chore: import upstream snapshot with attribution
This commit is contained in:
wehub-resource-sync
@@ -0,0 +1,21 @@
|
||||
# Copyright (c) Microsoft. All rights reserved.
|
||||
|
||||
from semantic_kernel.connectors.ai.ollama.ollama_prompt_execution_settings import (
|
||||
OllamaChatPromptExecutionSettings,
|
||||
OllamaEmbeddingPromptExecutionSettings,
|
||||
OllamaPromptExecutionSettings,
|
||||
OllamaTextPromptExecutionSettings,
|
||||
)
|
||||
from semantic_kernel.connectors.ai.ollama.services.ollama_chat_completion import OllamaChatCompletion
|
||||
from semantic_kernel.connectors.ai.ollama.services.ollama_text_completion import OllamaTextCompletion
|
||||
from semantic_kernel.connectors.ai.ollama.services.ollama_text_embedding import OllamaTextEmbedding
|
||||
|
||||
__all__ = [
|
||||
"OllamaChatCompletion",
|
||||
"OllamaChatPromptExecutionSettings",
|
||||
"OllamaEmbeddingPromptExecutionSettings",
|
||||
"OllamaPromptExecutionSettings",
|
||||
"OllamaTextCompletion",
|
||||
"OllamaTextEmbedding",
|
||||
"OllamaTextPromptExecutionSettings",
|
||||
]
|
||||
@@ -0,0 +1,39 @@
|
||||
# Copyright (c) Microsoft. All rights reserved.
|
||||
|
||||
from typing import Annotated, Any, Literal
|
||||
|
||||
from pydantic import Field
|
||||
|
||||
from semantic_kernel.connectors.ai.prompt_execution_settings import PromptExecutionSettings
|
||||
|
||||
|
||||
class OllamaPromptExecutionSettings(PromptExecutionSettings):
|
||||
"""Settings for Ollama prompt execution."""
|
||||
|
||||
format: Literal["json"] | dict[str, Any] | None = None
|
||||
options: dict[str, Any] | None = None
|
||||
|
||||
|
||||
class OllamaTextPromptExecutionSettings(OllamaPromptExecutionSettings):
|
||||
"""Settings for Ollama text prompt execution."""
|
||||
|
||||
system: str | None = None
|
||||
template: str | None = None
|
||||
context: str | None = None
|
||||
raw: bool | None = None
|
||||
|
||||
|
||||
class OllamaChatPromptExecutionSettings(OllamaPromptExecutionSettings):
|
||||
"""Settings for Ollama chat prompt execution."""
|
||||
|
||||
tools: Annotated[
|
||||
list[dict[str, Any]] | None,
|
||||
Field(
|
||||
description="Do not set this manually. It is set by the service based "
|
||||
"on the function choice configuration.",
|
||||
),
|
||||
] = None
|
||||
|
||||
|
||||
class OllamaEmbeddingPromptExecutionSettings(OllamaPromptExecutionSettings):
|
||||
"""Settings for Ollama embedding prompt execution."""
|
||||
@@ -0,0 +1,33 @@
|
||||
# Copyright (c) Microsoft. All rights reserved.
|
||||
|
||||
from typing import ClassVar
|
||||
|
||||
from semantic_kernel.kernel_pydantic import KernelBaseSettings
|
||||
|
||||
|
||||
class OllamaSettings(KernelBaseSettings):
|
||||
"""Ollama settings.
|
||||
|
||||
The settings are first loaded from environment variables with
|
||||
the prefix 'OLLAMA_'.
|
||||
If the environment variables are not found, the settings can
|
||||
be loaded from a .env file with the encoding 'utf-8'.
|
||||
If the settings are not found in the .env file, the settings
|
||||
are ignored; however, validation will fail alerting that the
|
||||
settings are missing.
|
||||
|
||||
Required settings for prefix 'OLLAMA' are:
|
||||
- chat_model_id: str - The chat model ID. (Env var OLLAMA_CHAT_MODEL_ID)
|
||||
- text_model_id: str - The text model ID. (Env var OLLAMA_TEXT_MODEL_ID)
|
||||
- embedding_model_id: str - The embedding model ID. (Env var OLLAMA_EMBEDDING_MODEL_ID)
|
||||
|
||||
Optional settings for prefix 'OLLAMA' are:
|
||||
- host: HttpsUrl - The endpoint of the Ollama service. (Env var OLLAMA_HOST)
|
||||
"""
|
||||
|
||||
env_prefix: ClassVar[str] = "OLLAMA_"
|
||||
|
||||
chat_model_id: str | None = None
|
||||
text_model_id: str | None = None
|
||||
embedding_model_id: str | None = None
|
||||
host: str | None = None
|
||||
@@ -0,0 +1,20 @@
|
||||
# Copyright (c) Microsoft. All rights reserved.
|
||||
|
||||
from abc import ABC
|
||||
from typing import ClassVar
|
||||
|
||||
from ollama import AsyncClient
|
||||
|
||||
from semantic_kernel.kernel_pydantic import KernelBaseModel
|
||||
|
||||
|
||||
class OllamaBase(KernelBaseModel, ABC):
|
||||
"""Ollama service base.
