Skip to content

changelog.rst

Latest commit

 

History

History
1733 lines (1314 loc) · 69.5 KB

changelog.rst

File metadata and controls

1733 lines (1314 loc) · 69.5 KB

Changelog

Version 4.14.3

This patch version includes:
  • Fixes return type for aggregate with group_by and zero results
  • Fix multi-vector with hybrid search

Version 4.14.2

not released

Version 4.14.1

This patch version includes:
  • Fixes a missing typing.overload specification for the query.bm25 method

Version 4.14.0

This minor version includes a major update to the docstrings and documentation of the client.
  • All docstrings have been aligned with Google's style format
  • The read-the-docs page has been streamlined and improved for ease of use: https://weaviate-python-client.readthedocs.io/en/stable/
  • Fixes a bug when adding a property to a collection configured with both named vectors and a generative module
  • Potential fixes for race conditions when instantiating async clients
  • A number of minor fixes to the public .pyi stub files found by developing a stub auto-generating tool for the codebase

Version 4.13.2

This patch version includes:
  • Fixes a parsing bug when providing URLs with GenerativeConfig for runtime generate queries

Version 4.13.1

This patch version includes:
  • Fixes generative.near_text() to not require generative_provider

Version 4.13.0

This minor version provides a complete top-down refactor of the underlying transport for the sync and async clients.
  • Introduction of a dependency injection system whereby the sync/async connections are injected into each client independently thereby removing their cross-dependency.
  • Removal of the event_loop.py file and all logic surrounding its usage by the sync client.
  • Refactoring of the batching algorithm to use blocking threads with a sync connection, rather than the event loop sidecar thread.

Version 4.12.1

This patch version includes:
  • Renames GenerativeConfig.openai_azure to GenerativeConfig.azure_openai to align with the convention of Configure.Generative.azure_openai

Version 4.12.0

This minor version includes:
  • Support for new 1.30 features:
    • Dynamic user management of OIDC and native DB users
    • Improved multi-dimensional/colBERT vectors
    • Query-time configuration of generative modules
    • Tenant-based filtering of data permissions in RBAC
    • The new generative-xai module

Version 4.11.3

This patch version includes:
  • Fixes a rare bug when the batching algorithm waits for async indexing to complete far beyond the maximum number of allowed retries
  • Removes the sync-in-async warning emitted by the sync client when used in an async context due to the warning being erroneously emitted in notebooks

Version 4.11.2

This patch version includes:
  • Adds exponential backoff on the UNAVAILABLE gRPC error code for the BatchObjects method
    • This is used in the batching algorithm and in data.insert_many
    • Max number of retries is configurable using the WEAVIATE_BATCH_MAX_RETRIES env var
  • Adds support for OIDC when using the text2vec-weaviate module
  • Adds support for updating descriptions of properties in collections (needs compatible server version)
  • Adds warning recommending to refactor to use AsyncClient if the SyncClient is used in an async context
  • Fixes bugs when importing/exporting collection configurations with respect to property module configs
  • Fixes bug when exporting config with multivector enabled between versions

Version 4.11.1

This patch version incldues:
  • Fixes exporting of named vector collections
  • Support for upcoming weaviate agents. User pip install weaviate-client[agents] to try it out
  • Raise error if a header with value None` has been added
  • Introduce collections.use as an alias for collections.get
    • collections.use is the preferred method to create an in-memory collection object
    • collections.get will be deprecated in the future and should not be used in new code

Version 4.11.0

This minor version includes:
  • Support for new 1.29 features:
    • RBAC is ready for production
    • Multi-dimensional/COLBert vectors experimental, breaking changes are expected
    • Aggregations are using GRPC now
  • Improved exception handling for errors returned by Weaviate
  • Add support for NVIDIA modules:
    • multi2vec-nvidia
    • text2vec-nvidia
    • generative-nvidia

Version 4.10.4

This patch version incldues:
  • Fix missing dimensions parameter in text2vec-azure-openai

Version 4.10.3

This patch version includes:
  • Add experimental debug namespace
  • Add support for dynamic backup location
  • Fix potential deadlock in batching
  • Fix backup and restore for embedded mode
  • Add extra method for text2vec_google_aistudio

Version 4.10.2

This patch version includes:
  • Fixes issue with query parameters in cluster.nodes method due to dependence on bugged behaviour of httpx that was fixed in 0.28.0

Version 4.10.1

This patch version includes:
  • Fixes compatibility with httpx to 0.28.X which requires updating the minimum required version to 0.26.0
  • Remove outdated docs

Version 4.10.0

This minor version includes:
  • Support for new 1.28 features:
    • RBAC (experimental, breaking changes are expected)
  • Add VoyageAI's new multimodal embedding model
  • Remove copy of v3-client. If you are still using v4 code please either:
  • Upgrade protobuf to v5
  • Add support for python 3.13
  • Weaviate v1.23 and v1.24 are NOT supported anymore

Version 4.9.6

This release is the last release that supports weaviate v1.23 and v1.24

This patch version includes:
  • Add support for the new Kagame_JA tokenizer

Version 4.9.5

This patch version includes:
  • Add support for the new text2vec-weaviate module
  • Wrap tenant.get AioRpcError in a specific exception

Version 4.9.4

This patch version includes:
  • Add support for multi2vec-jinaai
  • Add support for multi2vec-cohere
  • Add support for time-based deletion strategy
  • Fix type restriction on include_vector in QueryReference

Version 4.9.3

This patch version includes:
  • Fixes interface for updating generative and reranker collection settings

Version 4.9.2

This patch version includes:
  • Suppress gRPC library warnings for each event loop used in each process

