使用verl进行多模态RL训练的时候（图片存储在本地服务器），生成训练和测试数据的时候，这个"images"字段中应该填入什么？求大神解答一下哦

[lmingze]

data = { "data_source": data_source, "prompt": [ { "role": "user", "content": prompt, } ], "images": images, "ability": "math", "reward_model": {"style": "rule", "ground_truth": answer}, "extra_info": { "split": split, "index": idx, "answer": answer, "question": problem, }, }

[yszhli]

def process_image(image: dict | Image.Image) -> Image.Image: if isinstance(image, Image.Image): return image.convert("RGB")

if "bytes" in image:
    assert "image" not in image, "Cannot have both `bytes` and `image`"
    image["image"] = Image.open(BytesIO(image["bytes"]))

return fetch_image(image) 可以是一个Image.Image ，也可以是一个字典，存储图像的二进程，使用字段为bytes