|
||||
|
||||
Args:
|
||||
client [AsyncClient]: An Ollama client to use for the service.
|
||||
"""
|
||||
|
||||
MODEL_PROVIDER_NAME: ClassVar[str] = "ollama"
|
||||
|
||||
client: AsyncClient
|
||||
@@ -0,0 +1,367 @@
|
||||
# Copyright (c) Microsoft. All rights reserved.
|
||||
|
||||
import logging
|
||||
import sys
|
||||
from collections.abc import AsyncGenerator, AsyncIterator, Callable, Mapping, Sequence
|
||||
from typing import TYPE_CHECKING, Any, ClassVar, TypeVar
|
||||
|
||||
if sys.version_info >= (3, 12):
|
||||
from typing import override # pragma: no cover
|
||||
else:
|
||||
from typing_extensions import override # pragma: no cover
|
||||
|
||||
import httpx
|
||||
from ollama import AsyncClient
|
||||
from ollama._types import ChatResponse, Message
|
||||
from pydantic import ValidationError
|
||||
|
||||
from semantic_kernel.connectors.ai.chat_completion_client_base import ChatCompletionClientBase
|
||||
from semantic_kernel.connectors.ai.completion_usage import CompletionUsage
|
||||
from semantic_kernel.connectors.ai.function_choice_behavior import FunctionChoiceType
|
||||
from semantic_kernel.connectors.ai.ollama.ollama_prompt_execution_settings import OllamaChatPromptExecutionSettings
|
||||
from semantic_kernel.connectors.ai.ollama.ollama_settings import OllamaSettings
|
||||
from semantic_kernel.connectors.ai.ollama.services.ollama_base import OllamaBase
|
||||
from semantic_kernel.connectors.ai.ollama.services.utils import (
|
||||
MESSAGE_CONVERTERS,
|
||||
update_settings_from_function_choice_configuration,
|
||||
)
|
||||
from semantic_kernel.contents import AuthorRole
|
||||
from semantic_kernel.contents.chat_history import ChatHistory
|
||||
from semantic_kernel.contents.chat_message_content import CMC_ITEM_TYPES, ChatMessageContent
|
||||
from semantic_kernel.contents.function_call_content import FunctionCallContent
|
||||
from semantic_kernel.contents.streaming_chat_message_content import STREAMING_CMC_ITEM_TYPES as STREAMING_ITEM_TYPES
|
||||
from semantic_kernel.contents.streaming_chat_message_content import StreamingChatMessageContent
|
||||
from semantic_kernel.contents.streaming_text_content import StreamingTextContent
|
||||
from semantic_kernel.contents.text_content import TextContent
|
||||
from semantic_kernel.exceptions.service_exceptions import (
|
||||
ServiceInitializationError,
|
||||
ServiceInvalidExecutionSettingsError,
|
||||
ServiceInvalidResponseError,
|
||||
)
|
||||
from semantic_kernel.utils.telemetry.model_diagnostics.decorators import (
|
||||
trace_chat_completion,
|
||||
trace_streaming_chat_completion,
|
||||
)
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from semantic_kernel.connectors.ai.function_call_choice_configuration import FunctionCallChoiceConfiguration
|
||||
from semantic_kernel.connectors.ai.prompt_execution_settings import PromptExecutionSettings
|
||||
|
||||
CMC_TYPE = TypeVar("CMC_TYPE", bound=ChatMessageContent)
|
||||
|
||||
logger: logging.Logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
class OllamaChatCompletion(OllamaBase, ChatCompletionClientBase):
|
||||
"""Initializes a new instance of the OllamaChatCompletion class.
|
||||
|
||||
Make sure to have the ollama service running either locally or remotely.
|
||||
"""
|
||||
|
||||
SUPPORTS_FUNCTION_CALLING: ClassVar[bool] = True
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
service_id: str | None = None,
|
||||
ai_model_id: str | None = None,
|
||||
host: str | None = None,
|
||||
client: AsyncClient | None = None,
|
||||
env_file_path: str | None = None,
|
||||
env_file_encoding: str | None = None,
|
||||
) -> None:
|
||||
"""Initialize an OllamaChatCompletion service.
|
||||
|
||||
Args:
|
||||
service_id (Optional[str]): Service ID tied to the execution settings. (Optional)
|
||||
ai_model_id (Optional[str]): The model name. (Optional)
|
||||
host (Optional[str]): URL of the Ollama server, defaults to None and
|
||||
will use the default Ollama service address: http://127.0.0.1:11434. (Optional)
|
||||
client (Optional[AsyncClient]): A custom Ollama client to use for the service. (Optional)
|
||||
env_file_path (str | None): Use the environment settings file as a fallback to using env vars.
|
||||
env_file_encoding (str | None): The encoding of the environment settings file, defaults to 'utf-8'.
