Skip to main content

​
Deployment costs

Q: Are there costs associated with deploying fine-tuned models? Fine-tuned (LoRA) models require a dedicated deployment. Here’s what you need to know: What you pay for:
  • Deployment costs on a per-GPU-second basis for hosting the model
  • Usage costs on a per-token basis when the model is used for inference
  • The fine-tuning process itself, if applicable
Deployment options:
  • Live-merge deployment: Deploy your LoRA model with weights merged into the base model for optimal performance
  • Multi-LoRA deployment: Deploy up to 100 LoRA models as addons on a single base model deployment
For more details, see the Deploying Fine Tuned Models guide.

​
Model availability

Q: Do you provide notice before removing model availability? Yes, we provide advance notice before removing models from the serverless infrastructure:
  • Minimum 2 weeks’ notice before model removal
  • Longer notice periods may be provided for popular models, depending on usage
  • Higher-usage models may have extended deprecation timelines
Best Practices:
  1. Monitor announcements regularly.
  2. Prepare a migration plan in advance.
  3. Test alternative models to ensure continuity.
  4. Keep your contact information updated for timely notifications.

​
Additional resources