dreamzero0

dreamzero0 / dreamzero

Public

Code to pretrain, fine-tune, and evaluate DreamZero and run sim & real-world evals

dreamzero0.github.io
969
41
100% credibility
Found Feb 05, 2026 at 360 stars 3x -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

DreamZero provides a pretrained AI model that jointly predicts robot actions and future videos from observations, enabling zero-shot performance on unseen tasks through a distributed WebSocket inference server on multi-GPU hardware.

How It Works

1
🔍 Discover DreamZero

You find DreamZero, an AI helper that predicts robot moves and future sights from simple observations, perfect for new tasks without extra training.

2
📦 Set up your space

Create a quiet workspace on your powerful computer with multiple graphics cards, ready for heavy thinking.

3
📥 Grab the brain

Download the ready-trained AI brain from a trusted sharing site to your folder.

4
🚀 Wake it up

Start the smart server with a simple command, watching it hum to life across your graphics cards.

5
🧪 Test the magic

Run a quick check with the test tool, seeing it respond with actions and video previews in seconds.

6
🎮 Play in simulations

Connect to robot playgrounds like simulations, giving instructions and watching it act zero-shot.

✅ Zero-shot success

Your robot follows new commands perfectly, generating future videos and smooth actions effortlessly.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 360 to 969 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 dreamzero?

DreamZero is a Python GitHub repository for loading pretrained world action model checkpoints—like the 14B DreamZero-DROID—and running zero-shot policy inference on robotics sims such as DROID and Genie 3.0. It spins up a WebSocket server for distributed multi-GPU inference (GB200/H100 optimized), delivering joint action predictions and MP4 video generations via simple CLI commands, perfect for quick code github python evals without retraining.

Why is it gaining traction?

It delivers sub-second inference with DiT caching (~0.6s on GB200), beating typical world model latencies, plus seamless sim eval scripts and RoboArena integration for real-robot tests. Developers grab it for zero-shot task performance on unseen envs, video saving for analysis, and easy API hosting—no fuss with custom loaders or single-GPU bottlenecks.

Who should use this?

Robotics researchers benchmarking AI policies in DROID/Genie sims, sim eval teams needing fast WebSocket endpoints for distributed testing, or hardware labs with H100 clusters prototyping zero-shot manipulation like "pick up the cube."

Verdict

Grab it for GPU-heavy sim evals if you match the prereqs (PyTorch 2.8+, multi-GPU); the 591 stars signal buzz, but 1.0% credibility score flags early-stage risks—skim the README for setup gotchas before production.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.