paradigm-client 0.5.2

Python client for LightOn Paradigm

paradigm-client

Python client for LightOn Paradigm Large language Model.

Installation

You can install this package from PyPi with:

pip install paradigm-client

Or from source:

pip install -U git+https://github.com/lightonai/paradigm-client.git

Once the package is installed, make sure to define environment variables PARADIGM_API_KEY to your API key, e.g. by adding the following line to your .bashrc

export PARADIGM_API_KEY="<your api key>"

Quick Start

Using paradigm-client is pretty simple, here are a code example to show you how you can use it.

from paradigm_client.remote_model import RemoteModel

model = RemoteModel(model_name="llm-mini")

print(model.create("Hello, I am").completions[0].output_text)

Logging a feedback into Paradigm

After using the Create endpoint of Paradigm, you can log a feedback about it.

Feedback data is expected to be in a dictionary format with a key being one of "flag", "value", "tag" and "comment".

• flag: used for boolean feedbacks
• value: used for float feedbacks
• tag: used for short text feedbacks
• comment: used for free text feedbacks

Here are the different steps to log a feedback:

1. If the feedback type you want to use does not exist on Paradigm yet, go to the admin interface of Paradigm and create it.
2. Get the rating_id of the feedback type you want to use.
3. Instantiate a RemoteModel object.
4. Get the completion_id from the response of a RemoteModel.create() call.
5. Call the log_feedback() method of your RemoteModel object with the rating_id, the completion_id related to your feedback and your feedback data.

Important note: The API Key used to generate the completion_id and to send the feedback must authorize the logging of its requests from the admin interface of Paradigm.

Access to LightOn Paradigm

Learn more about Paradigm: https://www.lighton.ai/fr/paradigm. For a list of use cases: https://www.lighton.ai/fr/blog/ai-use-case-5.

Known Issues

If you find that a RemoteModel instantiation and subsequent completions are unusually slow, it may be that your network does not support IPv6. Try disabling IPv6 and see if that helps.