Fine-Tuned Model Comparison
The same question, asked to a plain base model and to a version fine-tuned on Xevara's own voice — see the difference for yourself.
This model was fine-tuned using LoRA (Low-Rank Adaptation) on a custom dataset of Xevara-style conversations — real examples of how Xevara talks about its services, tone, and personality. Rather than relying on a long instruction prompt every time, the model's own behavior has been shifted to consistently reflect that voice. It's a small, efficient form of training: only a lightweight set of additional parameters is trained on top of an open-source base model, using free GPU resources — demonstrating actual model adaptation, not just prompting.