Skip Vectorization Not Applying To Added Properties When Generative Config Is Not Null

Support
,
1

Description

When I add a property to an existing class skip_vectorization is not being propagated to the _PropertyVectorizerConfig making it impossible to add a property with skip_vectorization=True. It seems only to happen if generative_config is not null. If generative_config is null, this bug doesn’t happen.

Server Setup Information

  • Weaviate Server Version: Docker Image 1.28.2
  • Deployment Method: Docker
  • Multi Node? Number of Running Nodes: No
  • Client Language and Version: Python 3.12
  • Multitenancy?: No

Any additional Information

Code:
client = weaviate.connect_to_local(host=‘127.0.0.1’,port=8080,grpc_port=50051)
try:
client.collections.delete(‘Test’)
except:
pass

client.collections.create(
“Test”,
vectorizer_config=wvc.config.Configure.Vectorizer.text2vec_ollama(
api_endpoint=“http://localhost:11434”,
model=“snowflake-arctic-embed2:568m”
),
generative_config=Configure.Generative.ollama(
api_endpoint=“http://localhost:11434”, # If using Docker, use this to contact your local Ollama instance
model=“dolphin3:8b-llama3.1-q4_K_M” # The model to use, e.g. “phi3”, or “mistral”, “command-r-plus”, “gemma”
),
properties=[
Property(name=“property_made_during_class_creation”,data_type=DataType.BOOL,vectorize_property_name=False,skip_vectorization=True),

]

)
testCollection = client.collections.get(‘Test’)
testCollection.config.add_property(Property(name=“property_made_after_class_creation”,data_type=DataType.BOOL,vectorize_property_name=False,skip_vectorization=True))
testCollection.config.get().properties

Output:
[_Property(name=‘property_made_during_class_creation’, description=None, data_type=<DataType.BOOL: ‘boolean’>, index_filterable=True, index_range_filters=False, index_searchable=False, nested_properties=None, tokenization=None, vectorizer_config=_PropertyVectorizerConfig(skip=True, vectorize_property_name=False), vectorizer=‘text2vec-ollama’),
_Property(name=‘property_made_after_class_creation’, description=None, data_type=<DataType.BOOL: ‘boolean’>, index_filterable=True, index_range_filters=False, index_searchable=False, nested_properties=None, tokenization=None, vectorizer_config=_PropertyVectorizerConfig(skip=False, vectorize_property_name=False), vectorizer=‘text2vec-ollama’)]

Expected Behavior:
Both properties would have _PropertyVectorizerConfig.skip = True

Deployment Config:

services:
weaviate:
command:
- --host
- 0.0.0.0
- --port
- ‘8080’
- --scheme
- http
image: cr.weaviate.io/semitechnologies/weaviate:1.28.2
network_mode: ‘host’
volumes:
- /Volume:/var/lib/weaviate
restart: on-failure:0
environment:
QUERY_DEFAULTS_LIMIT: 25
AUTHENTICATION_ANONYMOUS_ACCESS_ENABLED: ‘true’
PERSISTENCE_DATA_PATH: ‘/var/lib/weaviate’
ENABLE_API_BASED_MODULES: ‘true’
CLUSTER_HOSTNAME: ‘node1’
ENABLE_MODULES: ‘text2vec-ollama,generative-ollama’
OLLAMA_API_BASE_URL: ‘http://127.0.0.1:11434/api’
~

2

Hello @clpurcell,

I’ve tried to replicate the behavior you described using the Python Client: 4.10.4, and I observed that my property reflects with skip=True as expected.

At this point, I’m not certain if this is a bug, but it’s important to verify this on the latest client version rather than on a deprecated one. Is there a specific reason you’re not using Client V4?

It’s possible that my replication setup isn’t exactly like yours—I’ve configured my vectorizer and generative model in a collection, then created properties, and later added a property with skip vectorization, and it all reflected fine.

Could you please retry your tests using latest client version, and if the issue persists, could you send me your script code for further debugging? The reason I recommend this is that if a bug is present in V3, which is deprecated, the likely recommendation would be to upgrade to V4. This way, we can determine if the issue is client-related or server-related.

Best regards,
Mohamed Shahin
Weaviate Support