diff --git a/ollama/_client.py b/ollama/_client.py index 0a85a74..4bcc1b1 100644 --- a/ollama/_client.py +++ b/ollama/_client.py @@ -363,6 +363,7 @@ class Client(BaseClient): truncate: Optional[bool] = None, options: Optional[Union[Mapping[str, Any], Options]] = None, keep_alive: Optional[Union[float, str]] = None, + dimensions: Optional[int] = None, ) -> EmbedResponse: return self._request( EmbedResponse, @@ -374,6 +375,7 @@ class Client(BaseClient): truncate=truncate, options=options, keep_alive=keep_alive, + dimensions=dimensions, ).model_dump(exclude_none=True), ) @@ -875,6 +877,7 @@ class AsyncClient(BaseClient): truncate: Optional[bool] = None, options: Optional[Union[Mapping[str, Any], Options]] = None, keep_alive: Optional[Union[float, str]] = None, + dimensions: Optional[int] = None, ) -> EmbedResponse: return await self._request( EmbedResponse, @@ -886,6 +889,7 @@ class AsyncClient(BaseClient): truncate=truncate, options=options, keep_alive=keep_alive, + dimensions=dimensions, ).model_dump(exclude_none=True), ) diff --git a/ollama/_types.py b/ollama/_types.py index 0482287..8d3226e 100644 --- a/ollama/_types.py +++ b/ollama/_types.py @@ -382,6 +382,9 @@ class EmbedRequest(BaseRequest): keep_alive: Optional[Union[float, str]] = None + dimensions: Optional[int] = None + 'Dimensions truncates the output embedding to the specified dimension.' + class EmbedResponse(BaseGenerateResponse): """