Untitled

        results = []
        for i in tqdm(range(0, len(dataset), batch_size)):
            # Prepare batch inputs
            batch_inputs_raw = dataset[i : i + batch_size]

            batch_texts = [
                self.processor.apply_chat_template(
                    input_raw["messages"], add_generation_prompt=True
                )
                for input_raw in batch_inputs_raw
            ]
            batch_images = [
                [load_image(input_raw["image_path"], model.device)] for input_raw in batch_inputs_raw
            ]

            start_time = time.time()
            # Process batch
            inputs = self.processor(
                batch_images,
                batch_texts,
                add_special_tokens=False,
                return_tensors="pt",
                padding=True,
            ).to(model.device)

            end_time = time.time()
            print(f"Time taken for processing batch: {end_time - start_time} seconds")

            start_time = time.time()

            with torch.no_grad():
                outputs = model.generate(
                    **inputs, max_new_tokens=256, do_sample=False, top_p=None
                )
            end_time = time.time()
            print(f"Time taken for generating batch: {end_time - start_time} seconds")  

            start_time = time.time()
            output_strs = self.processor.batch_decode(outputs, skip_special_tokens=True)
            end_time = time.time()
            print(f"Time taken for decoding batch: {end_time - start_time} seconds")