Skip to main content
Continual Learning as a Service (CLaaS) personalizes open-source language models in real time using SDPO (Self-Distillation from Preference Optimization). Every piece of user feedback triggers a single distillation step that updates the model’s LoRA adapter, without forgetting what it already knows.

​
Explore the docs

Quick Start

Get up and running with CLaaS in minutes using Local, Tinker, or Modal backends.

Training Backends

Compare Local GPU, Tinker SDK, and Modal cloud backends for SDPO training.

Evaluation

Run automated feedback loops to measure preference compliance and detect collapse.

Dashboards

Built-in web dashboards for training metrics and eval results served by the CLaaS API.
RuntimePython >= 3.11, uv
Repositorygithub.com/kfallah/CLaaS
LicenseMIT