|
||||
"""
|
||||
try:
|
||||
ollama_settings = OllamaSettings(
|
||||
chat_model_id=ai_model_id,
|
||||
host=host,
|
||||
env_file_path=env_file_path,
|
||||
env_file_encoding=env_file_encoding,
|
||||
)
|
||||
except ValidationError as ex:
|
||||
raise ServiceInitializationError("Failed to create Ollama settings.", ex) from ex
|
||||
|
||||
if not ollama_settings.chat_model_id:
|
||||
raise ServiceInitializationError("Ollama chat model ID is required.")
|
||||
|
||||
super().__init__(
|
||||
service_id=service_id or ollama_settings.chat_model_id,
|
||||
ai_model_id=ollama_settings.chat_model_id,
|
||||
client=client or AsyncClient(host=ollama_settings.host),
|
||||
)
|
||||
|
||||
# region Overriding base class methods
|
||||
|
||||
# Override from AIServiceClientBase
|
||||
@override
|
||||
def get_prompt_execution_settings_class(self) -> type["PromptExecutionSettings"]:
|
||||
"""Get the request settings class."""
|
||||
return OllamaChatPromptExecutionSettings
|
||||
|
||||
# Override from AIServiceClientBase
|
||||
@override
|
||||
def service_url(self) -> str | None:
|
||||
if hasattr(self.client, "_client") and isinstance(self.client._client, httpx.AsyncClient):
|
||||
# Best effort to get the endpoint
|
||||
return str(self.client._client.base_url)
|
||||
return None
|
||||
|
||||
@override
|
||||
def _prepare_chat_history_for_request(
|
||||
self,
|
||||
chat_history: ChatHistory,
|
||||
role_key: str = "role",
|
||||
content_key: str = "content",
|
||||
) -> list[Message]:
|
||||
return [MESSAGE_CONVERTERS[message.role](message) for message in chat_history.messages]
|
||||
|
||||
@override
|
||||
def _verify_function_choice_settings(self, settings: "PromptExecutionSettings") -> None:
|
||||
if settings.function_choice_behavior and settings.function_choice_behavior.type_ in [
|
||||
FunctionChoiceType.REQUIRED,
|
||||
FunctionChoiceType.NONE,
|
||||
]:
|
||||
raise ServiceInvalidExecutionSettingsError(
|
||||
"Ollama does not support function choice behavior of type 'required' or 'none' yet."
|
||||
)
|
||||
|
||||
@override
|
||||
def _update_function_choice_settings_callback(
|
||||
self,
|
||||
) -> Callable[["FunctionCallChoiceConfiguration", "PromptExecutionSettings", FunctionChoiceType], None]:
|
||||
return update_settings_from_function_choice_configuration
|
||||
|
||||
@override
|
||||
def _reset_function_choice_settings(self, settings: "PromptExecutionSettings") -> None:
|
||||
if hasattr(settings, "tools"):
|
||||
settings.tools = None
|
||||
|
||||
@override
|
||||
@trace_chat_completion(OllamaBase.MODEL_PROVIDER_NAME)
|
||||
async def _inner_get_chat_message_contents(
|
||||
self,
|
||||
chat_history: "ChatHistory",
|
||||
settings: "PromptExecutionSettings",
|
||||
) -> list["ChatMessageContent"]:
|
||||
if not isinstance(settings, OllamaChatPromptExecutionSettings):
|
||||
settings = self.get_prompt_execution_settings_from_settings(settings)
|
||||
assert isinstance(settings, OllamaChatPromptExecutionSettings) # nosec
|
||||
|
||||
prepared_chat_history = self._prepare_chat_history_for_request(chat_history)
|
||||
|
||||
response_object = await self.client.chat(
|
||||
model=self.ai_model_id,
|
||||
messages=prepared_chat_history,
|
||||
stream=False,
|
||||
**settings.prepare_settings_dict(),
|
||||
)
|
||||
|
||||
if isinstance(response_object, ChatResponse):
|
||||
return [self._create_chat_message_content_from_chat_response(response_object)]
|
||||
if isinstance(response_object, Mapping):
|
||||
return [self._create_chat_message_content(response_object)]
|
||||
raise ServiceInvalidResponseError(
|
||||
"Invalid response type from Ollama chat completion. "
|
||||
f"Expected Mapping or ChatResponse but got {type(response_object)}."
|
||||
)
|
||||
|
||||
@override
|
||||
@trace_streaming_chat_completion(OllamaBase.MODEL_PROVIDER_NAME)
|
||||
async def _inner_get_streaming_chat_message_contents(
|
||||
self,
|
||||
chat_history: "ChatHistory",
|
||||
settings: "PromptExecutionSettings",
|
||||
function_invoke_attempt: int = 0,
|
||||
) -> AsyncGenerator[list["StreamingChatMessageContent"], Any]:
|
||||
if not isinstance(settings, OllamaChatPromptExecutionSettings):
|
||||
settings = self.get_prompt_execution_settings_from_settings(settings)
|
||||
assert isinstance(settings, OllamaChatPromptExecutionSettings) # nosec
|
||||
|
||||
prepared_chat_history = self._prepare_chat_history_for_request(chat_history)
|
||||
|
||||
response_object = await self.client.chat(
|
||||
model=self.ai_model_id,
|
||||
messages=prepared_chat_history,
|
||||
stream=True,
|
||||
**settings.prepare_settings_dict(),
|
||||
)
|
||||
|
||||
if not isinstance(response_object, AsyncIterator):
|
||||
raise ServiceInvalidResponseError(
|
||||
"Invalid response type from Ollama streaming chat completion. "
|
||||
f"Expected AsyncIterator but got {type(response_object)}."
