Open Neural Network Exchange - Generative Artificial Intelligence

Graiphic is developing ONNX GenAI, the first open, efficient, and sovereign runtime layer for generative AI models, based on LabVIEW, ONNX, and ONNX Runtime. Unlike current GenAI tools that rely on fragmented formats like GGUF or safetensors, ONNX GenAI provides a unified, inspectable execution stack that enables model conversion, optimization, and deployment across any hardware platform: CPU, GPU, NPU, FPGA, or edge.

ONNX GenAI is a component of ONNX Runtime, an open-source inference engine released under the MIT license. Graiphic is the only European contributor to ONNX GenAI and plays a key role in adapting it to edge, embedded, and sovereign AI environments.

Today, most LLMs are tied to closed runtimes and lack graph-level standardization, making deployment fragile and energy-intensive. ONNX GenAI introduces a portable abstraction layer, supporting quantization, pruning, and dynamic hardware dispatching. Integrated into SOTA, our visual AI suite based on LabVIEW, it allows engineers and scientists to orchestrate, fine-tune, and deploy LLMs without writing code.

The technology is at TRL 6. Graiphic has already validated key transformer components such as rotary attention, softmax, and quantization on real hardware, outperforming PyTorch baselines in latency. The next step is full model support, documentation, and industrial adoption.

ONNX GenAI enables trustworthy, efficient, and sovereign GenAI across the edge to cloud continuum. It supports the EU’s Green Deal, GenAI4EU, and digital sovereignty goals. We request €1.7M€ in grant funding to scale adoption and ensure this strategic infrastructure is anchored in Europe.