Turn Compute Into Intelligence
Train and deploy AI models efficiently on any device
Create neural networks that run better on any hardware, for both
training and inference.
Mabor's novel approach supports dynamic graphs and shapes with the performance of static graphs, thanks to our advanced multiplatform Just-in-Time compiler.
Core Principles
We do not compromise on what matters the most in artificial intelligence.
We believe in maximizing compute across all existing hardware, turning it into impactful intelligence.
Blazingly Fast
Combine multi-level compiler technology with runtime search to find the best way to run any neural network.
Flexible
Build neural networks with dynamic graphs and shapes, without sacrificing performance for both training and inference.
Portable
Run your models on cloud servers, laptops, mobile, from AMD, NVIDIA, Intel, Qualcom, Apple, or any computing device.
The Next-Generation AI Infrastructure
Accelerate Progress with Mabor
Increased Developer Productivity
Less performance tuning, no custom hacks, minimal debugging, no dependency hell. Optimized focus for more iterations.
More Iterations
More feedback accelerates model improvement and drives research breakthroughs, leading to better results.
Better Results
Whether you’re deploying models more cost-effectively, improving customer satisfaction, or increasing your number of publications, it will enhance your work.
THE AI COMPUTE LANDSCAPE
More compute is
only half the story.
NVIDIA’s GPUs are part of the infrastructure behind modern AI. Mabor focuses on the developer side: training models, running inference, and making better use of available hardware.
Dynamo 1.0 brings open-source inference to production. A closer look at the software behind GPU efficiency. See inference on your deviceDAILY CLOSE
Loading end-of-day data
Independent market context. $MABOR is separate from NVDA shares.
MNIST Inference Demo
Draw a number. See the prediction—and how long your browser takes to produce it.
Draw a digit here
Probability result
Loading the local model…
Measured around the model’s inference call only. Excludes loading, drawing and chart rendering; the first run may include runtime warm-up. This measures your browser, not an NVIDIA GPU benchmark.
Built with Rust
Mabor introduces a unique architecture based on tensor operation streams, fully optimized at runtime and auto-tuned for your hardware by a Just-in-Time compiler. This approach relies on Rust's ownership rules to precisely track tensor usage, a feature unattainable without a robust type system. It enables you to create highly dynamic models with the performance of an extensively optimized static graph.
Mabor adheres to the best practices of the Rust programming language, renowned for its reliability, enhancing development speed and helping you achieve your goals more quickly.
Be part of our community
Follow Mabor on X.
We want everyone to feel welcome, no matter their background. Ask questions, share your creations, and stay updated on Mabor's latest releases! 🔥
Shape the future
of AI with
us
Let's get started