YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Dexter provides a straight line from your first API call to autonomous AI research loops and production endpoints. Designed for
Gemma-4-31Band optimized for cheap, high-performance cloud GPUs.
๐ฎ The Dexter OS (Start Here!)
The easiest way to interact with the entire ecosystem, run labs, and manage your cloud swarm is through the interactive Dexter OS CLI.
# Clone the repository
git clone https://github.com/lyffseba/dexter.git
cd dexter
# Boot up the Dexter OS
python3 dexter_cli.py
The OS will guide you through running the local Mojo labs and deploying the cloud inference endpoints automatically.
๐๏ธ Ecosystem Structure
dexter/
โโโ scripts/ # Deployment scripts for Inference (Modular MAX)
โโโ labs/ # Interactive Tinkering & Fine-Tuning Labs
โโโ docs/ # ๐ Dedicated documentation per lab
โโโ 00_getting_started.ipynb # ๐ข START HERE: Data, Tokenizers & Mojo setup
โโโ 01_inference_test.ipynb # ๐ฌ API Prompting & Generation
โโโ 02_qlora_finetuning.ipynb # ๐ง QLoRA fine-tuning 31B models
โโโ autoresearch/ # ๐ค Autonomous AI Agent Research Loop (PyTorch -> Mojo port)
๐ The Best Platform: Modular Cloud
For cost-to-performance, Modular Cloud is currently the best platform. They have the fastest inference speeds for Gemma-4 using MAX, their GenAI native modeling & serving framework, completely outperforming vLLM on both NVIDIA and AMD platforms.
While Modular Cloud is ideal for inference and deployment, RunPod (RTX 3090/4090) remains the best value for bare-metal SSH access required for the autonomous PyTorch training labs.
๐ ๏ธ Getting Started on Modular Cloud
1. Account Setup
- Go to the Modular Console and sign up.
- Modular Cloud gives you a straight line from the first API call to a production endpoint.
2. Running Gemma-4-31B (Inference)
Inside your cloud environment, install the Modular CLI and MAX framework:
bash scripts/01_setup_inference.sh
Then start the MAX engine server:
bash scripts/02_start_server.sh
Now you can interact with the model via Python or using any OpenAI-compatible UI!
๐ฌ Tinkering & Fine-Tuning Labs
Check the labs/ directory for Jupyter Notebooks designed to let you play with the model locally and in the cloud.
- [START HERE]
labs/00_getting_started.ipynb: An interactive, local introduction to AI data, tokenizers, and compiling Mojo. Start here to learn the fundamentals! labs/01_inference_test.ipynb: Test prompting and generating text against your MAX endpoint.labs/02_qlora_finetuning.ipynb: Learn how to fine-tuneGemma-4-31Bon your own custom data using LoRA.labs/autoresearch/: An autonomous AI agent loop that tweaks LLM architectures. Includes the ongoing project to port the framework from PyTorch to Mojo ๐ฅ & MAX.