FLOP Network yellowpaper — normative specification for a verified-inference settlement layer. Feedback via Issues.
-
Updated
Nov 18, 2026 - Python
FLOP Network yellowpaper — normative specification for a verified-inference settlement layer. Feedback via Issues.
Provable LLM inference measured on the same hardware: overhead against production serving, proving time, proof size and check time, and what one proof establishes.
Bit-exact, cross-hardware deterministic matrix multiplication using Q16.16 fixed-point arithmetic and SHA-256 verification. Provides identical AI inference results across NVIDIA GPUs, CPUs, and OSs. Essential for ZK-ML, Fintech, and AI Safety. Includes a PyTorch drop-in replacement, API and GPT-2 demo.
Official TypeScript implementation of the IECC protocol. Features Ed25519 signing, RFC 8785 canonicalization, and AI-native verification via MCP.
Research on zero-knowledge proofs of LLM inference, building on FlashProve.
Verifiable inference proof pipeline — cryptographic commit, TEE, and ZK verification tiers anchored on-chain
Signed, auditable receipts for Gemma 4 inference. The provider commits to the execution trace; an auditor challenges random positions and verifies them on a CPU without the weights: Freivalds checks, exact decode, Merkle-bound traces. CommitLLM protocol.
Publicly verifiable proof that an LLM endpoint actually spent the compute you paid for — zero provider cooperation. Closes the effort gap (Hollow-LLM) and the transferability gap (IRIS).
Cryptographic proofs of LLM inference and exact classifier decisions.
Publicly verifiable zero-knowledge proofs for closed-weight LLM APIs—unchanged native-FP serving (vLLM + CUDA Graphs), private weights/architecture/CoT, Lean 4 soundness proofs, and seconds-scale CPU verification.
Requant Network research: SM120 inference-prover support, complete synthetic-model proofs, independent CPU verification and measured limits.
SPOT: a gateway that binds every model response to a published weight commitment, with a Bitcoin-block-sampled subset carrying a zero-knowledge proof. Working mockup.
Protocol record for KORTX: what an inference trace fixes, what each evidence tier covers, and where the coverage stops.
Proof of Edition: verify that an inference API serves the model edition it claims. Signed manifests and receipts, a transparency-logged registry, re-execution audits and the watch.
Verifiable inference: model outputs executed by consensus with receipts any third party can verify. Rust testnet node.
To associate your repository with the verifiable-inference topic, visit your repo's landing page and select "manage topics."