|
||||
)
|
||||
|
||||
async for part in response_object:
|
||||
if isinstance(part, ChatResponse):
|
||||
yield [self._create_streaming_chat_message_content_from_chat_response(part, function_invoke_attempt)]
|
||||
continue
|
||||
if isinstance(part, Mapping):
|
||||
yield [self._create_streaming_chat_message_content(part, function_invoke_attempt)]
|
||||
continue
|
||||
raise ServiceInvalidResponseError(
|
||||
"Invalid response type from Ollama streaming chat completion. "
|
||||
f"Expected mapping or ChatResponse but got {type(part)}."
|
||||
)
|
||||
|
||||
# endregion
|
||||
|
||||
def _create_streaming_chat_message_content_from_chat_response(
|
||||
self,
|
||||
response: ChatResponse,
|
||||
function_invoke_attempt: int,
|
||||
) -> StreamingChatMessageContent:
|
||||
"""Create a chat message content from the response."""
|
||||
items: list[STREAMING_ITEM_TYPES] = []
|
||||
if response.message.content:
|
||||
items.append(
|
||||
StreamingTextContent(
|
||||
choice_index=0,
|
||||
text=response.message.content,
|
||||
inner_content=response.message,
|
||||
)
|
||||
)
|
||||
self._parse_tool_calls(response.message.tool_calls, items)
|
||||
return StreamingChatMessageContent(
|
||||
choice_index=0,
|
||||
role=AuthorRole.ASSISTANT,
|
||||
items=items,
|
||||
inner_content=response,
|
||||
ai_model_id=self.ai_model_id,
|
||||
metadata=self._get_metadata_from_chat_response(response),
|
||||
function_invoke_attempt=function_invoke_attempt,
|
||||
)
|
||||
|
||||
def _parse_tool_calls(self, tool_calls: Sequence[Message.ToolCall] | None, items: list[Any]):
|
||||
if tool_calls:
|
||||
for tool_call in tool_calls:
|
||||
items.append(
|
||||
FunctionCallContent(
|
||||
inner_content=tool_call,
|
||||
ai_model_id=self.ai_model_id,
|
||||
name=tool_call.function.name,
|
||||
arguments=tool_call.function.arguments,
|
||||
)
|
||||
)
|
||||
|
||||
def _create_chat_message_content_from_chat_response(self, response: ChatResponse) -> ChatMessageContent:
|
||||
"""Create a chat message content from the response."""
|
||||
items: list[CMC_ITEM_TYPES] = []
|
||||
if response.message.content:
|
||||
items.append(
|
||||
TextContent(
|
||||
text=response.message.content,
|
||||
inner_content=response.message,
|
||||
)
|
||||
)
|
||||
self._parse_tool_calls(response.message.tool_calls, items)
|
||||
return ChatMessageContent(
|
||||
role=AuthorRole.ASSISTANT,
|
||||
items=items,
|
||||
inner_content=response,
|
||||
ai_model_id=self.ai_model_id,
|
||||
metadata=self._get_metadata_from_chat_response(response),
|
||||
)
|
||||
|
||||
def _create_chat_message_content(self, response: Mapping[str, Any]) -> ChatMessageContent:
|
||||
"""Create a chat message content from the response."""
|
||||
items: list[CMC_ITEM_TYPES] = []
|
||||
if not (message := response.get("message", None)):
|
||||
raise ServiceInvalidResponseError("No message content found in response.")
|
||||
|
||||
if content := message.get("content", None):
|
||||
items.append(
|
||||
TextContent(
|
||||
text=content,
|
||||
inner_content=message,
|
||||
)
|
||||
)
|
||||
if tool_calls := message.get("tool_calls", None):
|
||||
for tool_call in tool_calls:
|
||||
items.append(
|
||||
FunctionCallContent(
|
||||
inner_content=tool_call,
|
||||
ai_model_id=self.ai_model_id,
|
||||
name=tool_call.get("function").get("name"),
|
||||
arguments=tool_call.get("function").get("arguments"),
|
||||
)
|
||||
)
|
||||
|
||||
return ChatMessageContent(
|
||||
role=AuthorRole.ASSISTANT,
|
||||
items=items,
|
||||
inner_content=response,
|
||||
metadata=self._get_metadata_from_response(response),
|
||||
)
|
||||
|
||||
def _create_streaming_chat_message_content(
|
||||
self, part: Mapping[str, Any], function_invoke_attempt: int
|
||||
) -> StreamingChatMessageContent:
|
||||
"""Create a streaming chat message content from the response part."""
|
||||
items: list[STREAMING_ITEM_TYPES] = []
|
||||
if not (message := part.get("message", None)):
|
||||
raise ServiceInvalidResponseError("No message content found in response part.")
