Hi, my name is Jules. I'm a 15-year-old high school student from France, and I'm teaching myself how to build and train open-source language models.

After saving money for several months, I bought an NVIDIA Jetson Orin Nano Super with only 8GB of memory. Despite the hardware limitations, I used it to learn the full process of developing a language model: preparing data, designing the architecture, training, evaluating, optimizing, and publishing the result.

I've since released two models publicly on Hugging Face:

https://huggingface.co/AZERDSQ/G0-nano-instruct

https://huggingface.co/AZERDSQ/G1-nano-instruct

Both are small models (60-63 million parameters), but they proved I could complete the entire development pipeline independently, on very limited hardware.

I'm now working on the next generation of the project, aiming for models in the hundreds of millions to a few billion parameters. That scale is out of reach on my Jetson. I've been able to borrow occasional access to a shared NVIDIA DGX Spark (GB10, 128GB unified memory) to keep experiments moving, but that access is temporary and not something I can rely on long-term.

My goal is to raise enough to fund the development itself and move toward owning a DGX Spark, or an equivalent GB10-based machine, so I have permanent, independent compute rather than borrowed time on someone else's hardware.

This would let me:

- train and fine-tune larger open-source models without depending on shared access;

- test new architectures and datasets on my own schedule;

- publish benchmarks, checkpoints, and technical findings;

- document both successful and failed experiments;

- make the resulting models available publicly.

The fundraising target is approximately €5,000. If the full target isn't reached, the funds will go toward cloud GPU compute, storage, and smaller-scale training experiments instead — every euro raised goes directly into the project either way, just at a different scale of hardware.

I'll document the use of the funds and the development process publicly: training configurations, compute costs, benchmarks, results, limitations, and failures.

Because I'm under 18, the fundraiser and withdrawal of funds will be managed with the involvement of a parent or legal guardian.

Any support, contribution, technical advice, or introduction to a hardware partner would help me continue learning and building.

Thank you for supporting independent, open-source AI development.

Jules

Hugging Face: https://huggingface.co/AZERDSQ