Open models,
sourced and fine-tuned properly.
Hugging Face is where we source open-weight models and the Transformers library we fine-tune them with — the foundation underneath any project where a client needs a model they can own and run themselves.
Serving UK and EU clients: GDPR, EU AI Act, and data residency are covered on our Trust & Safety page.
The library and the hub,
not just a dataset host.
Hugging Face is two things we use together: the Hub, where open-weight models like Llama are published and versioned, and the Transformers library, the standard tooling for loading, fine-tuning, and running those models. It's the foundation layer underneath any project involving an open-source model.
We treat model selection from the Hub as a real evaluation step — license terms, benchmark performance on tasks similar to the client's, and community support — not a default pick.
Selection before
fine-tuning.
Picking the right base model matters more than most fine-tuning technique choices.
Checking that a model's license actually permits the client's intended commercial use before committing to it.
Using Unsloth or Axolotl on top of Transformers for LoRA/QLoRA fine-tuning without full-scale training costs.
Fine-tuning runs tracked in Weights & Biases so results are comparable and reproducible.
Where this fits
The foundation layer underneath our open-source model and fine-tuning work.
Before you
book a call.
The questions we get asked most about Hugging Face and open-source models — answered straight, no sales pitch.
Is every model on Hugging Face safe to use commercially?
No — licenses vary significantly, and some restrict commercial use or require attribution. We check license terms as a real step before committing to a base model, not an afterthought.
Do you fine-tune every model you source from Hugging Face?
No — many projects use a base model as-is, with prompting and retrieval doing the customization work. Fine-tuning is a deliberate step when prompting alone doesn't get the behavior a project needs.
What's the actual deliverable here?
A model selected, optionally fine-tuned on a client's data, and deployed with production-grade serving — not just a notebook that ran once.
Tell us what
you're trying to build.
Book a 30-minute call — we'll tell you honestly whether an open-source model fits your project, and what fine-tuning would actually take.