|
||||
|
||||
if content := message.get("content", None):
|
||||
items.append(
|
||||
StreamingTextContent(
|
||||
choice_index=0,
|
||||
text=content,
|
||||
inner_content=message,
|
||||
)
|
||||
)
|
||||
if tool_calls := message.get("tool_calls", None):
|
||||
for tool_call in tool_calls:
|
||||
items.append(
|
||||
FunctionCallContent(
|
||||
inner_content=tool_call,
|
||||
ai_model_id=self.ai_model_id,
|
||||
name=tool_call.get("function").get("name"),
|
||||
arguments=tool_call.get("function").get("arguments"),
|
||||
)
|
||||
)
|
||||
|
||||
return StreamingChatMessageContent(
|
||||
role=AuthorRole.ASSISTANT,
|
||||
choice_index=0,
|
||||
items=items,
|
||||
inner_content=part,
|
||||
ai_model_id=self.ai_model_id,
|
||||
metadata=self._get_metadata_from_response(part),
|
||||
function_invoke_attempt=function_invoke_attempt,
|
||||
)
|
||||
|
||||
def _get_metadata_from_response(self, response: Mapping[str, Any]) -> dict[str, Any]:
|
||||
"""Get metadata from the response."""
|
||||
metadata = {
|
||||
"model": response.get("model"),
|
||||
}
|
||||
|
||||
if "prompt_eval_count" in response and "eval_count" in response:
|
||||
metadata["usage"] = CompletionUsage(
|
||||
prompt_tokens=response.get("prompt_eval_count"),
|
||||
completion_tokens=response.get("eval_count"),
|
||||
)
|
||||
|
||||
return metadata
|
||||
|
||||
def _get_metadata_from_chat_response(self, response: ChatResponse) -> dict[str, Any]:
|
||||
"""Get metadata from the response."""
|
||||
metadata: dict[str, Any] = {
|
||||
"model": response.model,
|
||||
}
|
||||
if response.prompt_eval_count and response.eval_count:
|
||||
metadata["usage"] = CompletionUsage(
|
||||
prompt_tokens=response.prompt_eval_count,
|
||||
completion_tokens=response.eval_count,
|
||||
)
|
||||
return metadata
|
||||
@@ -0,0 +1,163 @@
|
||||
# Copyright (c) Microsoft. All rights reserved.
|
||||
|
||||
import logging
|
||||
import sys
|
||||
from collections.abc import AsyncGenerator, AsyncIterator, Mapping
|
||||
from typing import TYPE_CHECKING, Any
|
||||
|
||||
if sys.version_info >= (3, 12):
|
||||
from typing import override # pragma: no cover
|
||||
else:
|
||||
from typing_extensions import override # pragma: no cover
|
||||
|
||||
import httpx
|
||||
from ollama import AsyncClient
|
||||
from ollama._types import GenerateResponse
|
||||
from pydantic import ValidationError
|
||||
|
||||
from semantic_kernel.connectors.ai.ollama.ollama_prompt_execution_settings import OllamaTextPromptExecutionSettings
|
||||
from semantic_kernel.connectors.ai.ollama.ollama_settings import OllamaSettings
|
||||
from semantic_kernel.connectors.ai.ollama.services.ollama_base import OllamaBase
|
||||
from semantic_kernel.connectors.ai.text_completion_client_base import TextCompletionClientBase
|
||||
from semantic_kernel.contents.streaming_text_content import StreamingTextContent
|
||||
from semantic_kernel.contents.text_content import TextContent
|
||||
from semantic_kernel.exceptions.service_exceptions import ServiceInitializationError, ServiceInvalidResponseError
|
||||
from semantic_kernel.utils.telemetry.model_diagnostics.decorators import (
|
||||
trace_streaming_text_completion,
|
||||
trace_text_completion,
|
||||
)
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from semantic_kernel.connectors.ai.prompt_execution_settings import PromptExecutionSettings
|
||||
|
||||
logger: logging.Logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
class OllamaTextCompletion(OllamaBase, TextCompletionClientBase):
|
||||
"""Initializes a new instance of the OllamaTextCompletion class.
|
||||
|
||||
Make sure to have the ollama service running either locally or remotely.
|
||||
"""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
service_id: str | None = None,
|
||||
ai_model_id: str | None = None,
|
||||
host: str | None = None,
|
||||
client: AsyncClient | None = None,
|
||||
env_file_path: str | None = None,
|
||||
env_file_encoding: str | None = None,
|
||||
) -> None:
|
||||
"""Initialize an OllamaChatCompletion service.
|
||||
|
||||
Args:
|
||||
service_id (Optional[str]): Service ID tied to the execution settings. (Optional)
|
||||
ai_model_id (Optional[str]): The model name. (Optional)
|
||||
host (Optional[str]): URL of the Ollama server, defaults to None and
|
||||
will use the default Ollama service address: http://127.0.0.1:11434. (Optional)
|
||||
client (Optional[AsyncClient]): A custom Ollama client to use for the service. (Optional)
|
||||
env_file_path (str | None): Use the environment settings file as a fallback to using env vars.
