[Question]: 使用Taskflow进行NER时出错

[tszxyuan123]

请提出你的问题

环境为windows10家庭版，paddle版本为3.0-beta的cpu版本

from paddlenlp.taskflow import Taskflow

ner = Taskflow("ner")
出现的问题是，当输入上面第二行代码时，报错，具体日志如下：
`[2024-11-12 16:05:44,254] [ INFO] - Downloading model_state.pdparams from https://bj.bcebos.com/paddlenlp/taskflow/knowledge_mining/wordtag_v1.5/model_state.pdparams

[2024-11-12 16:08:27,954] [ INFO] - Downloading config.json from https://bj.bcebos.com/paddlenlp/taskflow/knowledge_mining/wordtag_v1.1/config.json

[2024-11-12 16:08:28,355] [ INFO] - Downloading tags.txt from https://bj.bcebos.com/paddlenlp/taskflow/knowledge_mining/wordtag_v1.1/tags.txt

[2024-11-12 16:08:28,561] [ INFO] - Downloading vocab.txt from https://bj.bcebos.com/paddlenlp/taskflow/knowledge_mining/wordtag/vocab.txt

[2024-11-12 16:08:28,980] [ INFO] - Downloading special_tokens_map.json from https://bj.bcebos.com/paddlenlp/taskflow/knowledge_mining/wordtag/special_tokens_map.json

[2024-11-12 16:08:29,206] [ INFO] - Downloading tokenizer_config.json from https://bj.bcebos.com/paddlenlp/taskflow/knowledge_mining/wordtag/tokenizer_config.json

[2024-11-12 16:08:29,421] [ INFO] - Loading configuration file C:\Users\lenovo.paddlenlp\taskflow\wordtag\config.json
[2024-11-12 16:08:29,424] [ INFO] - Loading weights file C:\Users\lenovo.paddlenlp\taskflow\wordtag\model_state.pdparams
[2024-11-12 16:08:29,665] [ INFO] - Loaded weights file from disk, setting weights to model.
[2024-11-12 16:08:36,820] [ WARNING] - Some weights of the model checkpoint at C:\Users\lenovo.paddlenlp\taskflow\wordtag were not used when initializing ErnieCtmWordtagModel: ['sent_classifier.bias', 'sent_classifier.weight']

• This IS expected if you are initializing ErnieCtmWordtagModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).

• This IS NOT expected if you are initializing ErnieCtmWordtagModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
  [2024-11-12 16:08:36,821] [ WARNING] - Some weights of ErnieCtmWordtagModel were not initialized from the model checkpoint at C:\Users\lenovo.paddlenlp\taskflow\wordtag and are newly initialized: ['viterbi_decoder.transitions', 'crf_loss.crf.transitions']
  You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
  [2024-11-12 16:08:36,847] [ INFO] - Converting to the inference model cost a little time.
  D:\anaconda3\Lib\site-packages\paddle\jit\dy2static\program_translator.py:747: UserWarning: full_graph=False don't support input_spec arguments. It will not produce any effect.
  You can set full_graph=True, then you can assign input spec.

  warnings.warn(
  [2024-11-12 16:08:41,319] [ INFO] - The inference model save in the path:C:\Users\lenovo.paddlenlp\taskflow\wordtag\static\inference

---

RuntimeError Traceback (most recent call last)
Cell In[2], line 1
----> 1 ner = Taskflow("ner")

File D:\anaconda3\Lib\site-packages\paddlenlp\taskflow\taskflow.py:809, in Taskflow.init(self, task, model, mode, device_id, from_hf_hub, **kwargs)
807 self.kwargs = kwargs
808 task_class = TASKS[self.task][tag][self.model]["task_class"]
--> 809 self.task_instance = task_class(
810 model=self.model, task=self.task, priority_path=self.priority_path, from_hf_hub=from_hf_hub, **self.kwargs
811 )
812 task_list = TASKS.keys()
813 Taskflow.task_list = task_list

File D:\anaconda3\Lib\site-packages\paddlenlp\taskflow\named_entity_recognition.py:123, in NERWordTagTask.init(self, model, task, entity_only, **kwargs)
122 def init(self, model, task, entity_only=False, **kwargs):
--> 123 super().init(model="wordtag", task=task, **kwargs)
124 self.entity_only = entity_only
125 if self._user_dict:

