vae/config.json

{
  "_class_name": "AutoencoderKLCogVideoX",
  "_diffusers_version": "0.33.1",
  "_name_or_path": "/root/.cache/huggingface/hub/models--THUDM--CogVideoX-5b/snapshots/8fc5b281006c82b82d34fd2543d2f0ebb4e7e321/vae",
  "act_fn": "silu",
  "block_out_channels": [
    128,
    256,
    256,
    512
  ],
  "down_block_types": [
    "CogVideoXDownBlock3D",
    "CogVideoXDownBlock3D",
    "CogVideoXDownBlock3D",
    "CogVideoXDownBlock3D"
  ],
  "force_upcast": true,
  "in_channels": 3,
  "invert_scale_latents": false,
  "latent_channels": 16,
  "latents_mean": null,
  "latents_std": null,
  "layers_per_block": 3,
  "norm_eps": 1e-06,
  "norm_num_groups": 32,
  "out_channels": 3,
  "sample_height": 480,
  "sample_width": 720,
  "scaling_factor": 0.7,
  "shift_factor": null,
  "temporal_compression_ratio": 4,
  "up_block_types": [
    "CogVideoXUpBlock3D",
    "CogVideoXUpBlock3D",
    "CogVideoXUpBlock3D",
    "CogVideoXUpBlock3D"
  ],
  "use_post_quant_conv": false,
  "use_quant_conv": false
}