Announcing Zant v0.1 – an open-source TinyML SDK in Zig!
The first usable version of Zant is finally available!
All the details are in our Hacker News post—check it out and upvote to help us reach more people! 👉 https://t.co/bBMCA8oPd4
Code Generation, in Progress:
We’re developing an automated system to convert ONNX models into Zig code, optimizing for performance and memory usage on embedded devices.
Mathematical Optimizations:
Matrix multiplication performance has been significantly improved using cache blocking, SIMD vectorization, and optimized memory access patterns.
Ending the year with some important technical advancements:
Trained a CNN on the MNIST dataset, achieving state-of-the-art accuracy and loss—a crucial first step toward deploying YOLO, our big milestone for the coming months
Successfully ran a simple NN on a Raspberry Pi Pico
3️⃣ Faster Iteration: Community feedback helps us adapt quickly. Every contribution brings insights that keep Zant evolving to meet real user needs. Innovation thrives when the community is involved! 🚀
🚀 Why is Zant Open Source? 🚀
We're building Zant as a commercial open-source company for 3 key reasons. Here’s why open source is the right path for us: 🧵👇 #OpenSource#Zant#DeveloperTools
2️⃣ Adoption: OSS lets companies try Zant without high costs or risks. This independence builds trust and makes it easy for devs to advocate for Zant as a key tool within their organizations. 💡
🐜 Introducing Zant: an open-source SDK designed to simplify and cost-effectively deploy machine learning models on embedded and edge devices.
Our initial version will feature a static library that transforms NN models into optimized executables for specific devices.