ls_bert运行报错

[js19950726]

我安装了lightseq==2.1.1

但是运行时

找了半天bert在哪也没找到，可以告诉我咋解决吗，还是我少了什么步骤

[js19950726]

能不能贴一个运行流程？

[Taka152]

The pypi version hasn't update Bert support, you can build from source and try.
I'll try to release a new pypi version these days.

[js19950726]

我就是从源代码python setup.py install的啊
pypi上的版本不是2.1.0嘛
我的是2.1.1

[Taka152]

pip install only install training module if inference module installation failed, you can add -v to verify that.

lightseq/setup.py

Lines 130 to 139 in a57966a

	try:
	setup(ext_modules=ext_modules, **setup_kwargs)
	except Exception as e:
	logger.warning(e)
	logger.warning("The inference extension could not be compiled")
	# If this new 'setup' call don't fail, the module
	# will be successfully installed, without the C extension :
	setup(**setup_kwargs)
	logger.info("lightseq training installation succeeded.")

Check here to install inference env and run python setup.py install again.

[godweiyang]

我就是从源代码python setup.py install的啊
pypi上的版本不是2.1.0嘛
我的是2.1.1

We have updated the pypi version to 2.1.2. which includes the BERT inference. You can try again.

[ZTurboX]

我也遇到了这个问题，pypi version to 2.1.2也会报错

[Taka152]

2.1.2 works for me, could you share the error log?

[ZTurboX]

2.1.2 works for me, could you share the error log?

AttributeError: module 'lightseq.inference' has no attribute 'Bert'

[Taka152]

could you share the pip list version and your terminal screeshot

[ZTurboX]

could you share the pip list version and your terminal screeshot

[ZTurboX]

could you share the pip list version and your terminal screeshot

version = "2.1.2"

[Taka152]

Just want to make sure your pwd, this error may due to wrong package import, you can try running in different working directory

[ZTurboX]

pwd is correct

[godweiyang]

I try it successfully:

<Trial 3417505 worker_0> tiger $ pip3 install lightseq
Defaulting to user installation because normal site-packages is not writeable
Collecting lightseq
  Downloading lightseq-2.1.2-cp37-cp37m-manylinux_2_12_x86_64.manylinux2010_x86_64.whl (40.8 MB)
     |████████████████████████████████| 40.8 MB 106 kB/s 
Requirement already satisfied: ninja in /usr/local/lib/python3.7/dist-packages (from lightseq) (1.10.2)
Installing collected packages: lightseq
Successfully installed lightseq-2.1.2
<Trial 3417505 worker_0> tiger $ python3
Python 3.7.3 (default, Jul 25 2020, 13:03:44) 
[GCC 8.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import lightseq.inference as lsi
>>> lsi.Bert
<class 'lightseq.inference.Bert'>

[js19950726]

2.1.2 works for me, could you share the error log?

AttributeError: module 'lightseq.inference' has no attribute 'Bert'

你这个问题是因为里面有个叫lightseq的文件夹，你import的是文件夹而不是你安装的lightseq库，你可以在gpu上把lightseq文件夹删除就好了

[godweiyang]

2.1.2 works for me, could you share the error log?

AttributeError: module 'lightseq.inference' has no attribute 'Bert'

你这个问题是因为里面有个叫lightseq的文件夹，你import的是文件夹而不是你安装的lightseq库，你可以在gpu上把lightseq文件夹删除就好了

Another reason is that he directly clones and uses pip install -e . to install the source code, which will only install the training library for lack of inference environment. So we recommend installing by PyPi using pip install lightseq.