Summarizer Serve

An attempt to learn MLFlow, by inferencing T5 transformer architecture and using it for text summarization. Here we are running the experiment using mlflow and then using mlflow for serving the model.

set up

1. clone the repo, install poetry using pip install poetry
2. run poetry install
3. run poetry run python text_summarizer.py

ML Flow interface

ML Flow lets you track your different experiments along with serving models as APIs. you can check it out -

1. run mlflow ui
2. open localhost:5000 and checkout the run, and you can copy the run_id

Serving the model

You can serve any of your runs using MLFlow using the following command -

mlflow models serve -m runs:/<RUN_ID>/model --port=1234 --no-conda

and you can make curl requests to test the model like :

curl -X POST -H "Content-Type:application/json; format=pandas-split" --data '{"columns":["text"],"data":[["H.P.Lovecraft wrote his best books in Masachusettes."]]}' localhost:1234/invocations