I am building a tiny and intelligent language model.
Even a 5-year-old can run an AI model on his/her 15$ Raspberry Pi.
As builders, I believe we have a responsibility to empower the next generation with the tools they need to tinker and create. When I was growing up around the year 2000, my toolkit included things like Raspberry Pi, Arduino, motors, and Python code—affordable, accessible resources that made experimentation possible.
Today, AI is becoming an essential part of that toolkit. But most kids can’t pay for expensive APIs or buy powerful GPUs to run the latest models. If we want future innovators to thrive, we need to make these tools accessible to them—because one day, those kids will be the ones humanity depends on.
I hope to achieve this with many experiments:
All codes, datasets, model weights, budgets are shared openly. You can replicate/improve these experiments freely.
Training language models are expensive:
I have been fortunate enough to receive a few GPU compute credits. Combined with my ability to abuse startup credits from AWS & Azure. I also have a 8-year-old 1050 Ti gaming GPU.
Combined with the ability to efficiently use compute, I have been able to fund these experiments.
If you would like to donate to the cause, let me know at thomas@comfyspace.tech
So far, most of intelligent models are large. A handful of small models have proven “kinda dumb”. And all cost a lot of money and human resources.
There’s no guarantee of success. I’m not doing this because it’s easy. I do it because I believe it’s important.