Content-Moderation-Tagging

This repo does content moderation and content tagging.

Feature list

This list will be continually updated

1. Moderating videos using CLIP and Nudenet models to identify violence and sexual content
2. Content tagging videos using CLIP and Nudenet models to identify their primary and secondary categories.
3. Wrapped the code base with ML deployment framework, Rayserve.

Set-up .env file for testing in local

There needs to be a .env file with following parameters.

DownloadNumCPUPerReplica=0.2
DownloadNumReplicas=1
DownloadMaxCon=100


PreprocessNumCPUPerReplica=1
PreprocessNumReplicas=1
PreprocessMaxCon=100


NudenetNumCPUPerReplica=0.8
NudenetNumReplicas=1
NudenetMaxCon=100

MFDNumCPUPerReplica=0.8
MFDNumReplicas=1
MFDMaxCon=100

ClipNumCPUPerReplica=0.1
ClipNumGPUPerReplica=0.14
ClipNumReplicas=1
ClipMaxCon=100


ComposedNumCPUPerReplica=0.1
ComposedNumReplicas=1
ComposedMaxCon=100

SnowflakeResultsQueue=content_moderation_tagging-results_dev
RawResultsQueue=content_moderation_tagging-raw-results_dev
AiModelBucket=datalake-dev

Additional variables for internal testing

For DS Team internal testing, we also need to add the following env vars to the .env file:

AWS_ACCESS_KEY_ID=
AWS_SECRET_ACCESS_KEY=
AWS_DEFAULT_REGION=us-east-2

and uncomment these lines in tasks.py:

# from dotenv import load_dotenv

# load_dotenv("./.env")

To prepare the conda environment to test the script:

pip install -r requirements.txt
pip install git+https://github.com/openai/CLIP.git --no-deps
pip install -U "ray[default]==1.11.1"
pip install "ray[serve]"
pip install pytest

For use in g4dn.2xlarge instance, use the following variables instead

DownloadNumCPUPerReplica=0.2
DownloadNumReplicas=1
DownloadMaxCon=100


PreprocessNumCPUPerReplica=1
PreprocessNumReplicas=1
PreprocessMaxCon=100


NudenetNumCPUPerReplica=0.8
NudenetNumReplicas=2
NudenetMaxCon=100

MFDNumCPUPerReplica=0.8
MFDNumReplicas=2
MFDMaxCon=100

ClipNumCPUPerReplica=0.1
ClipNumGPUPerReplica=0.14
ClipNumReplicas=4
ClipMaxCon=100


ComposedNumCPUPerReplica=0.1
ComposedNumReplicas=1
ComposedMaxCon=100

SnowflakeResultsQueue=content_moderation_tagging-results_dev
RawResultsQueue=content_moderation_tagging-raw-results_dev
AiModelBucket=datalake-dev

Instructions (Docker)

1. Ensure there are environment variables or .env file, see section above for environment variables.
2. Ensure GPU for docker is enabled. See section below.
3. Once the container is able to detect the GPU, we can follow the normal process of

docker-compose build
docker-compose up

Enabling GPU for Docker

To enable the GPU for Docker, make sure Nvidia drivers for the system are installed. Refer link for details

Commands which can help install Nvidia drivers are:

unbuntu-drivers devices
sudo ubuntu-drivers autoinstall

Then nvidia-docker2 tools needs to be installed. To install follow the below instructions. Refer link for details

curl https://get.docker.com | sh   && sudo systemctl --now enable docker
   distribution=$(. /etc/os-release;echo $ID$VERSION_ID)    && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -    && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list

sudo apt-get update
sudo apt-get install -y nvidia-docker2
sudo systemctl restart docker

Pytest

1. Test if the code is working as expected. Firstly on terminal, do:

ray start --head --port=6300

2. Then, deploy the ray services:

python serve_tasks/tasks.py

3. Finally you can conduct the pytest and shutdown the ray cluster.

cd ./test
python -m pytest test_ray_deployments.py 
ray stop --force

Code structure

To be updated.