A small, self-verifiable corpus of proof-of-inference objects — some honest, some
forged — for testing whether a verifier can tell which model actually produced an
output. Part of TensorCash: a layer-1 whose proof-of-work
is LLM inference, where every generation carries a replayable proof.
Don't trust the provider. Verify the machine.
In GPU teacher-forced replay (150 proofs per class), the verifier accepted… See the full description on the dataset page:
https://huggingface.co/datasets/tensorcash/proof-of-inference-eval.