File D:\anaconda3\Lib\site-packages\paddlenlp\taskflow\knowledge_mining.py:235, in WordTagTask.init(self, model, task, tag_path, term_schema_path, term_data_path, user_dict, linking, spo_config_path, with_ie, **kwargs)
233 self._usage = usage
234 self._summary_num = 2
--> 235 self._get_inference_model()
237 if self._user_dict:
238 self._custom = Customization()

File D:\anaconda3\Lib\site-packages\paddlenlp\taskflow\task.py:371, in Task._get_inference_model(self)
369 self._static_params_file = self._static_fp16_params_file
370 if self._predictor_type == "paddle-inference":
--> 371 self._config = paddle.inference.Config(self._static_model_file, self._static_params_file)
372 self._prepare_static_mode()
373 else:

RuntimeError: (NotFound) Cannot open file C:\Users\lenovo.paddlenlp\taskflow\wordtag\static\inference.pdmodel, please confirm whether the file is normal.
[Hint: Expected paddle::inference::IsFileExists(prog_file_) == true, but received paddle::inference::IsFileExists(prog_file_):0 != true:1.] (at ..\paddle\fluid\inference\api\analysis_config.cc:111)`

有查到和这个问题类似：#3242
但是我的目录中并没有中文，用户名是lenovo，为什么也会出现这种情况？

[wawltor]

可以尝试删除再试试 C:\Users\lenovo.paddlenlp\taskflow\wordtag\static\

[tszxyuan123]

您是说删除static这个文件夹，再重新运行ner = Taskflow("ner")吗
我刚才试了一下，依然报同样的错误
我的图片上传似乎有些问题，C:\Users\lenovo.paddlenlp\taskflow\wordtag\static\，这个文件夹中只有inference.json和inference.pdiparams，没有inference.pdmodel

[wawltor]

可以尝试把paddle版本切换到2.5.2版本，同时删除 ，C:\Users\lenovo.paddlenlp\taskflow\wordtag\static\

[tszxyuan123]

您是说把paddlenlp的版本降到2.5.2吗，我试一下

[wawltor]

您是说把paddlenlp的版本降到2.5.2吗，我试一下

是的

[tszxyuan123]

确实没有出现之前的报错了，但是出现了另一个错误
`---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
Cell In[5], line 1
----> 1 ner = Taskflow("ner")

File D:\anaconda3\Lib\site-packages\paddlenlp\taskflow\taskflow.py:809, in Taskflow.init(self, task, model, mode, device_id, from_hf_hub, **kwargs)
807 self.kwargs = kwargs
808 task_class = TASKS[self.task][tag][self.model]["task_class"]
--> 809 self.task_instance = task_class(
810 model=self.model, task=self.task, priority_path=self.priority_path, from_hf_hub=from_hf_hub, **self.kwargs
811 )
812 task_list = TASKS.keys()
813 Taskflow.task_list = task_list

File D:\anaconda3\Lib\site-packages\paddlenlp\taskflow\named_entity_recognition.py:123, in NERWordTagTask.init(self, model, task, entity_only, **kwargs)
122 def init(self, model, task, entity_only=False, **kwargs):
--> 123 super().init(model="wordtag", task=task, **kwargs)
124 self.entity_only = entity_only
125 if self._user_dict:

File D:\anaconda3\Lib\site-packages\paddlenlp\taskflow\knowledge_mining.py:235, in WordTagTask.init(self, model, task, tag_path, term_schema_path, term_data_path, user_dict, linking, spo_config_path, with_ie, **kwargs)
233 self._usage = usage
234 self._summary_num = 2
--> 235 self._get_inference_model()
237 if self._user_dict:
238 self._custom = Customization()

File D:\anaconda3\Lib\site-packages\paddlenlp\taskflow\task.py:343, in _get_inference_model(self)
339 self._static_model_file = self.inference_model_path + ".pdmodel"
340 self._static_params_file = self.inference_model_path + ".pdiparams"
342 if paddle.get_device().split(":", 1)[0] == "npu" and self._infer_precision == "fp16":
--> 343 # transform fp32 model tp fp16 model
344 self._static_fp16_model_file = self.inference_model_path + "-fp16.pdmodel"
345 self._static_fp16_params_file = self.inference_model_path + "-fp16.pdiparams"