Version 4.9.1

This patch version includes:
  • Fix event loop being used from multiple processes
  • Add updating generative and reranker collection settings
  • Add experimental backoff for GRPC UNAVAILABLE error codes
  • Automatically pick up GRPC message size from Weaviate
  • Remove octoai vectorizer and generative methods. They are shutting down

Version 4.9.0

This minor version includes:
  • Support for new 1.27 features:
    • Multi vector search for near_vector and hybrid
    • Backup cancellation
    • Vector search filter strategy (acorn)
  • Renaming *-palm to *-google for generative and vectorization modules.
  • Typing fixes
  • Bump of default embedded version to 1.26.6
  • Allow return_properties={True, False} as equivalents to {None, []}
  • Dependency updates

Version 4.8.1

This patch version includes:
  • Suppresses warning emitted by protobuf library in relation to obsolate generated stubs
  • Fixes the authlib to <1.32.2 while a potential bug is investigated
  • Updates the validators dependency

Version 4.8.0

This minor version includes:
  • Support for new modules:
    • Mistral text2vec module
    • Friendli generative module
    • Databricks text2vec and generative modules
  • New fetch_objects_by_ids method
  • Support for hybrid search with vector distance
  • Fixes issue with adding a property on a vectorized collection
  • Bumps requests, importlib-metadata, pytest`, pytest-asyncio, types-requests, litestar and typing-extensions libs to newest versions

Version 4.7.1

This patch version includes:

  • Fixes log noise due to value of deprecated GRPC_VERBOSITY flag in underlying grpc library
  • Bumps requests and authlib versions to avoid security issues
  • Adds support for configuring the new reranker-jinaai module when creating collections
  • Fixes parsing of the timeout configuration on client instantiation
    • The query timeout now modifies the read= timeout in the httpx client for all GET, HEAD, and gql query requests
    • The insert timeout now modifies the read= timeout in the httpx client for all DELETE, POST, PATCH, and PUT requests
    • The init timeout now only modifies the timeouts in the httpx client for requests involved in the client.connect() method

Version 4.7.0

This minor version includes:

  • The introduction of the WeaviateAsyncClient class to support I/O requests to Weaviate using the async/await syntax with asyncio <https://docs.python.org/3/library/asyncio.html>

    • All methods that perform CRUD and search actions are now async def functions

    • To instantiate a client quickly, use the weaviate.use_async_with_x methods in an async context manager pattern, e.g.:
      async with weaviate.use_async_with_local() as client:
    # Your code
    • Note, you cannot do await weaviate.use_async_with_x if not using the context manager pattern. You have to create the client first and then connect manually:
      client = weaviate.use_async_with_local()
await client.connect()
# Your code
await client.close()

  • A refactoring of the underlying implementation of the WeaviateClient to use the WeaviateAsyncClient under-the-hood scheduling the necessary coroutines to run in a side-car event-loop thread

  • Support for new core Weaviate features in both the sync and async clients:
    • Multi-vector search in the .near_x and .hybrid methods within the .generate and .query collection namespaces
    • Scalar Quantization (SQ) vector index configuration
    • Async replication configuration for multi-node Weaviate deployments
    • Tenant offloading to S3 cloud storage using the newly intrduced OFFLOADED tenant activity status
    • Renaming of HOT to ACTIVE and COLD to INACTIVE for tenant activity statuses
    • NOTE: To use these features, you must have Weaviate version 1.26.0 or higher

Version 4.6.7

This patch version includes:

  • Fix batching with references. Under some circumstances a reference could be added before its to-object and the reference would be lost.
  • Fix node status for timed out nodes
  • Fix parsing the year 0. While weaviate allows to add dates with year zero ("0000-01-30T00:00:00Z"), the datetime library is based on the gregorian calendar which does not have a year zero. The client will years with 0 as the minimum date that is possible in datetime (datetime.datetime(1, 1, 1, 0, 0)) and emit a warning
  • Support for custom rerankers and generative modules using Configure.Generative.custom() and Configure.Reranker.custom()
  • Add support for kagome_kr tokenizer. Requires Weaviate 1.25.8
  • Increase default embedded version to 1.25.8

Version 4.6.6

This patch version includes:

  • Log batch errors
  • Only the last 100k successfully added UUIDs are kept in memory to prevent OOM situations.
  • Fix tenant creation with string input

In the v3 copy that is part of v4:

  • Fixes GraphQL query injection vulnerability caused by incorrect escaping of backslashes in plain text input builder methods. Many thanks to @adamleko, @bismuthsalamander, and @tardigrade-9 for their help in fixing this issue
  • Fixes batch retry with tenants

Version 4.6.5

This patch version includes:
  • Addition of new voyageai embedding and rerank models as string literal types
  • Added missing exports of submodules in weaviate.outputs to aid usage
  • Updated ollama modules docstrings
  • Added missing py.typed file for use by static type checkers

Version 4.6.4

This patch version includes:
  • Add support for auto-tenant-activation (available in Weaviate 1.25.2)

Version 4.6.3

This patch version includes:
  • Removal of top-level validation of vector in data.insert, data.replace, and data.update. This validation occurs within the _get_vector_v4 method now instead.

Version 4.6.2

This patch version includes:
  • Respect default vectorizer
  • Wait for Weaviate 1.25 to be ready before starting the embedded client
  • Add missing models for voyageai
  • Rename WCS to Weaviate Cloud and add new helper function weaviate.connect_to_weaviate_cloud

Version 4.6.1

This patch version includes:
  • Fixes for client.integrations.configure

Version 4.6.0

This minor version includes:

  • Support for Weaviate 1.25.0:
    • BM25/Hybrid support groupBy parameter
    • Hybrid supports MoveTo/MoveAwayFrom in near_vector and near_text through HybridVector class
    • Ollama text2vec and generative module
    • Octoai text2vec and generative module
    • multi2vev-palm module
    • dynamic vector index type
    • auto tenant creation
    • improved batching with vectorization
    • tenant exists endpoint
    • get tenant by name
  • Added client.integrations.configure to configure api-keys and model provider parameters for integration/module-providers without setting headers.
  • Improved error messages and deprecation warnings.

