Fix no half vae

modules/processing_vae.py

@@ -44,7 +44,10 @@ def full_vae_decode(latents, model):
     upcast = (model.vae.dtype == torch.float16) and getattr(model.vae.config, 'force_upcast', False) and hasattr(model, 'upcast_vae')
     if upcast: # this is done by diffusers automatically if output_type != 'latent'
         model.upcast_vae()
-    if getattr(model.vae, "post_quant_conv", None) is not None:
+
+    if shared.opts.no_half_vae:
+        latents = latents.to(torch.float32)
+    elif getattr(model.vae, "post_quant_conv", None) is not None:
         latents = latents.to(next(iter(model.vae.post_quant_conv.parameters())).dtype)
 
     # normalize latents

modules/sd_models_compile.py

@@ -270,7 +270,7 @@ def optimum_quanto_weights(sd_model):
             else:
                 sd_models.move_model(sd_model, devices.device)
             with quanto.Calibration(momentum=0.9):
-                sd_model(prompt="dummy", height=512, width=512, guidance_scale=4.0, num_inference_steps=10)
+                sd_model(prompt="dummy prompt", num_inference_steps=10)
             sd_model = apply_compile_to_model(sd_model, optimum_quanto_freeze, shared.opts.optimum_quanto_weights, op="optimum-quanto-freeze")
             if shared.opts.diffusers_offload_mode == "model":
                 sd_models.disable_offload(sd_model)