chroma-core/chroma · error · ValueError

The model name cannot be changed after the embedding functio

Error message

The model name cannot be changed after the embedding function has been initialized.

What it means

When a collection's embedding function is replaced, chromadb calls update_ef.validate_config_update(old_config, new_config) via overwrite_embedding_function (chromadb/api/collection_configuration.py:711, reached from Collection.modify). AmazonBedrockEmbeddingFunction rejects any new_config that contains the "model_name" key, because changing the embedding model would make all existing vectors in the collection incompatible. Caveat: the guard checks key presence, not value change — and get_config() always includes "model_name" — so effectively any bedrock-to-bedrock embedding function update raises, even with an identical model.

Source

Thrown at chromadb/utils/embedding_functions/amazon_bedrock_embedding_function.py:123

        else:
            session = boto3.Session(**session_args)

        return AmazonBedrockEmbeddingFunction(
            session=session, model_name=model_name, **kwargs
        )

    def get_config(self) -> Dict[str, Any]:
        return {
            "model_name": self.model_name,
            "session_args": self._session_args,
            "kwargs": self.kwargs,
        }

    def validate_config_update(
        self, old_config: Dict[str, Any], new_config: Dict[str, Any]
    ) -> None:
        if "model_name" in new_config:
            raise ValueError(
                "The model name cannot be changed after the embedding function has been initialized."
            )

    @staticmethod
    def validate_config(config: Dict[str, Any]) -> None:
        """
        Validate the configuration using the JSON schema.

        Args:
            config: Configuration to validate

        Raises:
            ValidationError: If the configuration does not match the schema
        """
        validate_config_schema(config, "amazon_bedrock")

View on GitHub (pinned to aecdd12c8a)

Solutions

1. Create a new collection with the desired bedrock model and re-embed your documents — model changes cannot be applied in place.
2. If you only meant to change other collection settings (metadata, hnsw config), call modify() without an embedding_function so this validation never runs.
3. If you believe unchanged model_name should be allowed, file an upstream issue: the check should compare old vs new values, not mere key presence.

Example fix

# before: raises ValueError "The model name cannot be changed..."
client.get_collection("docs", embedding_function=old_ef).modify(
    configuration=CollectionConfiguration(
        embedding_function=AmazonBedrockEmbeddingFunction(session=new_session)
    )
)

# after: model is fixed per collection — make a new one and re-embed
new_col = client.create_collection("docs_v2", embedding_function=AmazonBedrockEmbeddingFunction(session=new_session))
for batch in read_old_batches():
    new_col.add(**batch)
# or, to change only non-EF settings: col.modify(metadata={...}) with no embedding_function

from chromadb.utils.embedding_functions import AmazonBedrockEmbeddingFunction

def can_update_bedrock_ef(new_ef) -> bool:
    try:
        new_ef.validate_config_update(new_ef.get_config(), {k: v for k, v in new_ef.get_config().items() if k != "model_name"})
        return True
    except ValueError:
        return False  # model_name change attempted — must create a new collection instead

if can_update_bedrock_ef(new_ef):
    col.modify(configuration=CollectionConfiguration(embedding_function=new_ef))
else:
    raise RuntimeError("Bedrock model_name is fixed per collection; create a new collection and re-embed")

try:
    col.modify(configuration=CollectionConfiguration(embedding_function=new_ef))
except ValueError as e:
    if "model name cannot be changed" in str(e):
        # model is immutable per collection — migrate instead
        new_col = client.create_collection(f"{col.name}_v2", embedding_function=new_ef)
        migrate_documents(col, new_col)
    else:
        raise

When it happens

Trigger: collection.modify(configuration=CollectionConfiguration(embedding_function=AmazonBedrockEmbeddingFunction(...))) on a collection whose current embedding function is also amazon_bedrock; overwrite_embedding_function passes the new function's get_config() (which always contains model_name) straight into this check.

Common situations: Trying to tweak session args or kwargs on an existing collection; attempting a model-version migration in place; copy-pasting the original constructor into a modify() call while debugging.

Related errors

• Keyword argument {key} is not a primitive type
• The boto3 python package is not installed. Please install it
• Failed to register embedding function: {e}
• Failed to register sparse embedding function: {e}
• Config must contain a 'name' field.

AI-assisted analysis of chroma-core/chroma@aecdd12c8a (2026-08-16). Data as JSON: /api/errors/1eff8b58affcc4c8. Report an issue: GitHub.