Ministral 3B now available on Azure

Microsoft is committed to driving AI innovation by continuously improving our products. As we celebrate the one-year anniversary of Mistral 7B, we’re excited to announce that we’re continuing our collaboration with Mistral with a new, cutting-edge model added to the Azure AI model catalog. Ministerial 3B. Despite its small size, this model sets new standards in performance and efficiency.

A new horizon in AI performance

According to Mistral, Ministral 3B represents a significant advancement from the lower 10B categories, focusing on knowledge, common sense reasoning, function calls, and efficiency. maximum support 128k context length These models have been adapted for a variety of applications, from coordinating agent workflows to developing specialized task workers.

Improve workflow efficiency

When used with large-scale language models such as Mistral Large, Ministral 3B can act as an efficient intermediary for function calls in multi-step agent workflows. The ability to fine-tune allows you to excel at tasks such as:

• Input parsing: Simplify processing by quickly understanding user input.

• Job routing: Directs actions to the appropriate model or function based on user intent.

• API call: Efficiently interface with APIs while minimizing latency and operational costs.

Creating powerful agents from small models

Two main use cases

Ministral 3B excels in two main macro use cases:

• Multi-step agent workflow: This use case involves orchestrating complex workflows that require agents to selectively call larger models. Ministral 3B acts as a highly efficient intermediary, identifying the appropriate large-scale model to call and ensuring that the right model is used for the right task in the workflow.

• Low-latency, high-volume use cases: For applications that require fast, high-throughput responses, such as real-time customer support, data processing, and high-volume API calls, Ministral 3B delivers exceptional performance with ultra-low latency, allowing businesses to process high volumes of requests with minimal latency. .

Various use cases

Ministral 3B can also be leveraged for a wide range of agent use cases, including:

• Customer support automation: Improve customer interactions with efficient, automated responses.

• Back office and process automation: Streamline operations and improve productivity.

• Code migration and CI/CD: Facilitates smoother transitions in the software development cycle.

• Improved RAG architecture and search: Optimize your search augmentation creation efforts.

• Arbitration and LLM Output Verification: Ensure the quality and relevance of AI output.

Agentic advantages of small models

• ability: Ministral 3B and other compact models in the same category are powerful, fast and cost-effective. they are excellent function callIdeal for agent workflows.

• security: These models can be deployed securely and efficiently in your environment, keeping your internal data private. These can be implemented in: VPCAccessible in the cloud, on-premises, or via API.

• Customization possibilities: Smaller models can be easily fine-tuned for specific tasks and may outperform larger models in certain areas. Large scale allows for efficient fine-tuning and retraining, facilitating adaptability to changing requirements.

Agent architecture using small models

Ministral 3B is designed to operate within an efficient agent architecture that leverages specialized compact models. Here’s how it works:

• Handling user requests: When a user makes a request (e.g. “Give me my customer number”), it is processed through a router, which can be an embedding model or a large-scale language model (LLM).

• Connect with an expert advisor: The router forwards the request to the following specialized agents:
  1. account management agent
  2. fraud detection agent
  3. Billing Details Agent
  4. Customer Support Representative

• Efficiency Advantages: This architecture is much faster and cheaper than using a single large model. Small-scale and edge models are fine-tuned for specific domain tasks and outperform large-scale generic models in many scenarios.
• Microservice approach: This architecture makes it easier to apply the model compared to a large LLM that handles everything. This microservices approach improves the performance of function calls and the overall user experience.

How do I use MInistral 3B on Azure?

Here’s how to effectively leverage the new les Ministraux models in the Azure AI Model Catalog.

Prerequisites:

• Create Azure AI Studio Hub and project. You must select East US, West US3, South Central US, West US, North Central US, East US2, or Sweden Central as the Azure region for your hub.

Create a deployment to obtain the inference API and keys.

• Open the model card in the Model Catalog in Azure AI Studio.

• Click Deploy and select the Pay as you go option.

• Subscribe to and distribute Marketplace offers. You can also review API pricing at this stage.

• Within a minute you should be taken to the deployment page showing your API and key. You can try out the prompts on the playground.

Prerequisites and deployment steps are described in the following topics: product documentation. The API and keys can be used by a variety of clients. check it out sample To get started.

conclusion

The introduction of Ministral 3B marks an exciting milestone in our journey to enhance AI capabilities on Azure. Integrating these cutting-edge models into the Azure AI Model Catalog allows developers and enterprises to confidently innovate with advanced AI solutions for edge computing and on-device applications.

It combines low-latency performance, versatility across use cases, and cost-effectiveness. $0.04 per million tokens, Ministerial 3B A groundbreaking solution for businesses looking to harness the power of AI without breaking the bank.

Join us as we explore the future of AI with Ministral 3B, where cutting-edge technologies meet practical applications for a smarter, more efficient world.