype": "float32",
  "transformer_config": {
    "_class_name": "SD3Transformer2DModel",
    "_diffusers_version": "0.31.0.dev0",
    "_name_or_path": "stabilityai/stable-diffusion-3.5-medium",
    "attention_head_dim": 64,
    "caption_projection_dim": 1536,
    "dual_attention_layers": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7,
      8,
      9,
      10,
      11,
      12
    ],
    "in_channels": 32,
    "joint_attention_dim": 5,
    "num_attention_heads": 24,
    "num_layers": 24,
    "out_channels": 16,
    "patch_size": 2,
    "pooled_projection_dim": 400,
    "pos_embed_max_size": 384,
    "qk_norm": "rms_norm",
    "sample_size": 128
  },
  "transformers_version": "4.50.0",
  "use_empty_prompt": true,
  "use_tail_drop": true,
  "videomae_from_pretrained": "OpenGVLab/VideoMAEv2-Large",
  "vocab_size": 4375
}