finetuna

Terms of service

Last updated 19 September 2026. Plain language on purpose; every section binds both sides.

1. Who

finetuna is operated by Lintigo Damian Siembor, ul. Korczunkowa 4B, 05-503 Głosków, Poland, VAT ID PL1231217409 ("we"). You are the person or company that opens an account and orders a job ("you"). By starting a job you accept these terms.

2. What you buy

One fine-tuning run of the base model you picked, on the dataset you uploaded, with the parameters you chose (epochs, system prompt, GPU tier). The result is a set of files listed on the job page: GGUF files, a Modelfile, the LoRA adapter and a report. The price is the package shown before payment, in USD, VAT included, and does not change after payment.

3. Delivery

Files appear in your panel and you get an e-mail. Most jobs are delivered the same day. The outside limit is 48 hours from payment for packages S and M and 5 days for package L. If nothing is delivered by then you may ask for a full refund and we pay it within 7 days.

4. Failures and refunds

• Dataset error. We report the line number. You fix the file and we run the corrected job once at no extra charge. If you do not resend a corrected file within 30 days, the order is closed without refund.
• Failure on our side. We retry. If the second run also fails, we refund the full price within 7 days.
• Quality. Fine-tuning changes a model's behaviour; it does not guarantee a given accuracy. The report shows base vs tuned answers so you can judge. We do not refund for results you consider not good enough when the run itself succeeded.
• Refunds go back to the card used at Stripe.

5. Your dataset and your responsibilities

You confirm, for every job, that:

• you own the dataset or hold the rights needed to use it for model training and to give us the instruction to process it;
• it contains no content that is illegal in Poland, the EU or your country, and no material that would make the tuned model unlawful to distribute;
• on the Community tier it contains no personal data (GDPR Art. 4(1)), no health, financial or other special-category data, and nothing you hold under a confidentiality obligation. Community GPUs are machines of individual hosts on RunPod's marketplace and we cannot give the guarantees such data requires;
• on the Secure tier personal data is allowed only where you are the controller or an authorised processor and the Data Processing Agreement in section 6 applies.

We may refuse or stop a job that appears to breach this section and will tell you why. We do not review datasets by hand; automated checks only look at the file format.

6. Data processing (GDPR)

For account data (e-mail, job metadata, invoice data) we are the controller; see the privacy notice. For any personal data inside a dataset you upload on the Secure tier, you are the controller and we are your processor under GDPR Art. 28. This section is the processing agreement:

• We process the data only to run the job you ordered and to deliver the files, for the duration in section 7, and on your documented instruction, which is the job configuration.
• Sub-processors: Cloudflare (storage, EU jurisdiction setting), RunPod Inc. (GPU compute, Secure Cloud datacenters in the EU or US), Stripe (payments), Resend (e-mail). We tell you 14 days before adding one; you may end the contract if you object.
• Datasets travel encrypted in transit (TLS) and are stored encrypted at rest. Access is limited to the operator and to the automated training job.
• We help with data-subject requests and breach notices within 72 hours of learning about an incident.
• You may audit our compliance by written questions once a year; we answer within 30 days.
• Transfers outside the EU (RunPod US datacenters) rely on the EU Standard Contractual Clauses in RunPod's terms. On a Secure job, tick "Keep the GPU in the EU" if the GPU must stay in the Union.

7. Retention

The dataset, the checkpoints and the output files are deleted 14 days after delivery. What stays: your e-mail, the job configuration (model, package, epochs, tier, sizes, timestamps), the report's metrics and the sample answers shown in the report. Ask at hello@finetuna.io to delete the account and everything with it at once.

8. Model licences

The tuned model inherits the base model's licence, listed next to each model in the catalogue. Llama derivatives must carry "Built with Llama" and follow Meta's Acceptable Use Policy; Gemma derivatives follow Google's Gemma Terms. You are responsible for how you distribute and use the tuned model and for any output it produces.

9. What we do not promise

The service is provided as described in these terms and nothing more. We are not liable for indirect losses, lost profit or the behaviour of a tuned model in your product. Our total liability for a job is capped at the price you paid for it. Nothing here limits liability that cannot be limited under Polish law, including towards consumers.

10. Consumers

If you buy as a consumer in the EU: the job starts on your explicit request right after payment, so the 14-day withdrawal right ends once training has begun (Art. 38(13) of the Polish Consumer Rights Act). Everything in section 4 applies in your favour.

11. Changes and law

We may change these terms; the version in force when you paid applies to that job. Polish law governs, and disputes go to the court competent for our seat, unless consumer law gives you another forum.

Contact: hello@finetuna.io