AbdelStark

AbdelStark / llm-provable-computer

Public

Can LLMs be provable computers?

llms transformer-architecture zero-knowledge-proofs
19
5
100% credibility
Found Mar 23, 2026 at 19 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Rust
AI Summary

This Rust project implements a transformer-based virtual machine for deterministic program execution with STARK proofs to verify computations without re-execution.

How It Works

1
🔍 Discover Transformer VM

You hear about a clever way to run simple math programs inside a smart machine that can prove it worked exactly right every time.

2
📱 Grab a sample program

Pick an easy example like a Fibonacci calculator from the ready-made programs.

3
▶️ Run your program

Launch it and watch it compute step by step, seeing the final answer like Fibonacci(8)=21.

4
✨ Create a proof

With one command, generate a compact proof that confirms the entire computation happened correctly.

5
✅ Verify instantly

Check the proof super quickly without re-running the program, and it passes every time.

🎉 Proven computation!

You now have a tiny, shareable proof anyone can verify that your program ran perfectly.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 19 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is llm-provable-computer?

This Rust repo builds a transformer VM that executes assembly programs deterministically during the forward pass, turning LLMs into provable computers. You compile simple assembly (fibonacci, factorial) to transformer weights, run via CLI like `cargo run --bin tvm -- programs/fib.tvm`, then generate STARK proofs and verify without re-execution. It solves stochastic LLM outputs by using 2D attention for memory access and algebraic traces for proofs.

Why is it gaining traction?

Unlike typical local LLMs GitHub repos focused on inference, this delivers verifiable execution with four engines (transformer, native, tensor, ONNX) for lockstep checks, plus transparent post-quantum STARKs at O(log n) verify time. Developers dig the TUI debugger, benchmarks, and ONNX export for portable validation—no sampling, always same input/output. It's a fresh take on code LLMs GitHub experiments, blending Percepta's LLM computer idea with Rust's speed.

Who should use this?

ZK researchers prototyping provable offchain compute, blockchain devs needing STARK-wrapped VM traces, or LLM hackers building jailbreaking LLMs GitHub tools with deterministic guarantees. Ideal for Rust fans exploring llms GitHub Copilot alternatives or llms course GitHub projects requiring proofs.

Verdict

Intriguing proof-of-concept for provable LLMs in Rust, with solid CLI/docs/tests, but early at 19 stars and 1.0% credibility—expect rough edges. Grab it for experiments if verifiable transformers excite you.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.