Bug: ImportError: cannot import name 'get_class_name' from partially initialized module 'openspeech.utils'

[Jxu-Thu]

Hi, Thanks for your great codes.

I followed the readme to install the packages.

Torch 1.6.0
CUDA 10.1
Python 3.8

Code Branch: Master

And Run

DATASET_PATH=data/LibriSpeech_raw
MANIFEST_FILE_PATH=data/LibriSpeech_raw/openspeech/libri_character_manifest.txt
VOCAB_PARH=data/LibriSpeech_raw/openspeech/libri_labels.csv

CUDA_VISIBLE_DEVICES=1 python ./openspeech_cli/hydra_train.py
dataset=librispeech
dataset.dataset_download=True
dataset.dataset_path=$DATASET_PATH
dataset.manifest_file_path=$MANIFEST_FILE_PATH
tokenizer=libri_character
model=conformer_transducer
vocab_path=$VOCAB_PARH
audio=fbank
lr_scheduler=warmup_reduce_lr_on_plateau
trainer=gpu-fp16
criterion=transducer

And encounterd errors as below:

File "./openspeech_cli/hydra_train.py", line 30, in
from openspeech.tokenizers import TOKENIZER_REGISTRY
File "openspeech-main/openspeech/init.py", line 23, in
import openspeech.criterion
File "openspeech-main/openspeech/criterion/init.py", line 70, in
module = importlib.import_module(f"openspeech.criterion.{file}.{python_file}")
File "/home/python3.8/importlib/init.py", line 127, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "openspeech-main/openspeech/criterion/ctc/ctc.py", line 29, in
from ...tokenizers.tokenizer import Tokenizer
File "openspeech-main/openspeech/tokenizers/init.py", line 69, in
module = importlib.import_module(f"openspeech.tokenizers.{file}.{tokenizer_name}")
File "/home/python3.8/importlib/init.py", line 127, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "openspeech/tokenizers/librispeech/subword.py", line 28, in
from openspeech.datasets.librispeech.preprocess.subword import SENTENCEPIECE_MODEL_NAME
File "openspeech-main/openspeech/datasets/init.py", line 60, in
module = importlib.import_module(f"openspeech.datasets.{file}.{data_module_name}")
File "python3.8/importlib/init.py", line 127, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "openspeech-main/openspeech/datasets/aishell/lit_data_module.py", line 31, in
from openspeech.data.audio.dataset import SpeechToTextDataset
File "openspeech-main/openspeech/data/init.py", line 71, in
module = importlib.import_module(f"openspeech.data.audio.{file}.{module_name}")
File "/home/python3.8/importlib/init.py", line 127, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "openspeech-main/openspeech/data/audio/mfcc/mfcc.py", line 28, in
from ....utils import LIBROSA_IMPORT_ERROR
File "openspeech-main/openspeech/utils.py", line 33, in
from .models import OpenspeechModel
File "openspeech-main/openspeech/models/init.py", line 27, in
from .openspeech_encoder_decoder_model import OpenspeechEncoderDecoderModel
File "openspeech-main/openspeech/models/openspeech_encoder_decoder_model.py", line 29, in
from openspeech.utils import get_class_name
ImportError: cannot import name 'get_class_name' from partially initialized module 'openspeech.utils' (most likely due to a circular import) (openspeech-main/openspeech/utils.py)

[Jxu-Thu]

I make sure that I have installed everything as follows:

---

conda create -n acent_asr python=3.8
source activate acent_asr
sudo apt-get install libsndfile1 -y
#conda install -c pytorch torchaudio -y
#conda install -c conda-forge librosa -y

pip install --ignore-installed llvmlite
pip install librosa
pip install torch-1.6.0+cu101-cp38-cp38-linux_x86_64.whl
pip install torchaudio==0.6.0
pip install sentencepiece
pip install pytorch-lightning
pip install hydra-core --upgrade

install rnnt

git clone https://github.com/1ytic/warp-rnnt
cd warp-rnnt/pytorch_binding
python setup.py install
cd ../..
git clone --recursive https://github.com/parlance/ctcdecode.git
cd ctcdecode && pip install .
cd -
pip install openspeech-core

#sudo mkdir /opt/miniconda/pkgs/
#sudo touch /opt/miniconda/pkgs/urls.txt
#pip install librosa --user

install apex

cd apex
pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./
cd -

pip install -e .

---

[Jxu-Thu]

I try 0.2.1 and find the code can run. so the bug is caused in 0.3.0 version (Master)

[Jxu-Thu]

After carefully checking the codes, I find that deleting

from .models import OpenspeechModel
from .modules import Linear

from the openspeech/utils.py enable the normal running.

Hope to fix the bug for other users :)

[upskyy]

I'll fix the bug. Thank you !