Skip to content

README.md

Latest commit

 

History

History
65 lines (47 loc) · 2.7 KB

README.md

File metadata and controls

65 lines (47 loc) · 2.7 KB

Padatious Intent Cache

Pre-trained Padatious intent classifiers to significantly accelerate the first boot and skill loading of OpenVoiceOS.

The Problem

Padatious was originally designed for small datasets with a few sentences per intent. However, as the number of intents and languages grows, training times have increased drastically.

⚠️ Depending on the number of skills installed padatious intent training can slow down a raspberri pi to a crawl!

This is particularly noticeable in languages like Catalan and Galician, where utterance examples include many alternatives/optionals to account for verb tenses and grammatical gender. Large datasets can lead to training times exceeding 10 minutes for a single intent:

DEBUG - Training ask.intent with 110,784 positive + 20,212 negative samples...
... a few minutes later ...
DEBUG - Training ask.intent finished!

The Solution

This repository uses an automated utility (gen_cache.py) to bootstrap the training process. By simulating a complete ovos-core environment, we generate binary intent models for every supported language and configuration variation.

When these pre-trained models are present, Padatious skips the expensive training phase unless the local training data ( .intent files) has changed.

Technical Details

The gen_cache.py utility generates a matrix of 8 different configuration profiles to ensure compatibility with various user setups:

Feature Variations
Languages 12 (en, es, pt, fr, de, it, nl, da, ca, gl, eu)
Domain Engine Enabled / Disabled (Two-stage classification)
Stemming Enabled / Disabled
ASCII Normalization Enabled / Disabled

How it works

  1. Mock Environment: Uses FakeBus and SkillManager to load skills without a running messagebus.
  2. Pipeline Injection: Forces the ovos-padatious-pipeline-plugin-high pipeline.
  3. Lifecycle Monitoring: Listens for mycroft.ready.check to verify all Padatious threads have finished writing to disk before cycling to the next config profile.

Usage

  1. Download the intent_cache directories from the latest release/build.
  2. Place the folder in your XDG data directory (usually ~/.local/share/mycroft).
# Example structure
~/.local/share/mycroft/intent_cache/
├── en-us/
├── es-es/
└── ...

~/.local/share/mycroft/intent_cache_domain/
├── en-us/
├── es-es/
└── ...