Version 4.5.7

This patch version includes:

  • Deprecation of the bit_compression field in the PQConfig class
  • Improvements to closing possibly open objects and connections
  • Enhances the WeaviateGRPCUnavailableError message with added context relevant to the user's environment
  • Relaxes the httpx requirements to aid compatability with other packages

Version 4.5.6

This patch version includes:

  • Support for configuring collections with the new reranker-voyageai module
  • Providing an alpha parameter to collection.iterator() to control the beginning of the iteration
  • Update the default Timeout.init value from 1s to 2s

Version 4.5.5

This patch version includes:

  • Bugfix when parsing the result from v1/nodes API with shards: null
  • Bugfix when parsing the result from v1/schema API with class.properties.moduleConfig: null and class.vectoriser: !'none'
  • Dependency bumps

Version 4.5.4

This patch version includes:

  • Fix parsing of creation/update time from old weaviate versions that write them in ns instead of ms
  • Support video_fields in multi2vec-palm which was added in Weaviate 1.24.4:

Version 4.5.3

This patch version includes:

  • Fix bug with hybrid searches without vector.
  • Support for new modules in Weaviate 1.24.2: - text2vec-voyageai - generative-mistral - Support new parameters for inference URLs in text2vec-transformers and multi2vec-clip
  • Support for new modules in Weaviate 1.24.3: - multi2vec-palm

Version 4.5.2

This patch version includes:

  • Fixes endpoint parameter for text2vec-palm
  • Adds support for GSE and TRIGRAM tokenizers

Version 4.5.1

This patch version includes:

  • Implements an extension to the filtering syntax allowing to pass lists of filters
    • Filter.all_of([f1, f2]]) is a shortcut for f1 & f2
    • Filter.any_of([f1, f2]]) is a shortcut for f1 | f2
    • Can all be chained and mixed together to create dynamic and complex filters
  • Introduces weaviate.classes.init.Timeout class allowing to define the timeout used when performing client init checks, in addition to connect and query
  • Fixes a bug when performing contains_any/contains_all filtering using an empty list
  • Adds the ability to limit the top_occurences return when performing aggregation queries
  • Allows for defining gRPC proxying of the client and fixes the parsing of http and https proxies
  • Allow None as a query value in BM25 and hybrid queries
  • Fix missing named vectors support in data.update and data.replace
  • Reimplement support for updating named vector configurations alongside the patched 1.24.1 server version

Version 4.5.0

This minor version includes:

  • Full support for the new named vectors feature available in the Weaviate 1.24 release.
  • Bugfixes to passing of Weaviate schema objects as collection configurations in certain edge cases.
  • Support use of Sagemaker when vectorizing with the text2vec-aws module.
  • Allow creation of collections that use the hnsw index with the bq quantizing strategy.
  • Allow specifying dimensions when vectorizing with the text2vec-openai module.
  • Python in-memory performance improvements when making queries .

Version 4.4.4

This patch version includes:

  • A fix to the validation logic of the apiEndpoint field of GenerativePaLMConfig object.

Version 4.4.3

This patch version includes

  • Fixes batching with references. Under some circumstances a reference could be added before its from-object and the reference would be lost.
  • Fixes readthedocs page
  • Small performance improvements for queries

Version 4.4.2

This patch version includes

  • Fixes client.is_ready().
  • Adds option to skip input parameter validation if you need to squeeze out some extra performance.
  • All functions that accept vectors now also accept numpy arrays, tensorflow arrays and pandas/polars dataframes as input.
  • Hybrid search accepts None as query for a pure vector search.
  • Adds FilterValue to weaviate.outputs.
  • Allows group_by: str in aggregation queries.

Version 4.4.1

This patch version includes

  • Allows strings as input for groupBy arguments for aggregation.
  • Fixes for rate limit batching.

Version 4.4.0

This version is the first full release for the Python v4 client and _requires_ weaviate versions >= 1.23.7.

Since the previous RC, there have been a number of improvements and final bug fixes. - The type of object.vector has changed from Optional[Dict[str, List[float]]] to Dict[str, List[float]] so that object.vector is never None. - Exporting and importing of collections has been tidied up and improved. - A number of methods have had input validation added to them. - Most exceptions are now unified under a few common classes.

For more information around the new client, see here: https://weaviate.io/developers/weaviate/client-libraries/python

Version 4.4.rc1

This version is a release candidate for the python v4 client.

There is a significant breaking change in this version in anticipation of the named vectors functionality of future Weaviate versions. - The vector property of Object has had its type changed from Optional[List[float]] to Optional[Dict[str, List[float]]]. - Accessing of the vector property has changed from object.vector to object.vector["default"]. - When using the client with future releases, other named vectors will be accessible as object.vector["name"].

Newly created (as of 15:00UTC 01/30/24) WCS sandbox instances are now capable of handling gRPC connections and so the client has been updated accordingly in its connect_to_wcs method. If you are using an old sandbox, make a new one and use the new one instead.

Minor bugfixes are also included.

Version 4.4.rc0

This version is a release candidate for the python v4 client.

All backward compatibility code is being removed and _requires_ weaviate versions >= 1.23.5.

All deprecated code has been removed. Check the migration guide (https://www.weaviate.io/developers/weaviate/client-libraries/python#migration-guides) how to update your code.

