Jun 03, 2025

GroqCloud now supports Low-Rank Adaptation (LoRA) fine-tunes, exclusively by request, for our Enterprise tier customers. LoRA enables businesses to deploy adaptations of base models customized to their specific use cases on GroqCloud, offering a more efficient and cost-effective approach to model customization.

As a part of this release, we are introducing the ability to serve multiple LoRAs at the same latency and speed as the base model on GroqCloud. This means customers can deploy LoRA fine-tuned models without requiring a full dedicated hardware instance.

Real-World Applications of LoRA Fine-Tuning

Phonely, a company focused on AI phone support agents, partnered with Maitai to enhance the speed and accuracy of their AI agents. By leveraging Maitai’s platform, which enables hotswapping of LoRA models that run on GroqCloud, they achieved significant improvements in real-time response performance – this was a major milestone for conversational AI. They say a 73.4% reduction in time to first token, a 74.6% reduction in completion time, and an improvement in accuracy from 81.5% to 99.2% across four model iterations, surpassing GPT-4o by 4.5%. This partnership sets a new benchmark for phone support powered by agentic AI, enabling Phonely to deliver instantaneous, natural responses that improve customer satisfaction. Ultimately, this means enterprises can scale to tens of thousands of calls per day with lower latency and higher accuracy compared to use cases on any closed-source model.

What is a LoRA?

LoRA is a revolutionary Parameter-Efficient Fine-Tuning (PEFT) technique that allows enterprises to fine-tune the behavior of base models by adding small, task-specific adapters. These adapters remain separate from the base model and are applied during inference, enabling quick and efficient model adaptation without the need for full retraining. This approach is particularly beneficial for businesses seeking to tailor AI solutions to their unique needs while maintaining the integrity and performance of the base model.

Why LoRA Fine-Tuning Matters for Enterprises

LoRA fine-tuning offers several advantages over traditional fine-tuning methods, making it an ideal choice for businesses looking to optimize their AI workflows:

• Lower Total Cost of Ownership (TCO): LoRA significantly reduces fine-tuning costs by avoiding the need for full base model retraining. This makes it cost-effective to customize models at scale.
• Rapid Deployment with High Performance: Smaller, task-specific LoRA adapters can match or exceed the predictive accuracy of fully fine-tuned models, while also enabling faster inference. This allows for quicker experimentation, iteration, and real-world impact.
• Non-Invasive Model Adaptation: Since LoRA adapters don’t require changes to the base model, you avoid the complexity and liability of managing and validating a fully retrained system. Adapters are modular, independently versioned, and easily replaceable as your data evolves—simplifying governance and compliance.
• Full Control, Less Risk: Customers maintain complete control over how and when updates happen. LoRA adapters are lightweight, swappable, and integrate seamlessly into existing systems with minimal disruption. Plus, with self-service APIs, updating adapters is quick, intuitive, and doesn’t require heavy engineering lift.

Getting Started with LoRA Fine-Tuning on GroqCloud

To begin leveraging LoRA fine-tuned models on GroqCloud, follow these steps:

1. Prepare Your LoRA Adapters:

2. Fine-tune Groq-supported base models externally using providers like Maitai or custom PEFT solutions.

3. Ensure you fine-tune the exact base model supported by Groq to guarantee compatibility with our platform.

4. Request Access to LoRAs on GroqCloud:

5. Reach out to your Groq Sales Team representative or fill out our Enterprise request form to get access to deploy LoRA fine-tuned models on GroqCloud.

6. Once approved our team will provide you with instructions on how to upload and manage your LoRA adapters.

7. Deploy & Integrate:

8. Once your LoRA adapters are hosted on GroqCloud, integrate them into your applications by calling your unique LoRA model ID and start experiencing the benefits of efficient, customized AI solutions.

Note at the time of launch, LoRA support is only available for the Llama 3.1 8B and Llama 3.3 70B. Our team is actively working to expand support for additional models in the coming weeks, ensuring a broader range of options for our customers.

Conclusion

LoRA fine-tune support on GroqCloud is a big step – it means we’re making efficient, scalable, and cost-effective AI solutions more accessible to enterprises, on their terms. By combining the power of LoRA adapters with Groq’s fast AI inference infrastructure, businesses get faster deployment, lower costs, and greater control over their AI initiatives.

Whether you’re fine-tuning for a specific task or adapting models to meet evolving business needs, deploying LoRA fine-tuned models on GroqCloud offers the flexibility and performance you need to stay ahead in today’s competitive landscape.

Request access to LoRA fine-tuned models on GroqCloud today by filling out our Enterprise request form or reaching out to your Groq Sales Team representative.