0.5.0版本在ms2.6上提示导入错误

环境：

![输入图片说明](https://foruda.gitee.com/images/1753882185040219240/11f01e0c_7963367.png "屏幕截图")

**执行代码：** 
import mindspore
import mindnlp
from transformers import pipeline

chat = [
    {"role": "system", "content": "You are a sassy, wise-cracking robot as imagined by Hollywood circa 1986."},
    {"role": "user", "content": "Hey, can you tell me any fun things to do in New York?"}
]

pipeline = pipeline(task="text-generation", model="Qwen/Qwen3-8B", ms_dtype=mindspore.bfloat16, device_map="auto")
response = pipeline(chat, max_new_tokens=512)
print(response[0]["generated_text"][-1]["content"])

![输入图片说明](https://foruda.gitee.com/images/1753882065194220196/c343ce6c_7963367.png "屏幕截图")

**完整报错如下：**

ImportError                               Traceback (most recent call last)
Cell In[2], line 2
      1 import mindspore
----> 2 import mindnlp
      3 from transformers import pipeline
      5 chat = [
      6     {"role": "system", "content": "You are a sassy, wise-cracking robot as imagined by Hollywood circa 1986."},
      7     {"role": "user", "content": "Hey, can you tell me any fun things to do in New York?"}
      8 ]

File ~/miniconda/envs/jupyter/lib/python3.9/site-packages/mindnlp/__init__.py:50
     47 from .core._tensor import enable_mindspore_patch
     49 enable_mindspore_patch()
---> 50 initialize_torch_proxy()
     51 setup_metadata_patch()
     52 setup_safetensors_patch()

File ~/miniconda/envs/jupyter/lib/python3.9/site-packages/mindnlp/utils/torch_proxy.py:99, in initialize_torch_proxy()
     97 def initialize_torch_proxy():
     98     sys.meta_path.insert(0, RedirectFinder(REDIRECT_MAP))
---> 99     import torch
    100     torch.__version__ = "2.1.1+dev"

File ~/miniconda/envs/jupyter/lib/python3.9/site-packages/mindnlp/utils/torch_proxy.py:23, in RedirectFinder.find_spec(self, fullname, path, target)
     21             importlib.import_module(target_name)
     22         except Exception as e:
---> 23             raise e
     25         return importlib.machinery.ModuleSpec(
     26             name=fullname,
     27             loader=RedirectLoader(target_name),
     28             is_package=self._is_package(target_name),
     29         )
     30 return None

File ~/miniconda/envs/jupyter/lib/python3.9/site-packages/mindnlp/utils/torch_proxy.py:21, in RedirectFinder.find_spec(self, fullname, path, target)
     19 target_name = fullname.replace(proxy_prefix, target_prefix, 1)
     20 try:
---> 21     importlib.import_module(target_name)
     22 except Exception as e:
     23     raise e

File ~/miniconda/envs/jupyter/lib/python3.9/importlib/__init__.py:127, in import_module(name, package)
    125             break
    126         level += 1
--> 127 return _bootstrap._gcd_import(name[level:], package, level)

File ~/miniconda/envs/jupyter/lib/python3.9/site-packages/mindnlp/core/__init__.py:50
     48 from ._C.size import Size
     49 from .types import device
---> 50 from .autograd import *
     51 from .ops import *
     52 from .serialization import load, save

File ~/miniconda/envs/jupyter/lib/python3.9/site-packages/mindnlp/core/autograd/__init__.py:3
      1 """autograd"""
      2 from .node import Node
----> 3 from .function import Function, value_and_grad
      4 from .grad_mode import no_grad, enable_grad, inference_mode

File ~/miniconda/envs/jupyter/lib/python3.9/site-packages/mindnlp/core/autograd/function.py:9
      7 from mindspore.common.api import _pynative_executor
      8 from mindspore._c_expression import Cell_
----> 9 from .grad_mode import no_grad
     11 try:
     12     from mindspore import _Function as Function

File ~/miniconda/envs/jupyter/lib/python3.9/site-packages/mindnlp/core/autograd/grad_mode.py:17
     15 """core module"""
     16 from mindspore.common.api import _pynative_executor
---> 17 from ..utils._contextlib import _NoParamDecoratorContextManager, _DecoratorContextManager
     19 class no_grad(_NoParamDecoratorContextManager):
     20     """
     21     Context Manager to disable gradient calculation. When enter this context, we will disable calculate
     22     gradient. When exit this context, we will resume its prev state.
     23     Currently, it can only use in Pynative mode. It also can be used as decorator.
     24     """

File ~/miniconda/envs/jupyter/lib/python3.9/site-packages/mindnlp/core/utils/__init__.py:2
      1 """utils"""
----> 2 from . import data

ImportError: cannot import name 'data' from partially initialized module 'mindnlp.core.utils' (most likely due to a circular import) (/home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages/mindnlp/core/utils/__init__.py)

MindSpore Lab/mindnlp

Content Risk Flag

Comments (0)

MindSpore Lab/mindnlp .gitee-modal { width: 500px !important; }

Content Risk Flag