# Lagrange-Labs/deep-prove

Framework to prove inference of ML models blazingly fast

Repository: https://github.com/Lagrange-Labs/deep-prove
Canonical: https://ross.abutalabs.com/products/deep-prove
Homepage: https://lagrange.dev
Language: Rust
License: NOASSERTION
License Family: other
Topics: ai, ml, zk, zk-snarks, zkml
Last push: 2026-05-31T20:11:18+00:00

## Health v2 (maintenance only)
Score: 60/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 85, release rhythm 35, longevity 47
- inputs: {"age_days": 664, "days_push": 94, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3357, forks 102 (observed 2026-08-28T04:07:57.972594+00:00)

## What it is
DeepProve is a Rust framework for generating zero-knowledge proofs of neural network inference, including end-to-end proving of full LLM forward passes using sumcheck and logup GKR protocols. It includes a core zkml proving library, a client stack with a proof-generation worker and CLI, tensor storage, and telemetry tooling.

## Use cases
- prove that an LLM inference output came from a specific model
- generate zero-knowledge proofs of neural network forward passes
- verify ML model inference results without re-running the model
- benchmark zkML proving performance for GPT-2, Gemma 3, and Llama 2
- run a distributed proof generation network for AI inference
- quantize models and prove MLP, CNN, and transformer inference

## When to choose
- you need cryptographically verifiable AI inference with succinct proofs
- you want the fastest available zkML proving for transformer/LLM models
- you need to prove inference of GPT-2, Gemma, or Llama models end-to-end
- you want GPU-accelerated or horizontally distributed proof generation

## When to avoid
- you just need ML inference without verifiability
- you need a permissively licensed library (it uses a custom Lagrange License)
- your models are outside the supported MLP/CNN/transformer architectures
- you need tiny proof sizes or millisecond-level verification on constrained devices

## Facets
- artifact type: framework
- maturity: active
- function: machine-learning, llm-inference, cryptography, sdk, cli, monitoring
- domain: artificial-intelligence, machine-learning, large-language-models, security, developer-tools
- platform: rust, cli
- tags: zkml, zero-knowledge-proofs, zk-snarks, verifiable-inference, sumcheck, gkr, transformers, quantization, prover-network, cryptography, linux, gpu, docker

## Member repositories
- Lagrange-Labs/deep-prove (main) score 60

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:57.972594+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:40:54.648010+00:00, confidence not recorded.
  - readme: https://github.com/Lagrange-Labs/deep-prove (fetched 2026-08-28T04:07:57.972594+00:00, sha 220a5d51fcbb)
  - homepage: https://lagrange.dev (fetched 2026-08-29T09:33:57.530644+00:00, sha 897c4358c30d)
  - site_page: https://docs.lagrange.dev/ (fetched 2026-08-29T09:33:57.539700+00:00, sha ea4087d2c4e0)
- Data as of 2026-08-30T08:39:29.467469+00:00.