wq-will

wq-will / SimpleTES

Public

A general framework for strategically scaling evaluation-driven discovery loops, discovering state-of-the-art solutions on 21 open-ended problems.

www.wizardquant.comwillsimpletes
47
5
100% credibility
Found Apr 25, 2026 at 47 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
C++
AI Summary

SimpleTES is an AI-powered search engine that iteratively proposes, evaluates, and refines solutions to optimization problems in fields like quantum circuits, GPU kernels, and mathematics.

How It Works

1
🔍 Discover SimpleTES

You stumble upon this clever tool on GitHub that helps solve tough puzzles in science and math by smartly trying ideas over and over.

2
📦 Get it ready

With a simple command, you install it on your computer, like adding a new app.

3
🧙 Pick a challenge

An easy wizard guides you to choose a puzzle, like packing circles or optimizing quantum circuits, from ready-made examples.

4
🔗 Link your AI friend

You share a password with a smart AI service so it can help brainstorm solutions.

5
🚀 Launch the search

Hit go, and watch as it proposes ideas, tests them instantly, and refines the best ones in a clever loop.

6
📊 See it improve

A dashboard shows progress, with scores climbing as better solutions emerge before your eyes.

🏆 Claim your discovery

You get the top solution files, ready to use or share, having unlocked new records in your chosen challenge.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 47 to 47 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 SimpleTES?

SimpleTES is a Python framework for scaling test-time compute on evaluation-driven discovery loops, turning open-source LLMs into SOTA hunters on 21 hard problems like qubit routing, GPU kernel tuning, and circle packing. You point it at an evaluator script and seed program; it runs parallel chains of propose-evaluate-refine, tuning budget across exploration width, refinement depth, best-of-K, and history selectors for max score. CLI and interactive wizard handle setup, resuming from checkpoints, with LiteLLM for any model.

Why is it gaining traction?

It skips "just generate longer" for smarter allocation on real evaluators—timers, verifiers, compilers—echoing AlphaZero general search but for general framework problems where models need world feedback. Built-in tasks deliver paper SOTA artifacts immediately, and "bring your own evaluator" means no per-domain hacks. Community general GitHub buzz builds from reproducible runs on quantum metrology or combinatorial bounds.

Who should use this?

AI researchers chasing SOTA on evaluator-gated tasks: kernel optimizers timing trimul or cumsum, quantum devs compiling circuits, math folks maximizing Hadamard determinants or sum-difference ratios. Perfect for heuristic contest pros or data scientists fitting scaling laws without custom RL.

Verdict

Solid pick for eval-heavy discovery—47 stars and 1.0% credibility signal early promise with strong docs, tests, and artifacts; expect rough edges on heavy setups like Docker evals. Fork now if it matches your general framework needs.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.