MfM-Pipeline-2B / vae /config.json
LetsThink's picture
Upload folder using huggingface_hub
9bf4fe6 verified
{
"_class_name": "AutoencoderKLCogVideoX",
"_diffusers_version": "0.33.1",
"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": [
0.1161,
-1.01,
0.7151,
0.7965,
0.6372,
1.0438,
0.5592,
-1.2404,
0.9421,
-0.6948,
-1.3285,
-0.7456,
-0.2101,
1.3217,
-0.3254,
-0.9015
],
"latents_std": [
1.4375,
0.4531,
0.5508,
0.5273,
0.7227,
0.543,
0.5781,
0.6992,
0.5117,
0.6406,
0.7148,
0.5664,
0.7109,
0.5547,
0.8008,
0.8633
],
"layers_per_block": 3,
"norm_eps": 1e-06,
"norm_num_groups": 32,
"out_channels": 3,
"sample_height": 480,
"sample_width": 720,
"scaling_factor": 1.15258426,
"shift_factor": null,
"temporal_compression_ratio": 4,
"up_block_types": [
"CogVideoXUpBlock3D",
"CogVideoXUpBlock3D",
"CogVideoXUpBlock3D",
"CogVideoXUpBlock3D"
],
"use_post_quant_conv": false,
"use_quant_conv": false
}