Update worker_runpod.py

worker_runpod.py

@@ -94,7 +94,6 @@ with torch.inference_mode():
     
     scheduler = sampler_dict["DPM++"].from_pretrained(model_id, subfolder="scheduler")
 
-    pipeline = merge_lora(pipeline, lora_path, 1.00)
 
     if transformer.config.in_channels != vae.config.latent_channels:
         pipeline = CogVideoX_Fun_Pipeline_Inpaint.from_pretrained(
@@ -115,6 +114,8 @@ with torch.inference_mode():
             torch_dtype=torch.bfloat16
         )
 
+    pipeline = merge_lora(pipeline, lora_path, 1.00)
+
 
 # if low_gpu_memory_mode:
 #     pipeline.enable_sequential_cpu_offload()