Improvements include: - Input validation - Embedded weaviate shows an error when the chosen port(s) are already occupied

Fixes include: - Filter chained references by reference count - Various bug with filtered aggregation - Aggregation with move to/away_from objects - Timeouts also apply to GRPC calls

Version 4.4.b9

This beta version has breaking changes, a migration guide is available at https://www.weaviate.io/developers/weaviate/client-libraries/python#migration-guides:

  • The batching algorithm has been streamlined and improved in its implementation and API surface.
    • There are now three types of batching that can be performed:
      • client.batch.dynamic() where the algorithm will automatically determine the optimal batch size and number of concurrent requests.
      • client.batch.fixed_size() where the user can specify the batch size and number of concurrent requests.
      • client.batch.rate_limit() where the user specifies the number of requests per minute that their third-party vectorization API can support.
    • If an exception is thrown in the background batching thread then this is surfaced to the main thread and re-raised in order to stop the batch.
      • Previously, this would silently error.
  • Enforces that all optional arguments to queries must be supplied as keyword arguments.
  • Adds runtime validation to all queries.
  • Renaming of prop to name in Filter.by_property.
  • Moving of the timeout argument in weaviate.connect_to_x methods into new argument additional_config: Optional[AdditionalConfig].

Improvements include: - Introduction of the .by_ref_count() method on Filter to filter on the number of references present in a reference property of an object.

  • This was previously achievable with Filter([refProp]).greater_than(0) but is now more explicit using the chaining syntax.
  • The syntax for sorting now feels similar to the new filtering syntax.
    • Supports method chaining like Sort.by_property(prop).by_creation_time() which will apply the sorting in the order they are chained, i.e., this chain is equivalent to the previous syntax of [Sort(prop), Sort("_creationTimeUnix")].

Fixes include: - The potential for deadlocks and data races when batching has been reduced. - Fixes a number of missing properties and poor docstrings in weaviate.connect_to_x methods. - Adds the missing offset parameter to all queries.

Version 4.4.b8

This beta version has breaking changes, a migration guide is available at https://www.weaviate.io/developers/weaviate/client-libraries/python#migration-guides:

  • Filters have been reworked and have a new syntax.
    • Coming from <=4.4.b6 you can replace:
      • Filter(path=property) with Filter.by_property(property)
      • Filter(path=["ref","target_class", "target_property"]) with Filter.by_ref("ref").by_property("target_property")
      • FilterMetadata.ByXX``with ``Filter.by_id/creation_time/update_time()
    • Coming from =4.4b7 you can replace:
      • Filter.by_ref().link_on("ref").by_property("target_property") with Filter.by_ref("ref").by_property("target_property")

Bugfixes include: - Error message when creating the client directly without calling connect_to_XXX. - Fix deadlock in new batching algorithm. - Fix skip_init_checks=True resulting in compatibility with Weaviate 1.22 only.

Version 4.4.b7

This beta version has breaking changes, a migration guide is available at https://www.weaviate.io/developers/weaviate/client-libraries/python#migration-guides:

  • For client.batch the add_reference method was revised. The to_object_collection parameter was removed and the other parameters were harmonized with collection.batch. Available parameters are now: from_uuid, from_collection, from_property, to and tenant.
  • It is no longer possible to use client.batch directly, you must use it as a context manager (with client.batch as batch)
  • Manual batch mode has been removed.
  • Dynamic batching (for batch_size and number of concurrent requests) is now default. Fixed-size batching can be configured with batch.configure_fixed_size(..).
  • Filters have been reworked and have a new syntax. You can replace:
    • Filter(path=property) with Filter.by_property(property)
    • Filter(path=["ref","target_class", "target_property"]) with Filter.by_ref().link_on("ref").by_property("target_property")
    • FilterMetadata.ByXX``with ``Filter.by_id/creation_time/update_time()
  • Importing directly from weaviate has been deprecated. Use import weaviate.classes as wvc instead and import from there.
  • Multi-target references functions have been moved to:
    • ReferenceProperty.MultiTarget
    • DataReference.MultiTarget
    • QueryReference.MultiTarget
  • Exception names are now compatible with PEP8, old names are still available but deprecated.
  • References can now be provided directly as UUIDs, str and Reference.XXX() has been deprecated. For multi-target references use ReferenceToMulti.

New functionality includes: - New batching algorithm that supports dynamic scaling of batch-size and number of concurrent requests. - New filter syntax that also supports structured filtering on references for normal properties and metadata. - All reference functions have unified input formats and now accept UUID, str and (where applicable) List[str], List[UUID]. - Returned types are now available in weaviate.output. - Add missing classes to weaviate.classes. - Add missing parameters to connect_to_XXX, all functions should support skipping of init checks and auth. - The client can now be used in a context manager with connect_to_XX(..) as client and all connections will be closed when exiting the manager. - New close function client.close() that needs to be called when not using a context manager to avoid stale connections and potential memory leaks. - Support for Phonenumber datatype. - Referenced objects now contain the name of their collection. - Adds collection.config.update_shards().

Bugfixes include: - object.reference is empty instead of None, if an object does not have a reference. - Fixes creating backups on weaviate master. - Add missing classes to wvc.

New client usage: - Client as a context manager:

with weaviate.connect_to_local() as client:
    # Your code

  • Client without a context manager:

    try:
    client = weaviate.connect_to_local()
    # Your code
finally:
    client.close()

Version 4.4.b6

This beta version includes:

  • A fix to the _Property dataclass returned within collection.config.get() to include any nested_properties of object and object[] type properties
  • Fix batch inserts with empty lists

Version 4.4.b5

This beta version includes:

  • fetch_object_by_id with Weaviate 1.22 returned None for non-existing references
  • empty strings in returned objects caused a panic with weaviate 1.22
  • Support for nodes/cluster API
  • Speed up client creation when connecting to WCS using connect_to_wcs
  • Checks GRPC availability of Weaviate instance and return an error if it is not supported yet
  • Adds skip_init_checks to connect_to_wcs

With the next Weaviate version (1.23.1) this beta version supports: - Blob properties - Reranker

Version 4.4.b4

This beta version fixes an issue with being unable to disable PQ once enabled

Version 4.4.b3

This beta version fixes a naming issue: - All instances of quantitizer have been renamed to quantizer

Version 4.4.b2

This version works best with Weaviate 1.23 which was released on 2023-12-18.

