DeepSeek API Unleashes New Features: JSON Output, Function Calling, and More

The DeepSeek API has recently undergone a significant upgrade, introducing a suite of powerful new features designed to enhance its capabilities and broaden its application across diverse use cases. This update focuses on providing developers with more control, flexibility, and efficiency when working with the DeepSeek models.

What's New in the DeepSeek API?

The latest update brings several key enhancements:

• JSON Output: Ensures structured and easily parsable responses.
• Function Calling: Enables interaction with external tools and the physical world.
• Chat Prefix Completion (Beta): Offers more granular control over output generation.
• FIM Completion (Beta): Allows filling in the middle of text for creative applications such as story or code completion.
• Extended Token Limit: 8K max_tokens (Beta) for handling longer text outputs.

These features are available for both deepseek-chat and deepseek-coder models, providing a versatile toolkit for developers.

Diving Deeper into the New Features

Let's explore each feature in more detail:

1. JSON Output: Structured Data Made Easy

The new JSON Output feature is a game-changer for applications requiring structured data processing. By enforcing a valid JSON format for the model's output, it streamlines subsequent parsing and automation workflows.

How to use JSON Output:

• Set response_format to {'type': 'json_object'}.
• Guide the model to output JSON in the prompt.
• Set max_tokens appropriately to prevent truncation.

This feature is particularly useful for tasks like data extraction, knowledge representation, and generating structured reports. For a more step-by-step guide, check out the JSON Output Guide.

2. Function Calling: Bridging the Gap to the Real World

Function Calling empowers the DeepSeek API to interact with external tools and services, enabling a wide range of applications beyond simple text generation.

Key Capabilities:

• Supports multiple functions in a single call (up to 128).
• Enables parallel function calls for increased efficiency.

Imagine integrating DeepSeek with a website crawler tool like in LobeChat, demonstrated in the API documentation. This allows the model to fetch information from websites and provide summaries, opening doors to automated research, information gathering, and more. Explore practical applications in the Function Calling Guide.

3. Chat Prefix Completion (Beta): Fine-Grained Output Control

Chat Prefix Completion offers developers greater control over the generated output by allowing them to specify a prefix for the last assistant message. This is particularly useful for:

• Enforcing specific output formats (e.g., code blocks).
• Continuing truncated messages that hit the max_tokens limit.

How to use Chat Prefix Completion:

• Set base_url to https://api.deepseek.com/beta to enable Beta features.
• Ensure the last message's role is "assistant" and set the "prefix" parameter to True.

For example, you can force the model to start with a code block by setting the prefix to ```python\n`, further refining the output using the 'stop' parameter. Learn more in the Chat Prefix Completion Guide.

4. FIM Completion (Beta): Unleashing New Completion Scenarios

FIM (Fill-In-the-Middle) Completion allows users to provide a prefix and suffix, prompting the model to complete the content in between.

Ideal Use Cases:

• Story writing: Perfect a story by filling the gap present in it.
• Code completion: Letting the AI fill in missing functionality in the existing function.

To use FIM Completion, remember to set base_url to https://api.deepseek.com/beta to enable the Beta features. See how to best utilize this feature in the FIM Completion Guide.

5. 8K max_tokens (Beta): Expanding Output Possibilities

The increase in the max_tokens parameter to 8K in the Beta API addresses the need for handling longer text outputs. Allowing models to produce more detailed, nuanced, and comprehensive responses.

How to Enable:

• Set base_url to https://api.deepseek.com/beta.
• Set max_tokens up to 8192. For a better understanding of how tokens work, refer to this Token & Token Usage guide.

Important Considerations for Beta Features

It's important to remember that features marked as "Beta" are considered unstable. This means their behavior and availability may change as DeepSeek continues to refine and improve them. To use any of the beta features, ensure that you set the base_url to https://api.deepseek.com/beta.

Stay Updated and Engaged

DeepSeek encourages users to explore these new features and provide feedback. The company plans to release related model versions to the open-source community once the functionality is stable. Stay tuned for further updates and announcements!

For more resources, you can check out DeepSeek's integrations and other Other Resources to help start your journey.