Update mini.py

mini.py

@@ -147,7 +147,7 @@ def process_image(image):
         # Move Qwen2VL models to GPU
         logger.info("Moving Qwen2VL models to GPU...")
         # qwen2vl.to(DEVICE)
-        connector.to(DEVICE)
+        # connector.to(DEVICE)
         
         message = [
             {
@@ -179,8 +179,8 @@ def process_image(image):
             result = (image_hidden_state.cpu(), image_grid_thw)
         
         # Move models back to CPU
-        qwen2vl.cpu()
-        connector.cpu()
+        # qwen2vl.cpu()
+        # connector.cpu()
         torch.cuda.empty_cache()
         
         return result
@@ -221,7 +221,7 @@ def compute_t5_text_embeddings(prompt):
     
     prompt_embeds = text_encoder_two(text_inputs.input_ids)[0]
     prompt_embeds = t5_context_embedder.to(DEVICE)(prompt_embeds)
-    t5_context_embedder.cpu()
+    # t5_context_embedder.cpu()
     
     return prompt_embeds
 
@@ -268,11 +268,11 @@ def generate(input_image, prompt="", guidance_scale=3.5, num_inference_steps=28,
         # Move Transformer and VAE to GPU
         logger.info("Moving Transformer and VAE to GPU...")
         # transformer.to(DEVICE)
-        vae.to(DEVICE)
+        # vae.to(DEVICE)
         
         # Update pipeline models
-        pipeline.transformer = transformer
-        pipeline.vae = vae
+        # pipeline.transformer = transformer
+        # pipeline.vae = vae
         logger.info("Models moved to GPU")
         
         # Get dimensions