As part our 2024 Global engineering AI Combinator Hackathon the idea was born to create AI aided support mechanism for our platforms like OpenShift (including SNO), RHEL (incl RHDE) and MicroShift. This is what we call AIOps - an AI aided operations support for preventative, pre-emptive and reactive break fix support.
The process was to follow the recommended way for any AI project which is to work through the solutioning end-to-end and not just focusing on one particular area. So we went through Data Engineering -> Model Training -> Model Deployment and running inference.
We worked through several architecture iterations, with each iteration making it simpler in order to be able to finish a run-through end-to-end.
We ended up with this architecture which the code of this repository represents:
To get things up and running you need to clone this repo, download the model from hf (hugging face) as descried in the model operations folder.
The fine-tuning notebook was run on collab, which we recommend to do either with a T4 or a A100 GPU (you will need to buy credits for that).
If you get stuck with the model you can download the quantized version directly from hf from https://huggingface.co/aspanner/llama-2-7b-aiopsfinetuned-q8_0-gguf
- I recommend to create and activate a python virtual environment on your laptop (here: Fedora 39) locally, as llama.cpp, huggingface, streamlit an other components might clash with things you already got installed. To do that run (aiopsVE is the name of my virttual environment(VE): ' python3.12 -m venv aiopsVE ' and then activate it with ' source aiopsVE/bin/activate'
if you do a 'pip list' you should see an empty list, meaning no pip packags are install yet.
❯ pip list
Package Version
------- -------
pip 23.2.1
[notice] A new release of pip is available: 23.2.1 -> 24.0
[notice] To update, run: pip install --upgrade pip
Your zsh prompt should show the activated virtual environment name - below on the right hand side:
Then upgrade your pip via pip install --upgrade pip
Then run pip install -r requirements.txt
The main appliation which brings the UI to life is in app.py. If you work through the code you can see how we use langchain and Llama.cpp to load and interact with the model via the llama2 specific prompt template.
Then run streamlit run app.py from this folder and you should be greeted by a UI like this:
Other (non mandatory to run the WebUI) components are the logfile scraper and other data engineering script we used (in the data folder). There is also a helper python script that helps you to download the model.
I also committed the MicroShift logfiles if you'd like to improve the data engineering, add more data or remove more duplicates, timestamps or any other noise.
#Slack Integration I've moved all the slack integration work into the subfolder 'slack_integration'
To send slack message :
curl -H 'Content-Type: application/json' --data '{"slack_id":"Hello, send aiops message!"}' -X POST http://localhost:5000/send-to-slack