TypeError: 'type' object is not subscriptable，请问改问题怎么解决？

[zhanghaobucunzai]

1.执行：bash exps/finetune/mm/alpacaLlava_llamaQformerv2_13B.sh ./LLaMA2-Accessory ./13B_params.json ./tokenizer.model 报错如下： raceback (most recent call last): File "main_finetune.py", line 46, in from accessory.util.tensor_type import default_tensor_type, promote_trainable_params_to_fp32 File "/home/LLaMA2-Accessory-main/accessory/util/tensor_type.py", line 7, in class default_tensor_type: File "/home/LLaMA2-Accessory-main/accessory/util/tensor_type.py", line 37, in default_tensor_type exc_type: Optional[type[BaseException]], TypeError: 'type' object is not subscriptable

2。修改LLaMA2-Accessory-main/accessory/util/tensor_type.py 37行 将exc_type: Optional[type[BaseException]] 改为exc_type: Optional[type([BaseException])]，暂时规避问题，继续执行

3.遇到新报错 Traceback (most recent call last): File "main_finetune.py", line 47, in from accessory.model.meta import MetaModel File "/home/LLaMA2-Accessory-main/accessory/model/meta.py", line 14, in from accessory.util import misc, tensor_parallel File "/home//LLaMA2-Accessory-main/accessory/util/tensor_parallel.py", line 86, in ) -> OrderedDict[str, torch.Tensor]: TypeError: 'type' object is not subscriptable

[ChrisLiu6]

Are you using python 3.8? Updating to python 3.10 should fix it

[zhanghaobucunzai]

是的，我是用的3.8的版本，后续还遇到了RuntimeError: Tensors must be CUDA and dense报错。