LoRA and full fine-tuning
Coming soon
Arnict’s planned training offering includes LoRA and full fine-tuning (FFT). Training jobs are not available yet.
Fine-tuning adapts a base model to examples from your task. LoRA trains an adapter, while full fine-tuning updates the model’s weights. Arnict plans to support both approaches, followed by inference with the resulting model.
- Choose a base model appropriate for your task and license.
- Prepare representative examples and keep an evaluation set separate from training data.
- Compare the tuned model with the base model on your own success criteria.
- Join the training waitlistGet notified when training access opens.
- Custom-model inferenceLearn how custom models fit the dedicated-inference plan.
Training tools and data handling
The planned training stack includes Unsloth, Axolotl and Hugging Face TRL. Dataset formats, job controls, supported configurations and pricing will be documented before training opens.
Training datasets are different from inference prompts. Training requires access to the examples you submit. Arnict will publish the training-specific storage, access and deletion policy before accepting datasets.
Dataset preparation
Use examples you have permission to process. Remove credentials and unnecessary personal information, check labels for consistency and define an evaluation task before uploading data.