Release Notes on v0.4.4:

What has changed?
- When an exception is raised during model inference, it is checked whether the exception indicates that the passed text prompt is too big for the model to handel. If this is the case, a 413 HTTP code is returned with the message indicating the text prompt is too big.

release notes on v0.4.3:

What has been fixed:
- There has been a change made to the GPT J 6B model name for referencing it, which caused downloading the model params to fail. This has been fixed now.
- Concurrency limit has been increased to max. 15 concurrent requests.

Bugfix v0.4.2:

When calling the variable-extraction API and the provided text had too few characters in it, the response did not have the expected fields in it. This is now fixed and the API response in success case is consistent.

Release notes v0.4.0
New endpoint to perform lemmatization on a list of terms is now available.
Also, the extract-variables endpoint now returns not only the already lemmatized extracted variables, but also the unlemmatized.

RELEASE NOTE v0.3.0

This version introduces support for using the gpt-j-6B model to extract variables from a text input given some examples.
For further details consult the README of this project.

Also, the deployment scripts have been refactored to be more robust and extendable. The required models are now downloaded into a model cache used by the app before the uwsgi server is started. This speeds up start time.
It does still take around 2min for the server to start up, this is due to loading the models into memory, which takes some time. Currently, this happens not asynchronous, i.e. the app is not ready to server requests as long as it is loading the models into memory.

Note that this version still does not support multiprocessing, e.g. there might be a limitation to the number of concurrent requests the API is able to handle.

minor version releas to v0.2.0 includes addition of ocr model with corresponding endpoint to retrieve text from an image. Note, that in this version, the uwsgi server is limited to run in 1 single process due to cuda multiprocessing / forking issues.

first minimal running version providing api server with /api/ping operation. This version also includes scripts to deploy & run the service on a remote machine.