config.json

{
  "_class_name": "AutoencoderKLQwenImage",
  "_diffusers_version": "0.35.0",
  "_name_or_path": "pretrained_models/qwenimage-vae-e2e-lr2e-5-400k",
  "attn_scales": [],
  "base_dim": 96,
  "dim_mult": [
    1,
    2,
    4,
    4
  ],
  "dropout": 0.0,
  "latents_mean": [
    -0.04180605337023735,
    -0.015671083703637123,
    -0.005345471668988466,
    -0.01270077470690012,
    -0.04445185139775276,
    0.03513055667281151,
    -0.03672828525304794,
    0.023888064548373222,
    -0.03632868081331253,
    -0.004377194680273533,
    0.03795469179749489,
    -0.001475810189731419,
    -0.08210355043411255,
    -0.10997337102890015,
    -0.04832911491394043,
    0.007730678655207157
  ],
  "latents_std": [
    2.334869623184204,
    2.366467237472534,
    2.3873367309570312,
    2.3957691192626953,
    2.377338409423828,
    2.405400514602661,
    2.390782356262207,
    2.3725364208221436,
    2.3622777462005615,
    2.382412910461426,
    2.4043378829956055,
    2.36690354347229,
    2.380032777786255,
    2.3778722286224365,
    2.388890027999878,
    2.3639066219329834
  ],
  "num_res_blocks": 2,
  "temperal_downsample": [
    false,
    true,
    true
  ],
  "z_dim": 16
}