VAGOsolutions

VAGOsolutions / SauerkrautLM-Doom-MultiVec

Public

A tiny 1.3M parameter model that plays DOOM, outperforming LLMs up to 92,000x its size.

16
3
100% credibility
Found Apr 07, 2026 at 16 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

A compact AI model that interprets simplified text versions of DOOM game screens to choose actions, achieving superior performance to massive language models in gameplay benchmarks.

How It Works

1
🎮 Discover the tiny game master

You stumble upon a fun project where a super-small AI brain learns to play the classic shooter game DOOM better than huge smart assistants.

2
📹 Watch the thrilling demo

Click the video to see this little AI zoom around, shoot enemies, and rack up kills in real-time action.

3
🖥️ Set it up easily

Follow simple steps to get the AI player ready on your computer, no hassle.

4
▶️ Launch and watch it play

Hit play and enjoy seeing your tiny AI defend the center, turning and firing like a pro.

5
Explore more options
👀
Keep watching AI

Sit back as it battles waves of enemies on its own.

🎤
Record your plays

Play yourself and capture moves to teach the AI your style.

6
🏆 Challenge big AIs

Run quick contests to prove your mini player outsmarts giants.

🪙 Tiny AI triumphs!

Celebrate as your small creation scores way more kills and survives longer than the big ones.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 16 to 16 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 SauerkrautLM-Doom-MultiVec?

This Python repo delivers a 1.3M parameter model that plays DOOM in VizDoom, converting frames to ASCII art plus depth maps for real-time action prediction (shoot, move, turn). Trained on 31k human demos, it scores 178 frags over 10 episodes in defend_the_center—more than GPT-4o-mini, Nemotron-120B, and others combined. Run demos via CLI scripts to watch it play, benchmark vs LLMs, or train your own on custom data.

Why is it gaining traction?

Like github tiny shakespeare or tiny c compiler, it's a github tiny model punching way above its weight: outperforms LLMs 92,000x larger at DOOM with 31ms CPU inference (35 FPS), no GPU needed. Depth-aware tokens and human-play data enable aggressive play (LLMs just evade). HF model/dataset uploads, benchmarks, and Pi-ready export make testing instant.

Who should use this?

Game AI hobbyists scripting bots for VizDoom scenarios. Embedded ML devs deploying vision-like tasks to ARM (RPi Zero). Researchers probing tiny models vs LLMs in control tasks, or forking sauerkrautlm multivec for custom games.

Verdict

Grab it for the LLM-beating demo and paper—solid docs/scripts despite 16 stars and 1.0% credibility score. Early but forkable; train your variant before it matures.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.