Update handler.py

handler.py

@@ -70,7 +70,8 @@ class EndpointHandler:
         inputs = self.processor(images=images, texts=prompt, return_tensors="pt")
         inputs["pixel_values"] = inputs["pixel_values"].unsqueeze(0)
         inputs["image_sizes"] = inputs["image_sizes"].unsqueeze(0)
-        inputs.to("cuda").to(torch.bfloat16)
+        inputs = inputs.to("cuda").to(torch.bfloat16)
+        logger.info(f"Inputs contains {inputs=}")
 
         generation_args = {
             "max_new_tokens": data.get("max_new_tokens", data.get("max_tokens", 128)),
@@ -82,7 +83,9 @@ class EndpointHandler:
         logger.info(f"Running text generation with the following {generation_args=}")
 
         with torch.inference_mode():
+            logger.info(f"Inputs contains {inputs['input_ids']=}")
             generate_ids = self.model.generate(**inputs, **generation_args)
+            logger.info(f"Generate IDs contains {generate_ids=}")
 
         logger.info(f"Generated {generate_ids=}")
         generate_ids = generate_ids[:, inputs["input_ids"].shape[-1] :]