This beta version has breaking changes, a migration guide is available at https://www.weaviate.io/developers/weaviate/client-libraries/python#migration-guides:

  • Refactor weaviate.classes structure
  • Rename various classes and methods:
    • In all vectorizer configuration methods: vectorize_class_name => vectorize_collection_name
    • object.metadata.creation_time_unix => object.metadata.creation_time which is now a datetime
    • object.metadata.last_update_time_unix => object.metadata.last_update_time which is now a datetime
    • MetadataQuery(creation_time_unix=.., last_update_time_unix= ..) => MetadataQuery(creation_time=.., last_update_time=..)
    • FromReference => QueryReference when querying references
  • Splits out references from properties when creating, changing and querying collections
  • UUID and UUID_ARRAY properties are now returned as typed UUID objects
  • DATE and DATE_ARRAY properties are now returned as typed datetime objects
  • vector_index_type``has been remove from ``collection.create() and is now determined automatically
  • Configure.vector_index() has been moved to Configure.VectorIndex.hnsw()
  • PQ can now be configured using Configure.VectorIndex.hnsw(quantitizer=Configure.VectorIndex.Quantitizer.pq(..options..))
  • object.metadata.vector was moved to object.vector and can be requested by using include_vector=True/False when querying
  • object.metadata.uuid was moved to object.uuid and is always available
  • Order of arguments in .data.update() and .replace() changed to accommodate not providing properties when updating.
  • In .data.reference_add, .reference_delete and .reference_replace the ref keyword was renamed to to
  • In collections.create() and .get() the keyword to provide generics was renamed from data_model to data_model_properties

New functionality includes:

  • Adds backup functionality to v4 client (client.backup) and directly to the collection (collection.backup)
  • Adds support for FLAT vector index
  • Adds binary quantization for FLAT vector index
  • Adds text2vec_jinaai static method to Configure.Vectorizer
  • Adds anyscale static method to Configure.Generative
  • Adds collection.batch for uploading to a single collection in batches
  • Adds methods for creating a collection from dict and exporting a collection config as dict
  • Adds support for geo-coordinates
  • Adds metadata filtering with FilterMetadata
  • Adds client.graphql_raw_query to use Weaviate features that are not directly supported.
  • Adds DataReferenceOneToMany which allows to add multiple references at once.
  • Adds validation of input parameters for non-mypy users.
  • Various performance improvements and bugfixes

Version 4.4.b1

This patch beta version includes:

  • Performance improvements when making queries

Version 4.4.b0

This minor beta version includes:

  • Adds support for connecting to WCS using the connect_to_wcs helper function
  • Changes default num_workers in client.batch from 1 to Python's ThreadPoolExecutor default
  • Adds text2vec-aws and generative-aws static methods to Configure.Vectorizer and Configure.Generative
  • Tidy up stale docstrings
  • Add missing class exports

Version 4.3.b2

This patch beta version includes:

  • Fixes to the dataclass types returned by aggregate queries

Version 4.3.b1

This patch beta version includes:

  • Bump default Weaviate embedded version

Version 4.3.b0

This minor beta version includes:

  • Refactoring of the _Object class
    • _Object.metadata.uuid moved to _Object.uuid and is not Optional
    • _Object.metadata.vector moved to _Object.vector
  • Addition of include_vector argument to all queries
    • include_vector is False by default
  • return_metadata in queries is now Optional and defaults to None
    • _Object.metadata is now Optional as a result
  • Addition of include_vector to FromReference
  • Addition of ReferenceAnnotation for use when defining generic annotated cross references

Version 4.2.b2

This patch beta version includes:

  • Allow None when batch inserting using DataObject and BatchObject

Version 4.2.b1

This patch beta version includes:

  • Bug fix of the default alpha argument to query.hybrid
  • Extend the Configure.Vectorizer.multi2vec_ methods to accept lists of strings
  • Correctly export StopwordsPreset from weaviate.classes
  • Add generative_config and vectorizer_config to _CollectionConfig
  • Add skip_vectorization and vectorize_class_name to _PropertyConfig

Version 4.2.b0

This minor beta version includes:

  • A refactoring of the collection.aggregate namespace methods
  • Change Metrics to no longer accept the type_ argument
  • Instead, Metrics has multiple methods, e.g. .text(), for each type of metric
  • Allow return_metrics to be a single metric object or a list of metric objects in each aggregate query

Version 4.1.b2

This patch beta version includes:

  • Correctly exporting weaviate.collections.classes.aggregate.Metrics from weaviate.classes

Version 4.1.b1

This patch beta version includes:

  • Bumping the default embedded version to Weaviate latest
  • Adding the version argument to weaviate.connect_to_embedded to allow users to specify the embedded version

Version 4.1.b0

This minor beta version includes:

  • Makes total_count=True the default in aggregation queries to avoid unintentional GraphQL errors
  • Catches empty GraphQL errors in aggregation queries in case of user error
  • Renames class_name to collections within the collections.batch namespace
  • Adds get_vector to the collections.data namespace so that users can supply numpy and pytorch vectors
  • Adds __str__ magic method to Collections class so that print(collection) outputs the collection's schema as pretty JSON

Version 4.0.b5

This patch beta version includes:

  • Update changelog

Version 4.0.b4

This patch beta version includes:

