Weaviate is completely modularized. The Core of Weaviate, without any modules attached, is a pure vector-native database and search engine. Data is stored as vectors, and these vectors are searchable by the provided vector index (ANN) algorithm. The functionality of the vector-native database can be enriched by modules. One or more modules can be attached, for example to vectorize data (dense retrievers) or other functionalities like question answering (reader/generator modules). Retriever modules are used to quickly find a relevant set of data to the query, which then can be passed to a Reader or Generator module to enhance the search results.
Vectorization modules (Dense Retriever modules)
Vectorization modules, like the
img2vec-neural, transform data into vectors. These modules are also called “Dense Retriever” modules. Retrievers function as a filter to quickly find a relevant set of data to the query.
Dense retrievers vs. sparse retrievers
Sparse vector retrievers for the task of finding relevant data (by calculating the similarity of two pieces of data) in a database are for example TF-IDF or BM25. These retrievers are not trainable; instead they rely on word frequency in documents. This type of retrievers are not possible as Weaviate module, because the implementation sits deep in the core of Weaviate (BM25 will be released soon!). Dense vector retrievers are a relatively new approach to data retrieval, which usually outperforms traditional sparse retrieval methods because they take semantics into account and they are trainable. Dense vector retrievers can be chosen as Weaviate Module. Depending on the type of data you want to store and search (text, images, etc), and depending on the use case domain (science, healthcare, common daily language, etc), you can choose and attach a dense retriever (vectorization) module that best fits your use case.
Modules with additional functionalities (Reader or Generator modules)
Reader or Generator modules can be used on top of a Retriever/Vectorization module. A Reader module takes the set of relevant documents that are retrieved by the Retriever module, and extracts a piece of relevant information per document. An example Reader module is
qna-transformers module, which extracts an answer directly from a document that is retrieved by a Retriever module. A Generator module would, on the other hand, use language generation to generate an answer from the given document. Currently, there are no Generator modules available out-of-the-box yet, but they may be added soon!.
Check here how you can create and use your own modules.
Modules can be dependent on other modules to be present. For example, to use the
qna-transformers module, exactly one text vectorization module is required.
This graphic shows the available modules of the latest Weaviate version (v1.13.2).
Modules can be “vectorizers” (defines how the numbers in the vectors are chosen from the data) or other modules providing additional functions like question answering, custom classification, etc. Modules have the following characteristics:
- Naming convention:
- Vectorizer (Retriever module):
<media>2vec-<name>-<optional>, for example
- Other modules:
<functionality>-<name>-<optional>, for example
- A module name must be url-safe, meaning it must not contain any characters which would require url-encoding.
- A module name is not case-sensitive.
text2vec-bertwould be the same module as
- Vectorizer (Retriever module):
- Module information is accessible through the
- General module information (which modules are attached, version, etc.) is accessible through Weaviate’s
- Modules can add
additionalproperties in the RESTful API and
_additionalproperties in the GraphQL API.
- A module can add filters in GraphQL queries.
- Which vectorizer and other modules are applied to which data classes is configured in the schema.
Weaviate without modules
Weaviate can also be used without any modules, as pure vector native database and search engine. If you choose not to include any modules, you will need to enter a vector for each data entry. You can then search through the objects by a vector as well.
If you can’t find the answer to your question here, please look at the: