To truly preserve and disseminate access to LLMs we can’t keep mirroring huggingface.
All of the models, weights, datasets, evals, traces, repos hosted on centralized websites are just a DCMA take down away from disappearing
TensorBay solves this.
Our next goal: decentralized inference.
Peer-seeded weights. Distributed model shards. P2P request routing. Signed compute receipts verifiable by the network.
Eventually, users will be able to connect P2P, share compute, and coordinate inference across distributed model shards.
