bug描述 Describe the Bug

bug描述

paddle.io.Dataloader使用自定义的collate_fn，返回数据为list时，即使 len(loader)>0 ，Dataloader迭代也会直接结束，，且未抛出异常。

修复建议

调试定位到 paddle/fluid/dataloader/flat.py中的_flatten_batch()函数无法正确处理数组类型

复现代码

from paddle.io import DataLoader, Dataset

class DataDummy(Dataset):
    def __init__(self):
        self.data = [(1, 2, 3) for _ in range(300)]

    def __len__(self):
        return len(self.data)

    def __getitem__(self, idx):
        return self.data[idx]

    def collate_fn(self, batch):
        # # 可正常运行
        # data = paddle.to_tensor([s[0] for s in batch]), paddle.to_tensor([s[1] for s in batch]), paddle.to_tensor([s[2] for s in batch])

        # BUG: 无法正常获取数据，且没有抛出任何异常, 调试定位到paddle/fluid/dataloader/flat.py中的_flatten_batch函数无法正确处理数组类型
        data = [s[0] for s in batch], [s[1] for s in batch], [s[2] for s in batch]
        return data

if __name__ == "__main__":
    ds = DataDummy()
    datad = DataLoader(ds, batch_size=128, collate_fn=ds.collate_fn)

    print("len(datad) =", len(datad))

    read = []
    for i, batch in enumerate(datad):
        read.append(batch)
    print("实际读取batch数量 = ", len(read))

其他补充信息 Additional Supplementary Information

• No response*