Research into models that are easier to inspect and change.
Qriton HLM is our research line in energy-based (Hopfield) models for text, images, audio, 3D and sensor data. Approved teams can use it as one model option among several.
Why we research energy-based models.
Classical energy-based (Hopfield) networks store patterns as stable states. Our research asks whether that structure can make specific model behaviour easier to inspect, test and change.
In Qriton products, HLM is used only where it suits the task.
What it does.
Research models
Models for text, images, audio, 3D and sensor data, available to approved teams.
Inspection
Energy Language, published as the qriton-hlm package, lets researchers inspect model weights.
Reversible edits
Edits made with Energy Language can be reverted exactly. They are intended for research use.
How it works.
Define the task
Agree the decision, the data and the measures of success with your team.
Compare
Evaluate HLM against standard models on your own data.
Decide
Choose the model for production on the measured results.
GPU hours on Leonardo, through the EuroHPC AI Factory
At a glance.
- Model family
- HLM5 for language; HLM-Vision, HLM-Spatial and HLM-Audio; HLM-Micro for activity, keyword and gesture recognition.
- Inputs
- Text, documents, images, audio, 3D and sensor streams.
- Deployment
- Your servers, your private cloud or air-gapped sites.
- Research tools
- Energy Language: inspection of model weights, and edits that can be reverted exactly.
- Access
- Controlled early access for approved teams. Several research models, including HLM5-3B Base, are public on Hugging Face.