We finetune the pre-trained VAE ft-MSE of Stable Diffusion on the DeepFashion training set. But I find code stabilityai/sd-vae-ft-mse

[typhoon1104]

train
def instantiate_first_stage(self):
model = AutoencoderKL.from_pretrained("stabilityai/sd-vae-ft-mse").to(dist_util.dev())
self.first_stage_model = model.eval()
self.first_stage_model.train = False
for param in self.first_stage_model.parameters():
param.requires_grad = False
eval：
model_name = 'model_zoo/diffuser_finetune_vae'
vae = AutoencoderKL.from_pretrained(model_name).to(device)