Introduction

Azure OpenAI offers a wide range of models for various applications, from natural language processing and code generation to text search and embedding. In this blog, we'll explore the different Azure OpenAI models, their capabilities, and their use cases to help you decide which model best suits your needs.

GPT-3 Models

Azure OpenAI has several GPT-3 models, each with its own level of capability. The most capable GPT-3 model is Davinci, while the least capable model is Ada. The GPT-3 models can be used with Completion API requests, and get-35-turbo (ChatGPT) is available for both Completion API and Chat Completion API.

GPT-4 Models

GPT-4 models are more advanced than GPT-3 models, and they can only be used with the Chat Completion API. The two main GPT-4 models are gpt-4 and gpt-4-32k, with the latter having a larger token limit for processing longer texts.

Codex Models

Codex models are descendants of GPT-3 models and are designed for understanding and generating code. They are most proficient in Python and support over a dozen programming languages. The two main Codex models are code-davinci-002 and code-cushman-001, with Davinci being the most capable but more resource-intensive.

Embeddings Models

Azure OpenAI offers three families of Embeddings models for different functionalities: Similarity, Text search, and Code search. Each family includes models across a range of capabilities, with Davinci being the most capable but slower and more expensive, and Ada being the least capable but faster and cheaper.

1. Similarity embedding models are designed to capture semantic similarity between two or more pieces of text, making them suitable for clustering, regression, anomaly detection, and visualization tasks.

2. Text search embedding models help measure the relevance of long documents to a short search query, making them ideal for search, context relevance, and information retrieval tasks.

3. Code search embedding models are designed for code search and relevance tasks, with two input types: code for embedding code snippets and text for embedding natural language search queries.

Model Summary and Region Availability

Azure OpenAI provides a detailed summary table of the available models, their base model regions, fine-tuning regions, maximum request tokens, and the training data they were trained on up to a certain date.

Next Steps

To learn more about Azure OpenAI and how to fine-tune its models, explore the resources provided by Azure and OpenAI. Understanding the various models and their capabilities will help you make informed decisions when selecting the right model for your specific use case.

Conclusion

Azure OpenAI offers a diverse range of models to cater to different needs, from text generation and code generation to embeddings and search. By understanding the capabilities and limitations of each model, you can choose the one that best suits your project requirements and make the most out of the powerful tools provided by Azure OpenAI.