Provable LLM inference measured on the same hardware: overhead against production serving, proving time, proof size and check time, and what one proof establishes.
Provable inference for LLMs (Anthropic FlashProve style): weight-hiding Merkle commitment + Fiat-Shamir selective opening + CPU-only verifier. Four monotone 0-1 metrics + substitution/tampering detection + soundness sweep. Pure Python, zero deps.
Anthropic's FlashProve (zk proofs of AI inference) on a consumer RTX 5070 Ti: 13 patches, every proof checked by the unchanged upstream verifier, every limit written down. Replications wanted.
Which model am I really talking to? Statistical fingerprinting for LLM endpoints (GLM / DeepSeek baselines). Related: FlashProve provable inference
To associate your repository with the provable-inference topic, visit your repo's landing page and select "manage topics."