Skip to content

Latest commit

 

History

History
106 lines (71 loc) · 3.71 KB

README.md

File metadata and controls

106 lines (71 loc) · 3.71 KB
Raw

Intentless Policy

This repository contains a starter pack with a bot that uses the IntentlessPolicy. It's a good starting point for trying out the policy and for extending it.

The new intentless policy leverages large language models (LLMs) to complement existing rasa components

Please see the documentation on the Intentless Policy.

Beta State Notice 🚨

The IntentlessPolicy is a Rasa Pro beta feature. The current version of the policy is a prototype that we're releasing to gather feedback. Its use should be limited to development and experimentation. We are iterating on the model regularly, so expect model predictions to also change for the duration of the beta.

The beta allows you to try out the concept and to anticipate and influence our further developments of this feature.

Demo

Webinar demo showing that this policy can already handle some advanced linguistic phenomena out of the box.

The examples in the webinar recording are also part of the end-to-end tests defined in this repo (tests/e2e_test_stories.yml).

Installing this Example Project

The starter pack in this repository is ready to go after a few short steps:

  • Install the dependencies using poetry install

  • set-up the necessary environment variables, e.g. by putting them into a .env file:

    RASA_PRO_LICENSE=eyJhbGciOiJSU...
RASA_PRO_BETA_INTENTLESS=true
OPENAI_API_KEY=sk-...

    Please do not use any quotation for the values otherwise the makefile will not be able to read the values correctly.

  • activate the poetry environment with poetry shell

You can use the commands in the makefile, to train

make train

and run the bot:

make run

Generating End-to-End stories from existing training data

In this starter pack we provide a script that combines existing NLU data and stories to create end-to-end stories. You can use this script to compare different configurations and combinations of policies. For example, you can split your existing data, create more test cases and compare the performance with and without the intentless policy:

First, split your NLU data into train and test sets:

rasa data split nlu -u data/nlu.yml --training-fraction 0.5

Then run End-to-End test creation script from this starter pack. This script combines your NLU test data with your bot's stories:

python scripts/create_test_cases.py -u train_test_split/test_data.yml -s 'data/core/*yml'

Replacing the -s argument with the path to your stories file or a glob to match multiple filenames.

Inspect the generated_test_cases.yml to check that these test cases make sense.

Re-train your model- ensuring you only use the training data from your split for NLU training.

Run your test cases:

rasa test e2e -f generated_test_cases.yml

You can then try different policies, parameters, etc. in your config.yml to compare test performance.

Disclaimer

This software release, including the Intentless Policy feature, is beta software. It’s provided to you only for trial and evaluation, “as is”. You cannot use this beta software release as part of any production environment. Beta software releases are not intended for processing sensitive or personal data. There may be bugs or errors. We may never commercialize some beta software. We disclaim all liabilities, warranties, representations, or covenants related to this beta release.

Your use of the Intentless Policy feature and any other software or material included in this release is subject to the Beta Software Terms.

Please make sure you read these terms and conditions prior to using this beta release.