The code for method-level-summary with SparseCoder breaks python packages

[MariaDziuba]

Script that causes the problem: $(PROJECT_ROOT)/Method-Level-Summary/SparseCoder/run.sh
Dockerfile:

FROM pytorch/pytorch:2.0.0-cuda11.7-cudnn8-runtime
WORKDIR /app
COPY ./requirements.txt .
RUN apt update && \
    apt-get install git -y && \
    pip install -r requirements.txt

##############################################################################
# Installation/Basic Utilities
##############################################################################
RUN apt-get update && \
        apt-get install -y --no-install-recommends \
        software-properties-common build-essential autotools-dev \
        nfs-common pdsh \
        cmake g++ gcc \
        curl wget vim tmux emacs less unzip screen nano git \
        htop iftop iotop ca-certificates openssh-client openssh-server \
        rsync iputils-ping net-tools sudo \
        llvm-9-dev

requirements.txt

aiohttp==3.8.4
aiosignal==1.3.1
async-timeout==4.0.2
attrs==22.2.0
beautifulsoup4==4.11.2
bs4==0.0.1
certifi==2022.12.7
charset-normalizer==3.0.1
clearml==1.9.2
filelock==3.9.0
frozenlist==1.3.3
fsspec==2023.1.0
furl==2.1.3
huggingface-hub==0.12.1
idna==3.4
joblib==1.2.0
jsonschema==4.17.3
lightning-utilities==0.7.1
multidict==6.0.4
numpy==1.24.2
orderedmultidict==1.0.1
packaging==23.0
pandas==1.5.3
pathlib2==2.3.7.post1
Pillow==9.4.0
psutil==5.9.4
PyJWT==2.4.0
pyparsing==3.0.9
pyrsistent==0.19.3
python-dateutil==2.8.2
pytorch-lightning==1.9.3
pytz==2022.7.1
PyYAML==6.0
regex==2022.10.31
requests==2.28.2
scikit-learn==1.2.1
scipy==1.10.1
six==1.16.0
soupsieve==2.4
threadpoolctl==3.1.0
tokenizers==0.13.2
torch==2.0.0
torchmetrics==0.11.3
tqdm==4.64.1
transformers==4.28.1
typing_extensions==4.5.0
urllib3==1.26.14
yarl==1.8.2
boto3==1.26.94
datasketch
typer
fairscale
nltk
openai
peft==0.2.0
deepspeed==0.8.3
pygments
datasets==2.11.0
bitsandbytes==0.38.1
sentencepiece==0.1.97
notebook==6.5.4
jupyterlab==3.6.3

Error:

Traceback (most recent call last):
  File "/home/modzyuba1/SparseCoder/Method-Level-Summary/SparseCoder/run.py", line 44, in <module>
    from lsg_converter import LSGConverter
  File "/opt/conda/lib/python3.10/site-packages/lsg_converter/__init__.py", line 1, in <module>
    from .converter import LSGConverter
  File "/opt/conda/lib/python3.10/site-packages/lsg_converter/converter.py", line 6, in <module>
    from .bart.convert_bart_checkpoint import *
  File "/opt/conda/lib/python3.10/site-packages/lsg_converter/bart/convert_bart_checkpoint.py", line 1, in <module>
    from .modeling_lsg_bart import *
  File "/opt/conda/lib/python3.10/site-packages/lsg_converter/bart/modeling_lsg_bart.py", line 4, in <module>
    from transformers.modeling_attn_mask_utils import _prepare_4d_attention_mask, _prepare_4d_causal_attention_mask
ModuleNotFoundError: No module named 'transformers.modeling_attn_mask_utils'

OS version: Ubuntu 20.04.2 LTS (GNU/Linux 5.4.0-144-generic x86_64)

Steps to reproduce the error:

1. Build an image, run container
2. Pip install transformers ==4.37.1
3. Then run Python in interactive mode:

python
>>> import transformers
>>> transformers.__version__ 
>>> 4.37.1

4. Run $(PROJECT_ROOT)/Method-Level-Summary/SparseCoder/run.sh in the docker container and get the error
5. Do step 3 again
   >>> 4.32.0
6. Do step 2 again
   transformers==4.37.1 successfully installed
7. Do step 3 again
   >>> 4.32.0

Expected behavior:
This code should work, because missing modules are presented in transformers 4.37.1. https://github.com/huggingface/transformers/blob/v4.37.1/src/transformers/modeling_attn_mask_utils.py
The installed version of transformers is 4.37.1. After your code runs, Python packages become broken.
The transformers version does not change after the second pip install, but should