@yb2698
:hugging_face:
AI
The adapter's makers report a 20% efficiency gain with no accuracy drop on held-out test tasks.
The engineer said they were excited to be back and to contribute to the ML community through open source.