  • A small bug fix to remove a redundant print
  • Raising an exception from connect_to_wcs as gRPC support is not ready
  • Making _Collection a public class as Collection to be used in type hinting

Version 4.0.b3

This patch beta version includes:

  • Addition of batch_size to client.batch.configure for users who want automatic non-dynamic batching
  • Renaming of ConfigureUpdate to Reconfigure
  • Fixing of missing arguments to Configure.Vectorizer.text2vec_ methods

Version 4.0.b2

This patch beta version includes:

  • Fixes to the readthedocs documentation appearance

Version 4.0.b1

This beta version includes:

  • Introduction of the new beta Python collections client API
    • Streamlined and simplified client API for mutating and querying your data
    • Full support for gRPC batching and searching
    • End-to-end generics support for type safety
    • Python-native dataclasses for easy data manipulation
    • No more builder methods or raw dictionaries
  • Join the discussion and contribute your feedback here

Version 3.26.5

This patch version includes

  • Fixes GraphQL query injection vulnerability caused by incorrect escaping of backslashes in plain text input builder methods
  • Many thanks to @adamleko, @bismuthsalamander, and @tardigrade-9 for their help in fixing this issue

Version 3.26.4

This patch version includes

  • Fixes batch retry with tenants

Version 3.26.2

This patch version includes

  • Adds a timeout to wait_for_weaviate startup check

Version 3.26.1

This patch version includes

  • Fix backup creation with current weaviate master

Version 3.26.0

This minor version includes:

  • Support for Weaviate 1.23
  • Bump of the default version for Weaviate Embedded DB to v1.23.0
  • Adds support for nodes api verbosity option

Version 3.25.3

This patch version includes

  • Bump of the default version for Weaviate Embedded DB to v1.22.3

Version 3.25.2

This patch version includes

  • Fixes to the codebase naming convention and directory structure to prevent collision with Google's proto-plus library
  • Fixes to the build method so that readthedocs.io builds the documentation correctly again

Version 3.25.1

This patch version includes:

  • Bump default embedded version to 1.22.0

Version 3.25.0

This minor version includes:

  • Support for new Weaviate nested objects on insert and query
    • client.data_object.create() now supports nested objects
    • client.query.get() now supports nested objects
  • Updates to use Weaviate's v1 gRPC API
  • Support for batching with Weaviate>1.22.0 version and async vector indexing
  • Addition of the client.batch.wait_for_async_indexing() method to force block until async indexing is complete
  • Add tests for Python 3.12 to ensure compatibility

Version 3.24.2

This patch version includes:

  • Small fix to the batching process to ensure that failed multi-tenant objects are re-added to the batch with their tenant attached

Version 3.24.1

This patch version updates the changelog.rst that became stale over the last few releases

Version 3.24.0

This minor version includes:

  • Small fixes and improvements throughout the codebase:
    • Catching and reraising of JsonDecodeException for users to catch
    • Client-wide mypy error fixing and type hinting improvements
    • Fix for where filter operands in batch.delete_objects
    • Removal of buggy client-side schema validation
    • Package dependency updates

Version 3.23.2

This patch version includes:

  • Enforcing class name capitalization throughout the client
  • Further fixes to where filtering with ContainsAny/All

Version 3.23.1

This patch version includes:

  • Enabling of rerank-cohere module in EmbeddedWeaviate
  • Fixes for where filtering between query.get over GraphQL and batch.delete_objects over REST

Version 3.23.0

This minor version updates the client to work with Weaviate's 1.21 version and includes:

  • Adds support for near<Media> filters when using the new multi2vec-bind module for neural searching on different media types
    • client.query.get().with_near_audio()
    • client.query.get().with_near_depth()
    • client.query.get().with_near_image() (unchanged from previous versions but usable by the module)
    • client.query.get().with_near_imu()
    • client.query.get().with_near_thermal()
    • client.query.get().with_near_video()
  • Deprecates configuring client.batch using client.batch() in favour of using client.batch.configure()
    • client.batch() will be removed in a future version
    • client.batch.configure() will return None in a future version
    • with client.batch as batch should be the standard way to initiate a batch
  • Adds support for new ContainsAny and ContainsAll filters when using .with_where
  • Adds support for updating individual tenants within a multi-tenancy class configuration: client.schema.update_class_tenants
  • Improves client.batch algorithm to choose batch size dynamically maximizing throughput
  • Provides sensible defaults to client.batch that do not cause unexpected damaging consequences like infinite batch sizes
  • Fixes bugs when using .with_where with valueText, valueString, and valueGeoRange types

Version 3.22.1

This patch version includes:

  • Fix "is client outdated"-check in air-gaped environments
  • Add tenant to batch delete

Version 3.22.0

This minor version includes:

  • Multi-tenancy
  • Aggregate with limit
  • Autocut
  • Fusion type for hybrid search
  • Client emits a warning when it is outdated (three minor version behind last release on pypi)
  • Increase default embedded version to 1.19.12

Version 3.21.0

This minor version includes: - Weaviate Embedded supports MacOs

Version 3.20.1

This patch version includes:

  • Fix imports without GRPC package
  • Improve shutdown handling with Weaviate Embedded

Version 3.20.0

This minor version includes:

  • Increase maximum version of request library to 2.31.0. This also updates to urllib 2.0. This may contain minor breaking changes if you use urllib in other projects in the same virtual environment.
  • Add licensing information to pypi package
  • Increase default embedded version to 1.19.7

Version 3.19.2

This patch version includes:

  • Add custom headers to all requests
  • Support properties field in generative groupedResult field

Version 3.19.1

This patch version includes:

  • Fixes imports of of weaviate_pb2.