|
||||
env_file_encoding (str | None): The encoding of the environment settings file, defaults to 'utf-8'.
|
||||
"""
|
||||
try:
|
||||
ollama_settings = OllamaSettings(
|
||||
text_model_id=ai_model_id,
|
||||
host=host,
|
||||
env_file_path=env_file_path,
|
||||
env_file_encoding=env_file_encoding,
|
||||
)
|
||||
except ValidationError as ex:
|
||||
raise ServiceInitializationError("Failed to create Ollama settings.", ex) from ex
|
||||
|
||||
if not ollama_settings.text_model_id:
|
||||
raise ServiceInitializationError("Ollama text model ID is required.")
|
||||
|
||||
super().__init__(
|
||||
service_id=service_id or ollama_settings.text_model_id,
|
||||
ai_model_id=ollama_settings.text_model_id,
|
||||
client=client or AsyncClient(host=ollama_settings.host),
|
||||
)
|
||||
|
||||
# region Overriding base class methods
|
||||
|
||||
# Override from AIServiceClientBase
|
||||
@override
|
||||
def get_prompt_execution_settings_class(self) -> type["PromptExecutionSettings"]:
|
||||
return OllamaTextPromptExecutionSettings
|
||||
|
||||
# Override from AIServiceClientBase
|
||||
@override
|
||||
def service_url(self) -> str | None:
|
||||
if hasattr(self.client, "_client") and isinstance(self.client._client, httpx.AsyncClient):
|
||||
# Best effort to get the endpoint
|
||||
return str(self.client._client.base_url)
|
||||
return None
|
||||
|
||||
@override
|
||||
@trace_text_completion(OllamaBase.MODEL_PROVIDER_NAME)
|
||||
async def _inner_get_text_contents(
|
||||
self,
|
||||
prompt: str,
|
||||
settings: "PromptExecutionSettings",
|
||||
) -> list[TextContent]:
|
||||
if not isinstance(settings, OllamaTextPromptExecutionSettings):
|
||||
settings = self.get_prompt_execution_settings_from_settings(settings)
|
||||
assert isinstance(settings, OllamaTextPromptExecutionSettings) # nosec
|
||||
|
||||
response_object = await self.client.generate(
|
||||
model=self.ai_model_id,
|
||||
prompt=prompt,
|
||||
stream=False,
|
||||
**settings.prepare_settings_dict(),
|
||||
)
|
||||
|
||||
if not isinstance(response_object, (Mapping, GenerateResponse)):
|
||||
raise ServiceInvalidResponseError(
|
||||
"Invalid response type from Ollama chat completion. "
|
||||
f"Expected Mapping or GenerateResponse but got {type(response_object)}."
|
||||
)
|
||||
return [
|
||||
TextContent(
|
||||
inner_content=response_object,
|
||||
ai_model_id=self.ai_model_id,
|
||||
text=response_object.response
|
||||
if isinstance(response_object, GenerateResponse)
|
||||
else response_object["response"],
|
||||
)
|
||||
]
|
||||
|
||||
@override
|
||||
@trace_streaming_text_completion(OllamaBase.MODEL_PROVIDER_NAME)
|
||||
async def _inner_get_streaming_text_contents(
|
||||
self,
|
||||
prompt: str,
|
||||
settings: "PromptExecutionSettings",
|
||||
) -> AsyncGenerator[list[StreamingTextContent], Any]:
|
||||
if not isinstance(settings, OllamaTextPromptExecutionSettings):
|
||||
settings = self.get_prompt_execution_settings_from_settings(settings)
|
||||
assert isinstance(settings, OllamaTextPromptExecutionSettings) # nosec
|
||||
|
||||
response_object = await self.client.generate(
|
||||
model=self.ai_model_id,
|
||||
prompt=prompt,
|
||||
stream=True,
|
||||
**settings.prepare_settings_dict(),
|
||||
)
|
||||
|
||||
if not isinstance(response_object, AsyncIterator):
|
||||
raise ServiceInvalidResponseError(
|
||||
"Invalid response type from Ollama chat completion. "
|
||||
f"Expected AsyncIterator but got {type(response_object)}."
|
||||
)
|
||||
|
||||
async for part in response_object:
|
||||
yield [
|
||||
StreamingTextContent(
|
||||
choice_index=0,
|
||||
inner_content=part,
|
||||
ai_model_id=self.ai_model_id,
|
||||
text=part.response if isinstance(part, GenerateResponse) else part.get("response"),
|
||||
)
|
||||
]
