Arcee AI today introduced SuperNova, a 700-billion-parameter language model designed for enterprise deployment with advanced instruction-following capabilities and full customization options. The model aims to provide a robust, ownable alternative for enterprise data privacy, model stability, and customization, addressing key issues with API-based services such as OpenAI and Anthropic.
In the AI landscape dominated by cloud APIs, Arcee AI has taken a different approach with SuperNova. This large language model (LLM) can be deployed and customized within a company's own infrastructure. SuperNova was launched today, based on Meta's Llama-3.1-70B-Instruct architecture, and incorporates Arcee's superior instruction-following capabilities and a novel post-training process tailored to specific business needs.
Technological Innovation
SuperNova's development involved a multifaceted post-training approach.
Project lead engineer Lucas Atkins revealed the secret: "We trained three models: one distilled from Llama405B, one fed with our EvolKit-generated dataset, and one deeply modified with DPO on Llama3instruct. Finally, we combined them with a new magic, preserving the superpowers of each model."
Arcee claims this has resulted in SuperNova's instruction-following abilities, particularly from the distillation of the 405B parameter model, which not only shows SuperNova's ability to capture the essence of larger models while maintaining moderate hardware deployment.
Enterprise Deployment and Customization
SuperNova is designed to be deployed in the enterprise's own cloud environment, initially available on AWS Marketplace. Arcee is also working to make it available on Google and Azure marketplaces.
Arcee AI's co-founder Mark McQuade emphasized the benefits of this deployment method: "The model is deployed in your AWS VPC, but it also launches a web server, a chat interface, and a database to store your chat history. Everyone in your organization can interact closely with it."
This deployment method addresses concerns about data privacy and model stability. Unlike API-based services that may change without notice, SuperNova gives enterprises complete control. McQuade noted, given the recent turmoil in the AI industry, this is particularly important: "OpenAI just abandoned 3.5... many companies built their businesses around the 3.5 API. So when that API changes, your application crashes. But in our world, nothing changes unless you want it to, because it's your model, your way of running it."
Customization and Continuous Improvement
A major selling point of SuperNova is its ability to be fine-tuned and retrained in the enterprise environment.
Atkins explained the process and its benefits: "Over time, we can retrain the model entirely in your own environment to better align with your preferences. As we save these chats, if you want the model to improve comprehensively according to your unique preferences as a business, we have the ability to do so without letting data leave your system."
This capability allows technical teams to adapt the model to specific domain knowledge or company-specific requirements, a significant advantage over cloud-based API services that typically do not allow such customization.
Open Source Components
While the full 70B model is not open source, Arcee is releasing several components for the developer community:
A free API for testing and evaluation: This allows developers to try SuperNova without committing to full deployment.
SuperNova-Lite: An 8B parameter open-source version model. This smaller model may be useful for developers working in resource-constrained environments or those wanting to understand the architecture before deploying the full model.
EvolKit: Their dataset generation pipeline for creating complex QA pairs. This tool may be valuable for organizations looking to create custom training data for their specific use cases.
By open-sourcing these components, Arcee contributes to a broader AI community while also providing potential customers with tools to evaluate and customize their products. Arcee SuperNova is also available on AWS Marketplace.
Performance Claims and Benchmarking
Arcee claims SuperNova performs well across various domains, especially in mathematical reasoning. "Atkins noted: 'This performs exceptionally well on mathematical benchmarks.'" However, the company encourages third-party evaluations to validate their claims.
"We will provide an API for people to use. If third parties want to run trusted benchmarks to evaluate it themselves, we can arrange to provide them with access to the weights. We want full transparency for this model," Atkins said.
This openness to third-party evaluation is commendable as it allows independent verification of Arcee's claims. It will be particularly interesting to see how SuperNova performs against leading AI company models like OpenAI, Anthropic, etc., on standard benchmarks.
Impact on Enterprise AI Strategies
The launch of SuperNova comes at a time when many enterprises are reevaluating their AI strategies. While cloud-based API services have dominated the field, there is growing interest in deployable, customizable models that offer more control and flexibility.
SuperNova's approach addresses several key issues:
- Data Privacy: By deploying within the company's own infrastructure, SuperNova ensures sensitive data never leaves the organization's control.
- Model Stability: Unlike API services that may change or be deprecated without notice, SuperNova provides a stable foundation that only changes when the organization chooses to update.
- Customization: The ability to fine-tune and retrain the model on company-specific data allows for deep customization not possible with most API services.
- Cost Control: While the initial deployment may require significant resources, the long-term costs of running SuperNova may be lower than paying for large-scale API calls.
- Competitive Advantage: A customized, continuously improving AI model can provide a significant competitive advantage in industries where AI-driven insights are crucial.
The AI Sovereignty Dilemma
As enterprises navigate the rapidly evolving AI landscape, the launch of SuperNova reveals growing tensions in the industry: the convenience and capabilities of cloud-based AI services versus the control and customization offered by deployable models. This dichotomy presents what we might call the "AI Sovereignty Dilemma."
On one hand, cloud-based API services like GPT-4 and Claude offer state-of-the-art performance and continuous updates, but at the cost of data privacy issues and limited customization. On the other hand, models like SuperNova promise complete control and customization but require internal expertise for deployment and maintenance.
Arcee's approach with SuperNova attempts to bridge this gap, offering a model that can be deployed locally while still providing capabilities intended to rival leading cloud-based services. This hybrid approach may be particularly appealing to industries with strict regulatory requirements or those dealing with highly sensitive data.
Official Blog: https://blog.arcee.ai/meet-arcee-supernova-our-flagship-70b-model-alternative-to-openai/