Version 3.19.0

This minor version includes:

  • Increases default embedded version to 1.19.3

  • Clients emits warning if used weaviate version is too old (3 versions behind latest minor version)

  • Adds native support for querying reference properties

    result = client.query.get(
  "Article", ["title", "url", "wordCount", LinkTo(link_on="caller", linked_class="Person", properties=["name"])]
     )

  • Adds dataclasses to easier access to additional properties:

    query = client.query.get("Test").with_additional(
            weaviate.AdditionalProperties(
                uuid=True,
                vector=True,
                creationTimeUnix=True,
                lastUpdateTimeUnix=True,
                distance=True,
            )
        )

  • Typing fixes

  • Expand support for experimental GRPC API and add support for
    • BM25 and hybrid search
    • Additional properties (via dataclass shown above)
    • Querying reference properties (via dataclass shown above)

Version 3.18.0

This minor version includes:

  • Add support for properties with hybrid search
  • Fixes documentation publishing on readthedocs

Version 3.17.1

This patch version includes:

  • Fix schemas with new property keys indexFilterable and indexSearchable.

Version 3.17.0

This minor version includes:

  • Add support for groupBy to group objects:

    .with_group_by(properties=["caller"], groups=2, objects_per_group=3)

  • Add support for uuid and uuid[] datatypes.

  • Add schema.exists(class).

  • Add support for Support GQL Get{} tunable consistency:

    resp = (
    client.query.get("Article", ["name"])
    .with_additional("isConsistent")
    .with_consistency_level(ConsistencyLevel.ALL)
    .do()
)

Version 3.16.2

This patch version includes:

  • Fix url containing username and password.

Version 3.16.1

This patch version includes:

  • Fixes timeout error in detection of grpc.

Version 3.16.0

This minor version includes:

  • Experimental support for GRPC.
    • Can by enabled by installing the client with pip install weaviate-client[GRPC] or install the grpcio package manually.
    • To disable uninstall the grpcio package.
    • This will speed up certain GraphQL queries: Get with NearObject or NearVector if only non-reference queries are retrieved and no other options are set.
  • Removal of python 3.7 support. Minimum supported version is python 3.8
  • Removal of the WCS module. Note that the module was used to administrate old WCS instances and does not work anymore.

Version 3.15.6

This patch version includes:

  • Fix multi-line queries for BM25 and hybrid search.

Version 3.15.5

This patch version includes:

  • EmbeddedDB now supports latest and versions (eg 1.18.3) as version argument.
  • Removed cluster_hostname from EmbeddedOptions. It can still be set by using additional_env_vars.
  • Fix multi-line queries for generative search.

Version 3.15.4

This patch version includes:

  • Fix imports of EmbeddedDB on Mac. It now properly raises an exception that MacOS is currently unsupported.

Version 3.15.3

This patch version includes:

  • Improve embedded weaviate: Better folder structures, add support for env variables and support multiple versions.
  • Fix edge case for timeout retries: When all objects have been added, no empty batch will be send.
  • Fix authentication via additional_headers

Version 3.15.2

This patch version includes:

  • Fixes API keys with Weaviate setups that do not have OIDC enabled.

Version 3.15.1

This patch version includes:

  • Fixes refreshing of OIDC tokens on unstable connections

Version 3.15.0

This minor version includes:

  • GraphQL Multiple queries and aliases support:

    client.query.multi_get(
        [
           client.query.get("Ship", ["name"]).with_alias("one"),
           client.query.get("Ship", ["size"]).with_alias("two"),
           client.query.get("Person", ["name"])
        ]

  • Adds support for embedded weaviate version:

    from weaviate import Client
from weaviate.embedded import EmbeddedOptions

# Create the embedded client which automatically launches a Weaviate database in the background
client = Client(embedded_options=EmbeddedOptions())

Version 3.14.0

This minor version includes:

  • Support for API-Keys:

    client = weaviate.Client(url, auth_client_secret=AuthApiKey(api_key="my-secret-key"))

Version 3.13.0

This minor version includes:

  • Extend CRUD operations for single data objects and reference with consistency level.
  • Extend batch operations with consistency level.
  • Add Cursor api.
  • Add support for azure backup module.

Version 3.12.0

This minor version includes:

Version 3.11.0

This minor version includes:

Version 3.10.0

This minor version includes:

Version 3.9.0

This minor version includes:

Version 3.8.0

This minor version includes:

Version 3.7.0

This minor version includes:

Version 3.6.0

This minor version includes:

Version 3.5.1

This patch version fixes:

Version 3.5.0

This minor version contains functionality for the new features introduced in Weaviate v1.13.0.

Version 3.4.2

This patch version fixes another bug in :meth:`~weaviate.data.DataObject.exists`.

Version 3.4.1

This patch version fixes bug in :meth:`~weaviate.data.DataObject.exists`.

Version 3.4.0

This minor version fixes the bug in setting the Schema's invertedIndexConfig field.
New method :meth:`~weaviate.schema.Schema.get_class_shards` to get all shards configuration of a particular class.
New method :meth:`~weaviate.schema.Schema.update_class_shard` to update one/all shard/s configuration of a particular class.
Support for new Property field: tokenization.

Version 3.3.3

This patch version fixes the nearImage filter requests.

Version 3.3.2

This patch version allows using UUIDs in hex format for :class:`~weaviate.data.DataObject` too i.e. UUIDs without hyphens.

Version 3.3.1

This patch version allows using UUIDs in hex format too i.e. UUIDs without hyphens.

Version 3.3.0

This minor version adds a new :meth:`~weaviate.gql.get.GetBuilder.with_offset` for the Get queries. This method should be used with the :meth:`~weaviate.gql.get.GetBuilder.with_limit`. This new feature (introduced in weaviate version 1.8.0) allows to use pagination functionality with the Get queries. The offset represents the start index of the objects to be returned, and the number of objects is specified by the :meth:`~weaviate.gql.get.GetBuilder.with_limit` method.
For example, to list the first ten results, set limit: 10. Then, to "display the second page of 10", set offset: 10, limit: 10 and so on. E.g. to show the 9th page of 10 results, set offset: 80, limit: 10 to effectively display results 81-90.

