[Question]: Taskflow如何指定模型路径，task为information_extraction

[liujiachang]

请提出你的问题

我通过task_path和home_path设定均会报错

Traceback (most recent call last):
  File "/root/work/filestorage/liujc/paddle/main.py", line 7, in <module>
    model = Taskflow(
  File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/taskflow.py", line 809, in __init__
    self.task_instance = task_class(
  File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/information_extraction.py", line 536, in __init__
    self._get_inference_model()
  File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/task.py", line 372, in _get_inference_model
    self._prepare_static_mode()
  File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/task.py", line 227, in _prepare_static_mode
    self.predictor = paddle.inference.create_predictor(self._config)
RuntimeError: (Unavailable) Not allowed to load partial data via load_combine_op, please use load_op instead.
  [Hint: Expected buffer->eof() == true, but received buffer->eof():0 != true:1.] (at /paddle/paddle/phi/kernels/impl/load_combine_kernel_impl.h:81)
  [operator < load_combine > error]

而且保存的模型路径中，缺失了inference.pdmodel文件。

[wawltor]

你的paddle是什么版本？

[Decade-rider]

请提出你的问题

我通过task_path和home_path设定均会报错

Traceback (most recent call last):
  File "/root/work/filestorage/liujc/paddle/main.py", line 7, in <module>
    model = Taskflow(
  File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/taskflow.py", line 809, in __init__
    self.task_instance = task_class(
  File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/information_extraction.py", line 536, in __init__
    self._get_inference_model()
  File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/task.py", line 372, in _get_inference_model
    self._prepare_static_mode()
  File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/task.py", line 227, in _prepare_static_mode
    self.predictor = paddle.inference.create_predictor(self._config)
RuntimeError: (Unavailable) Not allowed to load partial data via load_combine_op, please use load_op instead.
  [Hint: Expected buffer->eof() == true, but received buffer->eof():0 != true:1.] (at /paddle/paddle/phi/kernels/impl/load_combine_kernel_impl.h:81)
  [operator < load_combine > error]

而且保存的模型路径中，缺失了inference.pdmodel文件。

请问您找到在哪里下载inference.pdmodel文件了吗

[liujiachang]

paddlepaddle 3.0.0.dev20241225
paddle-custom-npu 3.0.0.dev20241226
paddlenlp 3.0.0b3

通过设置环境变量
export FLAGS_enable_pir_api=0
修复了该问题

[liujiachang]

请提出你的问题

我通过task_path和home_path设定均会报错

Traceback (most recent call last):
  File "/root/work/filestorage/liujc/paddle/main.py", line 7, in <module>
    model = Taskflow(
  File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/taskflow.py", line 809, in __init__
    self.task_instance = task_class(
  File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/information_extraction.py", line 536, in __init__
    self._get_inference_model()
  File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/task.py", line 372, in _get_inference_model
    self._prepare_static_mode()
  File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/task.py", line 227, in _prepare_static_mode
    self.predictor = paddle.inference.create_predictor(self._config)
RuntimeError: (Unavailable) Not allowed to load partial data via load_combine_op, please use load_op instead.
  [Hint: Expected buffer->eof() == true, but received buffer->eof():0 != true:1.] (at /paddle/paddle/phi/kernels/impl/load_combine_kernel_impl.h:81)
  [operator < load_combine > error]

而且保存的模型路径中，缺失了inference.pdmodel文件。

请问您找到在哪里下载inference.pdmodel文件了吗

该文件不是下载下来的, static/目录是生成的

[Decade-rider]

我的意思是多次下载不成功，就想自己手动下载放到相应地址，但是没找到在仓库的哪里进行下载

…

---Original--- From: ***@***.***&gt; Sent at: 2025年1月2日(Thu) AM8:41 To: ***@***.***&gt;; Cc: "Chonglin ***@***.******@***.***&gt;; Subject: Re: [PaddlePaddle/PaddleNLP] [Question]: Taskflow如何指定模型路径，task为information_extraction (Issue #9698) 请提出你的问题 我通过task_path和home_path设定均会报错 Traceback (most recent call last): File "/root/work/filestorage/liujc/paddle/main.py", line 7, in <module&gt; model = Taskflow( File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/taskflow.py", line 809, in __init__ self.task_instance = task_class( File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/information_extraction.py", line 536, in __init__ self._get_inference_model() File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/task.py", line 372, in _get_inference_model self._prepare_static_mode() File "/usr/local/lib/python3.10/dist-packages/paddlenlp/taskflow/task.py", line 227, in _prepare_static_mode self.predictor = paddle.inference.create_predictor(self._config) RuntimeError: (Unavailable) Not allowed to load partial data via load_combine_op, please use load_op instead. [Hint: Expected buffer-&gt;eof() == true, but received buffer-&gt;eof():0 != true:1.] (at /paddle/paddle/phi/kernels/impl/load_combine_kernel_impl.h:81) [operator < load_combine &gt; error] 而且保存的模型路径中，缺失了inference.pdmodel文件。 请问您找到在哪里下载inference.pdmodel文件了吗 该文件不是下载下来的, static/目录是生成的 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt;

[liujiachang]

模型文件只有model_state.pdparams，static目录中的inference.pdmodel是运行taskflow代码生成的，他会自动检测你有没有inference.pdmodel这个文件，正常情况没有会主动生成，但是由于某些错误（我通过export FLAGS_enable_pir_api=0修复成功），没有生成，官方应该并没有提供该文件。

[Decade-rider]

好的，非常感谢

…

---Original--- From: ***@***.***&gt; Sent at: 2025年1月2日(Thu) AM10:04 To: ***@***.***&gt;; Cc: "Chonglin ***@***.******@***.***&gt;; Subject: Re: [PaddlePaddle/PaddleNLP] [Question]: Taskflow如何指定模型路径，task为information_extraction (Issue #9698) 模型文件只有model_state.pdparams，static目录中的inference.pdmodel是运行taskflow代码生成的，他会自动检测你有没有inference.pdmodel这个文件，正常情况没有会主动生成，但是由于某些错误（我通过export FLAGS_enable_pir_api=0修复成功），没有生成，官方应该并没有提供该文件。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt;

[Decade-rider]

模型文件只有model_state.pdparams，static目录中的inference.pdmodel是运行taskflow代码生成的，他会自动检测你有没有inference.pdmodel这个文件，正常情况没有会主动生成，但是由于某些错误（我通过export FLAGS_enable_pir_api=0修复成功），没有生成，官方应该并没有提供该文件。
非常感谢，在试了多次无果后，我重新配置了cuda、paddle，现在暂时问题解决了