|
||||
|
||||
# endregion
|
||||
@@ -0,0 +1,112 @@
|
||||
# Copyright (c) Microsoft. All rights reserved.
|
||||
|
||||
import logging
|
||||
import sys
|
||||
from typing import TYPE_CHECKING, Any
|
||||
|
||||
from ollama import AsyncClient
|
||||
from pydantic import ValidationError
|
||||
|
||||
from semantic_kernel.connectors.ai.ollama.ollama_prompt_execution_settings import OllamaEmbeddingPromptExecutionSettings
|
||||
from semantic_kernel.connectors.ai.ollama.ollama_settings import OllamaSettings
|
||||
from semantic_kernel.connectors.ai.ollama.services.ollama_base import OllamaBase
|
||||
from semantic_kernel.exceptions.service_exceptions import ServiceInitializationError
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from semantic_kernel.connectors.ai.prompt_execution_settings import PromptExecutionSettings
|
||||
|
||||
if sys.version_info >= (3, 12):
|
||||
from typing import override # pragma: no cover
|
||||
else:
|
||||
from typing_extensions import override # pragma: no cover
|
||||
|
||||
from numpy import array, ndarray
|
||||
|
||||
from semantic_kernel.connectors.ai.embedding_generator_base import EmbeddingGeneratorBase
|
||||
from semantic_kernel.utils.feature_stage_decorator import experimental
|
||||
|
||||
logger: logging.Logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
@experimental
|
||||
class OllamaTextEmbedding(OllamaBase, EmbeddingGeneratorBase):
|
||||
"""Ollama embeddings client.
|
||||
|
||||
Make sure to have the ollama service running either locally or remotely.
|
||||
"""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
service_id: str | None = None,
|
||||
ai_model_id: str | None = None,
|
||||
host: str | None = None,
|
||||
client: AsyncClient | None = None,
|
||||
env_file_path: str | None = None,
|
||||
env_file_encoding: str | None = None,
|
||||
) -> None:
|
||||
"""Initialize an OllamaChatCompletion service.
|
||||
|
||||
Args:
|
||||
service_id (Optional[str]): Service ID tied to the execution settings. (Optional)
|
||||
ai_model_id (Optional[str]): The model name. (Optional)
|
||||
host (Optional[str]): URL of the Ollama server, defaults to None and
|
||||
will use the default Ollama service address: http://127.0.0.1:11434. (Optional)
|
||||
client (Optional[AsyncClient]): A custom Ollama client to use for the service. (Optional)
|
||||
env_file_path (str | None): Use the environment settings file as a fallback to using env vars.
|
||||
env_file_encoding (str | None): The encoding of the environment settings file, defaults to 'utf-8'.
|
||||
"""
|
||||
try:
|
||||
ollama_settings = OllamaSettings(
|
||||
embedding_model_id=ai_model_id,
|
||||
host=host,
|
||||
env_file_path=env_file_path,
|
||||
env_file_encoding=env_file_encoding,
|
||||
)
|
||||
except ValidationError as ex:
|
||||
raise ServiceInitializationError("Failed to create Ollama settings.", ex) from ex
|
||||
|
||||
if not ollama_settings.embedding_model_id:
|
||||
raise ServiceInitializationError("Ollama embedding model ID is not set.")
|
||||
|
||||
super().__init__(
|
||||
service_id=service_id or ollama_settings.embedding_model_id,
|
||||
ai_model_id=ollama_settings.embedding_model_id,
|
||||
client=client or AsyncClient(host=ollama_settings.host),
|
||||
)
|
||||
|
||||
@override
|
||||
async def generate_embeddings(
|
||||
self,
|
||||
texts: list[str],
|
||||
settings: "PromptExecutionSettings | None" = None,
|
||||
**kwargs: Any,
|
||||
) -> ndarray:
|
||||
raw_embeddings = await self.generate_raw_embeddings(texts, settings, **kwargs)
|
||||
return array(raw_embeddings)
|
||||
|
||||
@override
|
||||
async def generate_raw_embeddings(
|
||||
self,
|
||||
texts: list[str],
|
||||
settings: "PromptExecutionSettings | None" = None,
|
||||
**kwargs: Any,
|
||||
) -> Any:
|
||||
if not settings:
|
||||
settings = OllamaEmbeddingPromptExecutionSettings()
|
||||
else:
|
||||
settings = self.get_prompt_execution_settings_from_settings(settings)
|
||||
|
||||
result = []
|
||||
for text in texts:
|
||||
response_object = await self.client.embeddings(
|
||||
model=self.ai_model_id,
|
||||
prompt=text,
|
||||
**settings.prepare_settings_dict(),
|
||||
)
|
||||
result.append(response_object["embedding"])
|
||||
|
||||
return result
|
||||
|
||||
@override
|
||||
def get_prompt_execution_settings_class(self) -> type["PromptExecutionSettings"]:
|
||||
return OllamaEmbeddingPromptExecutionSettings
|
||||
@@ -0,0 +1,133 @@
|
||||
# Copyright (c) Microsoft. All rights reserved.
|
||||
|
||||
import json
|
||||
from collections.abc import Callable, Mapping
|
||||
from typing import TYPE_CHECKING
|
||||
|
||||
from ollama._types import Message
|
||||
|
||||
from semantic_kernel.connectors.ai.function_calling_utils import kernel_function_metadata_to_function_call_format
|
||||
from semantic_kernel.connectors.ai.function_choice_behavior import FunctionChoiceType
|
||||
from semantic_kernel.contents.chat_message_content import ChatMessageContent
|
||||
from semantic_kernel.contents.function_call_content import FunctionCallContent
|
||||
from semantic_kernel.contents.function_result_content import FunctionResultContent
|
||||
from semantic_kernel.contents.image_content import ImageContent
|
||||
from semantic_kernel.contents.utils.author_role import AuthorRole
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from semantic_kernel.connectors.ai.function_call_choice_configuration import FunctionCallChoiceConfiguration
|
||||
from semantic_kernel.connectors.ai.prompt_execution_settings import PromptExecutionSettings
|
||||
|
||||
|
||||
def _format_system_message(message: ChatMessageContent) -> Message:
|
||||
"""Format a system message to the expected object for the client.
