Skip to content

self-adaptive-llms

Adapt an LLM to unseen tasks in two inference steps

Visit official site

What self-adaptive-llms is

The framework is described as Transformer squared from SakanaAI, which adapts large language models to unseen tasks during inference.

What you can do with it

  • Identify task attributes with a scheduler
  • Mix task specific expert vectors dynamically
  • Use reinforcement learning trained vectors

Getting started

  • Clone the repository from GitHub
  • Create a virtual environment and install libraries
  • Install the task evaluator
  • Run the train task expert script

Pros & cons

✓ What we like

  • Real time task adaptation is stated
  • Computation is described as efficient

! What to watch out for

  • Adaptation happens during inference
  • Behaviour depends on mixed expert vectors

FAQ

How does adaptation work?

The material says a scheduler identifies task attributes, then task specific expert vectors are mixed.

How popular is the repository?

The material states 195 stars and 12 forks on GitHub.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review