def generate_until()

in lmms_eval/models/gpt4v.py [0:0]

• 49 lines of code
• 18 McCabe index (conditional complexity)

    def generate_until(self, requests) -> List[str]:
        res = []
        pbar = tqdm(total=len(requests), disable=(self.rank != 0), desc="Model Responding")

        for contexts, gen_kwargs, doc_to_visual, doc_id, task, split in [reg.args for reg in requests]:
            # encode, pad, and truncate contexts for this batch
            visuals = [doc_to_visual(self.task_dict[task][split][doc_id])]
            visuals = self.flatten(visuals)
            imgs = []
            for visual in visuals:
                img = self.encode_image(visual)
                imgs.append(img)

            payload = {"model": "gpt-4-vision-preview", "messages": []}
            response_json = {"role": "user", "content": []}
            # When there is no image token in the context, append the image to the text
            if self.image_token not in contexts:
                payload["messages"].append(deepcopy(response_json))
                payload["messages"][0]["content"].append({"type": "text", "text": contexts})
                for img in imgs:
                    payload["messages"][0]["content"].append({"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{img}"}})
            else:
                contexts = contexts.split(self.image_token)
                for idx, img in enumerate(imgs):
                    payload["messages"].append(deepcopy(response_json))
                    payload["messages"][idx]["content"].append({"type": "text", "text": contexts[idx]})
                    payload["messages"][idx]["content"].append({"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{img}"}})

                # If n image tokens are in the contexts
                # contexts will be splitted into n+1 chunks
                # Manually add it into the payload
                payload["messages"].append(deepcopy(response_json))
                payload["messages"][-1]["content"].append({"type": "text", "text": contexts[-1]})

            if "max_new_tokens" not in gen_kwargs:
                gen_kwargs["max_new_tokens"] = 1024
            if "temperature" not in gen_kwargs:
                gen_kwargs["temperature"] = 0
            if "top_p" not in gen_kwargs:
                gen_kwargs["top_p"] = None
            if "num_beams" not in gen_kwargs:
                gen_kwargs["num_beams"] = 1

            # payload["max_tokens"] = gen_kwargs["max_new_tokens"]
            # payload["temperature"] = gen_kwargs["temperature"]

            for attempt in range(5):
                try:
                    response = url_requests.post(API_URL, headers=headers, json=payload, timeout=20)
                    response_data = response.json()

                    content = response_data["choices"][0]["message"]["content"].strip()
                    break  # If successful, break out of the loop

                except Exception as e:
                    eval_logger.info(f"Attempt {attempt + 1} failed with error: {str(e)}")
                    if attempt < 5 - 1:  # If we have retries left, sleep and then continue to next attempt
                        time.sleep(NUM_SECONDS_TO_SLEEP)
                    else:  # If this was the last attempt, log and return empty
                        eval_logger.error(f"All 5 attempts failed. Last error message: {str(e)}")
                        content = ""
            res.append(content)
            pbar.update(1)
        return res