Product
Solutions
Resources
Fine-Tuning LLM
Fine-tune open source models for your specific needs.
Fine-tuning takes an open model and teaches it your tone, your domain and your task. At HostYourAI this is project work, scoped and planned together, not a self-service button: your training data stays in the EU throughout, and the resulting model deploys to your own dedicated GPU instance served with vLLM. A concrete example is Loes, the sovereign Dutch model brand, fine-tuned from an open base for natural Dutch.
Training data is usually the most sensitive data a company owns: real conversations, real documents, real decisions. Sending it to a training pipeline outside Europe undoes years of careful data governance in one upload. Fine-tuning inside European datacenters keeps the whole loop, from dataset to deployed weights, under EU jurisdiction.
Why LLM fine-tuning in the EU
Your weights, your sovereignty
A fine-tuned open model is an asset you own. Unlike a proprietary API that can change pricing, behavior or availability overnight, your weights sit on infrastructure you control and can be redeployed anywhere vLLM runs. That is digital sovereignty in a practical form, and it is why open bases like Llama, Qwen and Mistral from the model catalog are the starting point.
Training data is the crown jewels
Datasets built from customer interactions almost always contain personal data, so the GDPR applies to the training run itself, not just to inference. Keeping preparation, training and evaluation in multiple European regions, with AES-256 encryption at rest and a Data Processing Agreement in place, means your legal basis does not depend on third-country transfer mechanisms.
Jurisdiction over the whole lifecycle
US cloud providers fall under the CLOUD Act, which can compel disclosure of data regardless of where the servers stand. A training pipeline on European infrastructure under a European provider removes that question entirely, which regulated sectors in particular tend to require.
EU Inference Router
‹ Insight 1 of 28 ›When fine-tuning pays off, and when it does not
Fine-tuning is the heaviest tool in the box, so reach for it last. If the model lacks knowledge, retrieval is cheaper and easier to update: the guide Build RAG on EU GPUs shows how a linked knowledge base injects your documents into completions without touching the weights. If the model just needs steering, a good system prompt often suffices. Fine-tuning wins when you need consistent style at scale, deep fluency in a language or domain the base handles poorly, or a smaller model that matches a bigger one on your task and runs cheaper on dedicated GPU instances.
| Approach | When to choose it |
|---|---|
| Prompting | Behavior fixes with instructions and examples in the prompt |
| RAG | The model needs your documents or frequently changing knowledge |
| Fine-tuning | Consistent style, domain language or task performance the base lacks |
| RAG plus fine-tuning | Domain fluency and current knowledge at the same time |
| Smaller fine-tuned model | Cut serving cost while keeping quality on one narrow task |
from openai import OpenAI client = OpenAI( base_url="https://hostyourai.com/api/v1", api_key="hyai-...") client.chat.completions.create( model="llama-3.3-70b", messages=[{"role":"user","content":"Hallo!"}])
How to start
A fine-tuned model is served exactly like any other: vLLM on your instance, OpenAI-compatible endpoint, your key. While the project is scoped you can already build against a base model such as Qwen3 8B and swap the slug later:
from openai import OpenAI
client = OpenAI(
base_url="https://hostyourai.com/api/v1",
api_key="hyai-...",
)
response = client.chat.completions.create(
model="qwen3-8b",
messages=[
{"role": "system", "content": "Answer in formal Dutch, two sentences max."},
{"role": "user", "content": "Explain what a knowledge base link does."},
],
)
print(response.choices[0].message.content)Questions about LLM fine-tuning
Is fine-tuning self-service?
No. It is scoped per project: dataset review, base model selection, training and evaluation are planned together, and the output deploys to your own instance.
Does my training data leave the EU?
No. Preparation, training and deployment all run in European datacenters, covered by a DPA, and HostYourAI never trains on customer data for its own purposes.
Which base models are suitable?
Open models with permissive licenses, typically Llama, Qwen, Mistral or Gemma variants. The right base depends on your language, task and the GPU budget for serving.
Who owns the fine-tuned weights?
You do, within the license terms of the base model. You can export them and run them on any vLLM-compatible stack, as described in Deploy your own LLM with vLLM.
Should I fine-tune or use RAG?
Start with RAG if the gap is knowledge, since documents change faster than weights. Fine-tune when the gap is style, format or language quality. Many teams end up combining both.
What does it cost?
Training is quoted per project; serving afterwards is pay-as-you-go from your prepaid credit balance. See the pricing page for how billing works.
Everything you need for AI
From model hosting to a customer-facing API, it is built for developers and businesses who want their AI running on infrastructure they actually control, inside the EU.
Your data and your models stay on European GPUs. GDPR-friendly by design.
Llama, Qwen, DeepSeek, Mistral, FLUX and plenty more. Pick one and it is warm in minutes, with no DevOps on your end.
Point your existing client at the Router and keep your tools. No rewrite, no lock-in.
From zero to a warm endpoint in minutes
No infra to manage. Pick a model, get an OpenAI-compatible URL, ship.
Pick a model
Choose from the Model Garden or paste any HuggingFace ID. Set the VRAM and pick an EU GPU.
Get your endpoint
We deploy vLLM, run readiness probes, and hand you a warm OpenAI- and Anthropic-compatible URL plus an API key.
Route and ship
Point your client at the Router. It auto-routes to a warm instance, idles GPUs when nobody is online, and logs every request.
Private by Default
HostYourAI keeps your models, prompts and data on European GPUs. It is built for teams that care about compliance, reliability and real control.
GPUs and data residency inside Europe. Your prompts never leave the EU.
Run open-weight models with no black boxes or hidden telemetry.
GPUs idle when nobody is online, so you only pay for what you run.
Your infra, your keys, your models. Leave whenever you want.
Works with the tools you already use
The Router speaks the OpenAI and Anthropic APIs, so it drops straight into the clients and SDKs your team already runs. Just change the base URL.
Try HostYourAI for free
Frequently asked questions
Can I run this in the EU?
Yes. HostYourAI runs open models on GPUs in European datacenters via vLLM. Your prompts and outputs never leave the EU and there is no US cloud provider in the chain.
Is it GDPR-compliant?
Yes. All processing happens inside the EU, a Data Processing Agreement (DPA) is available and the subprocessor list is public. Open weights also mean no training on your data.
Is the API OpenAI-compatible?
Yes. Point your existing OpenAI or Anthropic client at our Router (https://hostyourai.com/api/v1), change only the base URL and API key. No rewrite, no lock-in.
What does it cost?
Pay-as-you-go on one prepaid credit balance: the shared router per token or a dedicated GPU per hour. Free to start, no minimum, no fixed monthly fee.
Works with 390+ open models
Text and image models on dedicated EU GPUs. Every model tested on our own hardware.
Related pages
Explore more about EU-hosted AI on HostYourAI.
Browse a curated catalog of serveable open chat, image and embedding models with live warm/EU state. Deploy in one click on European GPUs.
Read more →Deploy any open model on a dedicated European GPU with vLLM. Your own endpoint, encrypted upstream keys, idle when nobody is online.
Read more →One OpenAI- and Anthropic-compatible endpoint that routes your requests to open models on European GPUs. Drop-in, EU-hosted, no lock-in.
Read more →A drop-in OpenAI-compatible API on European GPUs. Change only the base URL and run open models, GDPR-friendly, without rewriting your code.
Read more →Chat with dedicated instances and Router models in the Playground. Test prompts, compare models and then see per-request usage and latency.
Read more →Host AI chatbots on European infrastructure. Customer service, internal assistants, and more with GDPR compliance.
Read more →