Version 3.2.5

This patch fixes the 'Batch' object is not callable error.

Version 3.2.4

All class_name and cross-refs dataType are implicitly capitalized. (This functionality is added because if class_name is not capitalized then Weaviate server does it for you, and this was leading to errors where the client and server have different configurations.)

Fixes/updates in :class:`~weaviate.schema.Schema` class:

  • This patch fixes the :meth:`~weaviate.schema.Schema.contains` to accept separate class schemas as argument i.e. it does not expect to have only this format: {"classes": [CLASS_1, CLASS_2, ...]}; now it is possible to pass just CLASS_X as well.

Version 3.2.3

This patch fixes the :meth:`~weaviate.gql.get.GetBuilder.with_near_object`. It uses now explicit string literals for id/beacon in nearoOject clauses.

Version 3.2.2

This patch adds support for array data types: boolean[], date[].

Version 3.2.1

This patch adds support for array data types: int[], number[], text[], string[].

Version 3.2.0

Fixes/updates in :class:`~weaviate.wcs.WCS` class:

Updates in :mod:`~weaviate.gql` sub-package:

# with 'nearText' filter
client.query\
    .get('Article', ['title', 'author'])\
    .near_text(
        {
            'concepts': ['Ecconomy'],
            'autocorrect': True
        }
    )
    # the concept should be corrected to 'Economy'
# with 'ask' filter
client.query\
    .get('Article', ['title', 'author'])\
    .with_ask(
        {
            'question': 'When was the last financial crysis?',
            'autocorrect': True
        }
    )
    # the question should be corrected to 'When was the last financial crisis?'
# single additional property with this GraphQL query
'''
{
    Get {
        Article {
            title
            author
            _additional {
                id
            }
        }
    }
}
'''
client.query\
    .get('Article', ['title', 'author'])\
    .with_additional('id') # argument as `str`

# multiple additional property with this GraphQL query
'''
{
    Get {
        Article {
            title
            author
            _additional {
                id
                certainty
            }
        }
    }
}
'''
client.query\
    .get('Article', ['title', 'author'])\
    .with_additional(['id', 'certainty']) # argument as `List[str]`

# additional properties as clause with this GraphQL query
'''
{
    Get {
        Article {
            title
            author
            _additional {
                classification {
                    basedOn
                    classifiedFields
                    completed
                    id
                    scope
                }
            }
        }
    }
}
'''
client.query\
    .get('Article', ['title', 'author'])\
    .with_additional(
        {
            'classification' : [
                'basedOn',
                'classifiedFields',
                'completed',
                'id',
                'scope'
            ]
        }
    ) # argument as `Dict[str, List[str]]`

# or with this GraphQL query
'''
{
    Get {
        Article {
            title
            author
            _additional {
                classification {
                    completed
                }
            }
        }
    }
}
'''
client.query\
    .get('Article', ['title', 'author'])\
    .with_additional(
        {
            'classification' : 'completed'
        }
    ) # argument as `Dict[str, str]`

# additional properties as clause and clause settings with this GraphQL query
'''
{
    Get {
        Article {
            title
            author
            _additional {
                token (
                    properties: ["content"]
                    limit: 10
                    certainty: 0.8
                ) {
                    certainty
                    endPosition
                    entity
                    property
                    startPosition
                    word
                }
            }
        }
    }
}
'''
clause = {
    'token': [
        'certainty',
        'endPosition',
        'entity',
        'property',
        'startPosition',
        'word',
    ]
}
settings = {
    'properties': ["content"],  # is required
    'limit': 10,                # optional, int
    'certainty': 0.8            # optional, float
}
client.query\
    .get('Article', ['title', 'author'])\
    .with_additional(
        (clause, settings)
    ) # argument as `Tuple[Dict[str, List[str]], Dict[str, Any]]`

# if the desired clause does not match any example above, then the clause can always
# be converted to string before passing it to the `.with_additional` method

Version 3.1.1

  • Fixes in :class:`~weaviate.wcs.WCS` class:
    • Make :class:`~weaviate.wcs.WCS`'s methods' argument cluster_name case insensitive (lowercased inside the method) to match Weaviate Cloud Service' naming convention, this fixes the error when Weaviate Cloud Service lowercases the cluster_name but the users are not aware of this and get the exception KeyError.

Version 3.1.0

Version 3.0.0

  • weaviate.tools module is REMOVED.
    • Batcher class is REMOVED.
    • WCS class is moved from the weaviate.tools to the new module weaviate.wcs
    • weaviate.tools.generate_uuid is REMOVED.

  • :func:`weaviate.util.generate_uuid5` is ADDED.

  • New :class:`~weaviate.batch.Batch` class implementation to replace the old one. This implementation uses the BatchRequest objects under the hood, which means that there is no need to create BatchRequest's anymore. This new class implementation allows 3 different batch creations methods: manual, auto-create and auto-create with dynamic batching. See the :class:`~weaviate.batch.Batch` documentation for more information.
  • BatchRequest classes (ObjectsBatchRequest and ReferenceBatchRequest) are hidden from the user and should not be used anymore. This is due to the new :class:`~weaviate.batch.Batch` class implementation.
  • New :class:`~weaviate.schema.Schema` field is ADDED, "shardingConfig". It can bu used with Weaviate version >= 1.6.0.
  • New method :meth:`~weaviate.schema.Schema.update_config` used to update mutable schema configuration (like efConstruction, ...).