FLAS: Flow-based Activation Steering for Inference-Time Intervention
Official organization for FLAS.
FLAS is the Flow-based Activation Steering which learns a general, concept-conditioned velocity field that transports unsteered activations to steered ones for the inference-time intervention of LLMs.
Links
- Code Repository: FLAS
- Project Homepage: https://flas-ai.github.io
- Paper: Read more on arXiv
- Demo: Try FLAS on Hugging Face Space