การใช้โมดูล Weaviate

Weaviate's power comes from its flexible architecture, which can be extended using modules. These modules add specialized functionalities directly to your Weaviate instance.

Think of them as plugins that enhance Weaviate's core capabilities. They can handle tasks like generating embeddings, performing Q&A, or even processing images.

Modules streamline your data pipeline by integrating advanced AI functionalities directly into your vector database. This means:

• Automatic Vectorization: Weaviate can create embeddings for you.
• Enhanced Search: Add capabilities like Q&A or summarization to queries.
• Simplified Development: Less external code needed for common AI tasks.
• Multi-modal Support: Handle various data types like text and images.

Weaviate offers a rich ecosystem of modules, typically categorized by their function:

• Text2Vec: Generate vector embeddings from text (e.g., text2vec-openai, text2vec-huggingface).
• Generative: Add Large Language Model (LLM) capabilities for Q&A, summarization (e.g., generative-openai, generative-cohere).
• Multi-modal: Process different data types like images (e.g., img2vec-clip).
• Rerank: Improve search relevance by reordering results.

Before you can use a module, it must be enabled in your Weaviate instance. This is typically done during setup (e.g., via Docker Compose) or when using Weaviate Cloud.

When you initialize your client, you often specify the modules you intend to use. For example, to use text2vec-openai and generative-openai, you'd configure your client accordingly.

The text2vec modules are fundamental for automatically creating vector embeddings. When you define a class schema, you specify which vectorizer to use.

Weaviate then takes care of calling the embedding model for you whenever new data is imported, turning your text into searchable vectors.

Here's how to define a schema that uses the text2vec-openai module to automatically vectorize the description property of a 'Article' class:

Generative modules (like generative-openai) allow you to integrate Large Language Models (LLMs) directly into your Weaviate queries. This enables powerful features such as:

• Q&A: Ask questions about your retrieved data.
• Summarization: Get summaries of search results.
• Extraction: Pull specific information from context.

You use these modules via the _additional { generate { ... } } GraphQL syntax in your queries.

Here's a conceptual example of how to use a generative module to get an answer to a question based on retrieved data. Assume a 'Question' class exists with relevant data.

Beyond text, Weaviate supports multi-modal modules like img2vec-clip, which can generate embeddings for images. This allows you to perform similarity searches on visual data.

Other advanced modules include those for reranking search results (e.g., rerank-transformers) to boost relevance, ensuring users see the most pertinent information first.

When using Weaviate modules, consider these best practices:

• Choose Wisely: Select modules that align with your specific application needs (e.g., OpenAI for general text, Hugging Face for specialized models).
• Monitor Costs: Many modules rely on external APIs (like OpenAI), which incur costs. Monitor usage and set limits.
• Version Control: Keep track of module versions as they can impact embedding quality or generative output.
• Security: Protect your API keys and ensure proper access control to your Weaviate instance.

Which of the following are primary benefits of using Weaviate modules?

In this lesson, we explored Weaviate's powerful module ecosystem. We learned that modules extend Weaviate's capabilities for tasks like automatic vectorization (text2vec), generative AI (generative), and multi-modal data handling (img2vec).

By integrating these modules, you can build more sophisticated and efficient AI applications directly on top of your Weaviate instance. Understanding how to enable and configure them is key to unlocking advanced functionalities.