Spaces:
Running
on
Zero
Apply for community grant: Personal project (gpu)
Dual-Input Shunt Adapters meant to translate two completely different AI model's encodings into something guided with the knowledge of the other, and useful for either of those two models at runtime.
I've written multiple articles on the process, released the training notebook, introduced a series of solutions for problems, and found routes using innovative and recent AI discoveries.
As I finetune nomic-BERT the archetype will expand to a 26 shunt manifold, dedicated to translating a single clip and a single bert inference into a multi-stage timestep aware diffusion clip encoding.
I require additional hardware. I primarily use huggingface due to the ease and convenience of many elements and have provided multiple datasets along with the weights for the adapters.