Qretta Book a call
Trigger: Model training and fine-tuning

Model training and fine-tuning

Models trained or fine-tuned on your data to do one job better than a general model can.

Pillar
AI
Typical timeline
3 to 10 weeks
Price
Quoted in a written scope after the first call
Book a call
01Who it is for

Who it is for

Teams that tried a general AI model and found it too slow, too expensive, or not accurate enough on their task.

02The problem

The problem

General models know a little about everything. On a narrow task with your own terms, formats, and edge cases, a smaller model trained on your data is often more accurate and far cheaper to run.

03What you get5 deliverables

What you get

  1. 01Dataset preparation: cleaning, labelling guidelines, and train and test splits
  2. 02A fine-tuned language model (LoRA or full fine-tune) or a custom model for images, text, or tables
  3. 03Evaluation against the base model on a held-out test set
  4. 04A model card: data, metrics, known failure cases, and intended use
  5. 05Weights and training code delivered to your storage
04How it works4 steps
  1. Step 1

    Define the task and metric

    One job and one number to improve.

  2. Step 2

    Build the dataset

    Usually the most work, and the part that decides the result.

  3. Step 3

    Train and compare

    Several runs, each compared with the baseline on the same test set.

  4. Step 4

    Package

    The best model, its card, and the code to retrain it.

05Timeline

Timeline

3 to 10 weeks

A typical range for this kind of work. The written scope we send after the first call sets your actual dates.

06Related work

Related work

07Questions

Questions buyers ask

How much data do we need?

It depends on the task. Fine-tuning a language model can work with a few hundred good examples, while an image model trained from scratch needs far more. We tell you after seeing your data.

Who pays for the GPUs?

Training runs on cloud GPUs, usually in your own account. We estimate the compute cost before any run starts.

Who owns the model?

You own the weights, the dataset, and the training code. The base model's licence still applies, and we review it with you first.

08More in AI

More in AI

09Output: book a call

Tell us what you are trying to build

A 30-minute call. You explain the problem, we ask questions, and you leave knowing whether we are the right team and what the work would involve.

Book a call Email us