Exploring Different Large Language Models

Continuing with the scenario discussed previously, we created basic prompts that assign urgency, sentiment, and categories to customer messages that can be used in software.

We used advanced prompting techniques to arrive at better prompts. We evaluated techniques and their combinations to analyze the results.

Let's now evaluate the results for different models.

Here are some reasons to use different models to solve a business problem:

1. Specialization: Different LLMs are often trained on varied datasets and optimized for specific tasks. For example, some models excel at text generation, while others specialize in image or speech recognition.
2. Performance: Combining multiple LLMs can enhance overall performance. One model handles natural language understanding, while another generates high-quality responses, creating a more effective team.
3. Cost Efficiency: Allocating specialized models to specific tasks can be more cost-effective than relying on a single, general model. This approach optimizes resource allocation and reduces waste.
4. Flexibility: Integrating different LLMs enables handling multimodal inputs and outputs, such as text, images, and audio, providing a more comprehensive solution.
5. Redundancy and Reliability: Having multiple models in place ensures that if one model fails or underperforms, others can step in, leading to more reliable outcomes and minimizing downtime.

Generative AI hub provides access to various models in one place in a reliable and responsible manner for accelerating your innovation journey.

Generative AI hub provides the following benefits:

• Flexible access to broad range of models: Accelerate AI development with flexible access to a broad range of models and compute capacity. Our extensive set of frontier AI models, infrastructure, and tooling helps you move faster.
• Custom AI solutions: Build custom AI solutions and extend SAP applications by combining AI models with your unique data, creating powerful results.
• Responsible AI development: Take control of your AI lifecycle and safeguard your data with SAP's trusted privacy and security policies, ensuring responsible AI development through centralized orchestration.

Flexible access to the broadest range of models and compute capacity implies that generative AI hub can help in the following ways:

• Tailor: Leverage various model options from all leading providers in one platform to tailor AI to your needs.
  • Connect: Easily connect to any supported foundation model or bring your own model, giving you flexibility and control.
  • Switch: Switch seamlessly between models to find and upgrade to the best-suited technology for your needs, ensuring optimal results.
  • Optimize: Eliminate individual contracts and lock-in, ultimately boosting the ROI of your AI projects.
• Deploy: Deploy your custom AI solutions to any cloud infrastructure, freeing you from lock-in and giving you greater flexibility.
• Deliver: Simplify access, deployment, and management of the latest AI advancements to accelerate development.

We have seen how different models can be configured in generative AI hub. We also saw how orchestration services and workflow can help in configuring different models using minimum code.

Let's summarize these benefits here:

1. You can configure models using the Configuration tab in SAP AI Launchpad. You can refer to the configuration steps in Unit 1. Some configured models are shown in the following screenshot.

   You can see various foundation models from many vendors. You can also see various scenarios, such as aicore-opensource, gcp-vertexai, and azure-openai. You will learn more about scenarios and models in the next topic.

2. You can deploy models using the Deployments tab in SAP AI Launchpad. You can refer to the configuration steps in Unit 1 Some deployed models are shown in the following screenshot.
3. You have also seen how to use orchestration to use a common code for different models. Refer to "send_request" function details in the helper functions in Unit 2 Lesson 3.

SAP AI Core manages the global AI scenario foundation-models, which grant access to generative AI models. We offer individual models as executable serving templates, allowing users to select the desired model by choosing the corresponding template.

The available executables in foundation-models global scenario include:

• Azure OpenAI Service: Use the REST API to access OpenAI's powerful LLMs. This includes models such as gpt-4o, gpt-35, gpt-35-turbo, and others.
• SAP AI Core: Work with open-source models hosted and accessed directly through SAP AI Core. This includes models such as mistralai--mixtral-8x7b-instruct-v01 and meta--llama3-70b-instruct.
• GCP Vertex AI: Get access to Google's advanced PaLM 2 and Gemini models. This also includes models such gemini-1.5-pro and textembedding-gecko.
• AWS Bedrock: Explore and use Foundation models from leading providers like Anthropic, Amazon, and more. This includes models such as anthropic--claude-3.5-sonnet and amazon--titan-embed-text.

Orchestration services are provided under the global AI scenario orchestration, which is managed by SAP AI Core. It allows the use of different generative AI models with a common code, configuration, and deployment.

You can learn more about the models and versions available within generative AI hub along other details, such as their data center, and when they will be outdated, through the SAP Note: 3437766 - Availability of Generative AI Models. It also explains the generative AI Token conversion rates for each model and gives information on rate limits.

See the following video to know how to use different models in SAP AI Launchpad for the Facility Solutions Company.

You can see results using different models. In the next lesson, we'll evaluate the results from different models.