Fine-tune a separate customer model inside the FusionPointAI training SDK, using an agreed objective, prepared data and measurable evaluation.
The training SDK loads and fine-tunes the student model in the selected runtime. The website/API coordinates the session, configuration, prepared data, progress and artifact delivery. AI Hosting is optional and billed separately.
The website/API authenticates the `ai:training` key, opens a session for a paid job, prepares and transfers training and held-out validation data, and supplies the objective and configuration. The training SDK loads and fine-tunes the student model in FusionPointAI AI Hosting, your data center, or Google Colab or a similar runtime. It reports progress, evaluation and a locally produced artifact back to the session. Dataset-generation examples illustrate sample content and format only.
The Reference Model is an eligible teacher or performance reference. The Customer Model is a separate student whose weights are trained. A prompt describes the objective; it does not itself modify model weights or copy a proprietary model.
Describe the target language, specialisation or capability, the performance criteria and your delivery preference. The website prepares the training configuration and session; the selected runtime executes the training SDK.
Choose among the student models, training methods, dataset sizes, compute requirements and delivery formats eligible for the selected reference configuration. Options outside the approved scope require a separate commercial quote rather than an arbitrary calculator total.
Define the task and acceptance criteria before training. Use customer-authorised datasets or licensed educational resources approved for the project. Dataset preparation, provenance, permitted uses and evaluation splits must be established before execution.
The student model is loaded and fine-tuned inside the FusionPointAI training SDK, running on FusionPointAI AI Hosting, in your data center, or in Google Colab or a similar environment. The website VPS does not perform GPU training.
Measure the resulting model on a documented held-out evaluation set. Reports distinguish the requested target from the result actually achieved. Reference-level performance is an evaluation objective, not an unconditional promise of equivalence on every task.
The approved quote identifies the deliverable, model licence, artifact format and any adapter or base-model dependencies. Downloadable delivery does not require an AI Hosting subscription. Dedicated deployment is an optional delivery arrangement when supported.
An eligible purchased training plan includes reference-model API access for its approved training period. Access begins after confirmed payment and ends at the recorded expiry or earlier revocation under the agreement. It does not renew indefinitely; continued access requires a separate eligible subscription.
Project access follows the existing account and job permissions. Training inputs and delivered artifacts use the established private-file and retention controls. Do not upload data unless you have the required rights and the agreed handling terms cover it.
A generated training example could pair a German-language request with a target response, with separate held-out examples prepared for validation. This illustrates dataset preparation only.
Training develops capabilities in a separate customer model. Engineering adapts an existing artifact for size, precision, runtime or deployment. The workflows can reuse the same preparation, evaluation and delivery services without becoming duplicate backends.
It can initiate configuration. Actual training starts only after the required permissions, approved data, compatible student model, exact quote, payment and execution prerequisites are satisfied.
No automatic transfer is implied. The customer model is trained separately, and its deliverables and licence are defined in the approved quote.
Not unconditionally. Agreed capability targets are assessed using documented benchmarks and the measured results are reported.
No, when the approved deliverable is a downloadable artifact. Supported hosted deployments are priced separately from the one-time training project.
It expires at the end of its recorded training period. You can then choose a separate available Unlimited API subscription.