|
||||
|
||||
Args:
|
||||
message: The system message.
|
||||
|
||||
Returns:
|
||||
The formatted system message.
|
||||
"""
|
||||
return Message(role="system", content=message.content)
|
||||
|
||||
|
||||
def _format_user_message(message: ChatMessageContent) -> Message:
|
||||
"""Format a user message to the expected object for the client.
|
||||
|
||||
Args:
|
||||
message: The user message.
|
||||
|
||||
Returns:
|
||||
The formatted user message.
|
||||
"""
|
||||
if not any(isinstance(item, (ImageContent)) for item in message.items):
|
||||
return Message(role="user", content=message.content)
|
||||
|
||||
user_message = Message(role="user", content=message.content)
|
||||
|
||||
image_items = [item for item in message.items if isinstance(item, ImageContent)]
|
||||
if image_items:
|
||||
if any(not image_item.data for image_item in image_items):
|
||||
raise ValueError("Image item must contain data encoded as base64.")
|
||||
user_message["images"] = [image_item.data for image_item in image_items]
|
||||
|
||||
return user_message
|
||||
|
||||
|
||||
def _format_assistant_message(message: ChatMessageContent) -> Message:
|
||||
"""Format an assistant message to the expected object for the client.
|
||||
|
||||
Args:
|
||||
message: The assistant message.
|
||||
|
||||
Returns:
|
||||
The formatted assistant message.
|
||||
"""
|
||||
assistant_message = Message(role="assistant", content=message.content)
|
||||
|
||||
image_items = [item for item in message.items if isinstance(item, ImageContent)]
|
||||
if image_items:
|
||||
if any(image_item.data is None for image_item in image_items):
|
||||
raise ValueError("Image must be encoded as base64.")
|
||||
assistant_message["images"] = [image_item.data for image_item in image_items]
|
||||
|
||||
tool_calls = [item for item in message.items if isinstance(item, FunctionCallContent)]
|
||||
if tool_calls:
|
||||
assistant_message["tool_calls"] = [
|
||||
{
|
||||
"function": {
|
||||
"name": tool_call.function_name,
|
||||
"arguments": tool_call.arguments
|
||||
if isinstance(tool_call.arguments, Mapping)
|
||||
else json.loads(tool_call.arguments or "{}"),
|
||||
}
|
||||
}
|
||||
for tool_call in tool_calls
|
||||
]
|
||||
|
||||
return assistant_message
|
||||
|
||||
|
||||
def _format_tool_message(message: ChatMessageContent) -> Message:
|
||||
"""Format a tool message to the expected object for the client.
|
||||
|
||||
Args:
|
||||
message: The tool message.
|
||||
|
||||
Returns:
|
||||
The formatted tool message.
|
||||
"""
|
||||
function_result_items = [item for item in message.items if isinstance(item, FunctionResultContent)]
|
||||
if not function_result_items:
|
||||
raise ValueError("Tool message must have a function result content item.")
|
||||
|
||||
return Message(role="tool", content=str(function_result_items[0].result))
|
||||
|
||||
|
||||
MESSAGE_CONVERTERS: dict[AuthorRole, Callable[[ChatMessageContent], Message]] = {
|
||||
AuthorRole.SYSTEM: _format_system_message,
|
||||
AuthorRole.USER: _format_user_message,
|
||||
AuthorRole.ASSISTANT: _format_assistant_message,
|
||||
AuthorRole.TOOL: _format_tool_message,
|
||||
}
|
||||
|
||||
|
||||
def update_settings_from_function_choice_configuration(
|
||||
function_choice_configuration: "FunctionCallChoiceConfiguration",
|
||||
settings: "PromptExecutionSettings",
|
||||
type: FunctionChoiceType,
|
||||
) -> None:
|
||||
"""Update the settings from a FunctionChoiceConfiguration.
|
||||
|
||||
Since this function might be called before the settings are cast to Ollama Settings
|
||||
We need to try to use the tools attribute or fallback to the extension_data attribute.
|
||||
"""
|
||||
if function_choice_configuration.available_functions:
|
||||
tools = [
|
||||
kernel_function_metadata_to_function_call_format(f)
|
||||
for f in function_choice_configuration.available_functions
|
||||
]
|
||||
try:
|
||||
settings.tools = tools # type: ignore
|
||||
except Exception:
|
||||
settings.extension_data["tools"] = tools
|
||||
Reference in New Issue
Block a user