fix: syntax

app.py

@@ -60,7 +60,7 @@ samplers_k_diffusion = [
 # ]
 
 start_time = time.time()
-timeout = 60
+timeout = 90
 
 scheduler = DDIMScheduler.from_pretrained(
     base_model,
@@ -130,11 +130,12 @@ def setup_model(name, lora_state=None, lora_scale=1.0):
         te_cache[model] = text_encoder
         lora_cache[model] = LoRANetwork(text_encoder, unet)
 
-    if current_model != model and current_model not in keep_vram:
-        # offload current model
-        unet_cache[current_model].to("cpu")
-        te_cache[current_model].to("cpu")
-        lora_cache[current_model].to("cpu")
+    if current_model != model:
+        if current_model not in keep_vram:
+            # offload current model
+            unet_cache[current_model].to("cpu")
+            te_cache[current_model].to("cpu")
+            lora_cache[current_model].to("cpu")
         current_model = model
 
     local_te, local_unet, local_lora, = te_cache[model], unet_cache[model], lora_cache[model]
@@ -158,6 +159,10 @@ def setup_model(name, lora_state=None, lora_scale=1.0):
     pipe.setup_text_encoder(clip_skip, local_te)
     return pipe
 
+# precache on huggingface
+for model in models:
+    setup_model(model[0])
+
 def error_str(error, title="Error"):
     return (
         f"""#### {title}