File D:\anaconda3\Lib\site-packages\paddlenlp\taskflow\task.py:389, in _convert_dygraph_to_static(self)
386 def _check_input_text(self, inputs):
387 """
388 Check whether the input text meet the requirement.
--> 389 """
390 inputs = inputs[0]
391 if isinstance(inputs, str):

File D:\anaconda3\Lib\site-packages\decorator.py:232, in decorate..fun(*args, **kw)
230 if not kwsyntax:
231 args, kw = fix(args, kw, sig)
--> 232 return caller(func, *(extras + args), **kw)

File D:\anaconda3\Lib\site-packages\paddle\base\wrapped_decorator.py:40, in wrap_decorator..impl(func, *args, **kwargs)
33 @decorator.decorator
34 def impl(
35 func: Callable[_InputT, _RetT1],
36 *args: _InputT.args,
37 **kwargs: _InputT.kwargs,
38 ) -> _RetT2:
39 wrapped_func = decorator_func(func)
---> 40 return wrapped_func(*args, **kwargs)

File D:\anaconda3\Lib\site-packages\paddle\jit\api.py:895, in _run_save_pre_hooks..wrapper(layer, path, input_spec, **configs)
893 for hook in _save_pre_hooks:
894 hook(layer, input_spec, configs)
--> 895 func(layer, path, input_spec, **configs)

File D:\anaconda3\Lib\site-packages\decorator.py:232, in decorate..fun(*args, **kw)
230 if not kwsyntax:
231 args, kw = fix(args, kw, sig)
--> 232 return caller(func, *(extras + args), **kw)

File D:\anaconda3\Lib\site-packages\paddle\base\wrapped_decorator.py:40, in wrap_decorator..impl(func, *args, **kwargs)
33 @decorator.decorator
34 def impl(
35 func: Callable[_InputT, _RetT1],
36 *args: _InputT.args,
37 **kwargs: _InputT.kwargs,
38 ) -> _RetT2:
39 wrapped_func = decorator_func(func)
---> 40 return wrapped_func(*args, **kwargs)

File D:\anaconda3\Lib\site-packages\paddle\base\dygraph\base.py:101, in switch_to_static_graph..impl(*args, **kwargs)
99 def impl(*args: _InputT.args, **kwargs: _InputT.kwargs) -> _RetT:
100 with framework._dygraph_guard(None):
--> 101 return func(*args, **kwargs)

File D:\anaconda3\Lib\site-packages\paddle\jit\api.py:1209, in save(layer, path, input_spec, **configs)
1200 property_vals.append(
1201 (
1202 immediate_val,
1203 layer.class.name + '.' + attr_func,
1204 )
1205 )
1206 continue
1208 concrete_program = (
-> 1209 static_func.concrete_program_specify_input_spec(
1210 inner_input_spec,
1211 with_hook=with_hook,
1212 is_prim_infer=is_prim_infer,
1213 )
1214 )
1216 elif 'forward' == attr_func:
1217 if configs.skip_forward:
1218 # do not jit.save forward function

File D:\anaconda3\Lib\site-packages\paddle\jit\dy2static\program_translator.py:1026, in ASTStaticFunction.concrete_program_specify_input_spec(self, input_spec, with_hook, is_prim_infer)
1024 has_input_spec = desired_input_spec is not None
1025 if has_input_spec:
-> 1026 concrete_program, _ = self.get_concrete_program(
1027 *desired_input_spec,
1028 with_hook=with_hook,
1029 is_train=self._is_train_mode(),
1030 is_prim_infer=is_prim_infer,
1031 )
1032 return concrete_program
1033 else:

File D:\anaconda3\Lib\site-packages\paddle\jit\dy2static\program_translator.py:914, in ASTStaticFunction.get_concrete_program(self, *args, **kwargs)
908 (
909 concrete_program,
910 partial_program_layer,
911 ) = self._program_cache.get_program_without_cache(cache_key)
912 else:
913 # 3. check whether hit the cache or build a new program for the input arguments
--> 914 concrete_program, partial_program_layer = self._program_cache[
915 cache_key
916 ]
917 partial_program_layer._debug_name = self._debug_name
918 return concrete_program, partial_program_layer

