Training and unlimited API inference for Small Language Models

For a long time, I’ve had the feeling that generative AI models, for most of the tasks they’re used for, don’t actually need the oversized capabilities offered by frontier models. On top of that, they’re often prohibitively expensive when accessed via API. An open-source model with 4–14 billion parameters, multimodal capabilities, and the right harness of tools, data, and knowledge is more than enough for most use cases. With this idea in mind, I decided to build, together with my team, a simple platform that allows users to fine-tune and deploy these lightweight models quickly and without any coding. After several iterations, we have managed to develop what we believe is the optimal framework for anyone to deploy and customize a small 4B or 9B model, either through fine-tuning or RAG. And what’s most important t me, you can download the weights, run them locally, and own the IP and the property of the model, unlike in many cases where big vendors lock you in. We’ve also managed to sign an agreement with a local data center, and we now have our own GPUs. This means training and inference are not only extremely fast, but we can also offer unlimited inference. If you need a customized, affordable model with unlimited API inference, I think this could be really useful for you. There’s a free tier, and if you try it out and give us some feedback, I’ll give you the subscription plan for free for 6 months. Since we have our own GPUs in our data center, our infrastructure costs are fixed. I’m really happy with how it has turned out, and I use these models for many of my own projects. You can access the platform at neuro-block.com Would love to hear your thoughts in the comments.