File D:\anaconda3\Lib\site-packages\paddle\jit\dy2static\program_translator.py:1665, in ProgramCache.getitem(self, item)
1663 self._recent_key = item_id
1664 if item_id not in self._caches:
-> 1665 self._caches[item_id] = self._build_once(item)
1666 # Note: raise warnings if number of traced program is more than max_tracing_count
1667 current_tracing_count = len(self._caches)

File D:\anaconda3\Lib\site-packages\paddle\jit\dy2static\program_translator.py:1603, in ProgramCache._build_once(self, cache_key)
1600 enable_prim = cache_key.kwargs['build_strategy'].build_cinn_pass
1602 if use_pir_api():
-> 1603 concrete_program = ConcreteProgram.pir_from_func_spec(
1604 func_spec=cache_key.function_spec,
1605 input_spec=cache_key.input_args_with_spec,
1606 input_kwargs_spec=cache_key.input_kwargs_with_spec,
1607 class_instance=cache_key.class_instance,
1608 **cache_key.kwargs,
1609 )
1610 else:
1611 concrete_program = ConcreteProgram.from_func_spec(
1612 func_spec=cache_key.function_spec,
1613 input_spec=cache_key.input_args_with_spec,
(...)
1616 **cache_key.kwargs,
1617 )

File D:\anaconda3\Lib\site-packages\decorator.py:232, in decorate..fun(*args, **kw)
230 if not kwsyntax:
231 args, kw = fix(args, kw, sig)
--> 232 return caller(func, *(extras + args), **kw)

File D:\anaconda3\Lib\site-packages\paddle\base\wrapped_decorator.py:40, in wrap_decorator..impl(func, *args, **kwargs)
33 @decorator.decorator
34 def impl(
35 func: Callable[_InputT, _RetT1],
36 *args: _InputT.args,
37 **kwargs: _InputT.kwargs,
38 ) -> _RetT2:
39 wrapped_func = decorator_func(func)
---> 40 return wrapped_func(*args, **kwargs)

File D:\anaconda3\Lib\site-packages\paddle\base\dygraph\base.py:101, in switch_to_static_graph..impl(*args, **kwargs)
99 def impl(*args: _InputT.args, **kwargs: _InputT.kwargs) -> _RetT:
100 with framework._dygraph_guard(None):
--> 101 return func(*args, **kwargs)

File D:\anaconda3\Lib\site-packages\paddle\jit\dy2static\program_translator.py:1276, in ConcreteProgram.pir_from_func_spec(func_spec, input_spec, input_kwargs_spec, class_instance, **kwargs)
1274 error_data = getattr(e, error.ERROR_DATA, None)
1275 if error_data:
-> 1276 error_data.raise_new_exception()
1277 raise
1279 # 3. Gets all ParamBases and buffered VarBases in the function

File D:\anaconda3\Lib\site-packages\paddle\jit\dy2static\error.py:454, in ErrorData.raise_new_exception(self)
446 new_exception = self.create_exception()
447 # NOTE(liym27):
448 # Why raise new_exception from None?
449 #
450 # In Python 3, by default, an new exception is raised with trace information of the caught exception.
451 # This only raises new_exception and hides unwanted implementation details from tracebacks of the
452 # caught exception.
--> 454 raise new_exception from None

TypeError: In transformed code:

TypeError: create_forward.<locals>.forward() takes from 1 to 10 positional arguments but 11 were given

`

[tszxyuan123]

请问这个还是版本问题吗

[tszxyuan123]

当我设置paddlepaddle==2.6.1，paddlenlp==2.6.1时，就没有再报错了，可以正常使用了

[fengjunkuan2006]

兄弟，真心感谢你，我被这个问题困扰很久了，看了你的答复换了paddlenlp的版本也解决了。

当我设置paddlepaddle==2.6.1，paddlenlp==2.6.1时，就没有再报错了，可以正常使用了

[tszxyuan123]

客气，有帮助就好