from __future__ import annotations
from abc import ABC, abstractmethod
from typing import Any, Dict, Iterator, List, Literal, Union, cast, overload
from ai21.models import ChatMessage as J2ChatMessage
from ai21.models import RoleType
from ai21.models.chat import ChatCompletionChunk, ChatMessage
from ai21.stream.stream import Stream as AI21Stream
from langchain_core.messages import (
AIMessage,
AIMessageChunk,
BaseMessage,
BaseMessageChunk,
HumanMessage,
)
from langchain_core.messages.ai import UsageMetadata
from langchain_core.outputs import ChatGenerationChunk
_ChatMessageTypes = Union[ChatMessage, J2ChatMessage]
_SYSTEM_ERR_MESSAGE = "System message must be at beginning of message list."
_ROLE_TYPE = Union[str, RoleType]
[docs]
class ChatAdapter(ABC):
"""Common interface for the different Chat models available in AI21.
It converts LangChain messages to AI21 messages.
Calls the appropriate AI21 model API with the converted messages.
"""
[docs]
@abstractmethod
def convert_messages(
self,
messages: List[BaseMessage],
) -> Dict[str, Any]:
pass
def _convert_message_to_ai21_message(
self,
message: BaseMessage,
) -> _ChatMessageTypes:
content = cast(str, message.content)
role = self._parse_role(message)
return self._chat_message(role=role, content=content)
def _parse_role(self, message: BaseMessage) -> _ROLE_TYPE:
role = None
if isinstance(message, HumanMessage):
return RoleType.USER
if isinstance(message, AIMessage):
return RoleType.ASSISTANT
if isinstance(self, J2ChatAdapter):
if not role:
raise ValueError(
f"Could not resolve role type from message {message}. "
f"Only support {HumanMessage.__name__} and {AIMessage.__name__}."
)
# if it gets here, we rely on the server to handle the role type
return message.type
@abstractmethod
def _chat_message(
self,
role: _ROLE_TYPE,
content: str,
) -> _ChatMessageTypes:
pass
@overload
def call(
self,
client: Any,
stream: Literal[True],
**params: Any,
) -> Iterator[ChatGenerationChunk]:
pass
@overload
def call(
self,
client: Any,
stream: Literal[False],
**params: Any,
) -> List[BaseMessage]:
pass
[docs]
@abstractmethod
def call(
self,
client: Any,
stream: Literal[True] | Literal[False],
**params: Any,
) -> List[BaseMessage] | Iterator[ChatGenerationChunk]:
pass
def _get_system_message_from_message(self, message: BaseMessage) -> str:
if not isinstance(message.content, str):
raise ValueError(
f"System Message must be of type str. Got {type(message.content)}"
)
return message.content
[docs]
class J2ChatAdapter(ChatAdapter):
"""Adapter for J2Chat models."""
[docs]
def convert_messages(self, messages: List[BaseMessage]) -> Dict[str, Any]:
system_message = ""
converted_messages = [] # type: ignore
for i, message in enumerate(messages):
if message.type == "system":
if i != 0:
raise ValueError(_SYSTEM_ERR_MESSAGE)
else:
system_message = self._get_system_message_from_message(message)
else:
converted_message = self._convert_message_to_ai21_message(message)
converted_messages.append(converted_message)
return {"system": system_message, "messages": converted_messages}
def _chat_message(
self,
role: _ROLE_TYPE,
content: str,
) -> J2ChatMessage:
return J2ChatMessage(role=RoleType(role), text=content)
@overload
def call(
self,
client: Any,
stream: Literal[True],
**params: Any,
) -> Iterator[ChatGenerationChunk]: ...
@overload
def call(
self,
client: Any,
stream: Literal[False],
**params: Any,
) -> List[BaseMessage]: ...
[docs]
def call(
self,
client: Any,
stream: Literal[True] | Literal[False],
**params: Any,
) -> List[BaseMessage] | Iterator[ChatGenerationChunk]:
if stream:
raise NotImplementedError("Streaming is not supported for Jurassic models.")
response = client.chat.create(**params)
return [AIMessage(output.text) for output in response.outputs]
[docs]
class JambaChatCompletionsAdapter(ChatAdapter):
"""Adapter for Jamba Chat Completions."""
[docs]
def convert_messages(self, messages: List[BaseMessage]) -> Dict[str, Any]:
return {
"messages": [
self._convert_message_to_ai21_message(message) for message in messages
],
}
def _chat_message(
self,
role: _ROLE_TYPE,
content: str,
) -> ChatMessage:
return ChatMessage(
role=role.value if isinstance(role, RoleType) else role,
content=content,
)
@overload
def call(
self,
client: Any,
stream: Literal[True],
**params: Any,
) -> Iterator[ChatGenerationChunk]: ...
@overload
def call(
self,
client: Any,
stream: Literal[False],
**params: Any,
) -> List[BaseMessage]: ...
[docs]
def call(
self,
client: Any,
stream: Literal[True] | Literal[False],
**params: Any,
) -> List[BaseMessage] | Iterator[ChatGenerationChunk]:
response = client.chat.completions.create(stream=stream, **params)
if stream:
return self._stream_response(response)
return [AIMessage(choice.message.content) for choice in response.choices]
def _stream_response(
self,
response: AI21Stream[ChatCompletionChunk],
) -> Iterator[ChatGenerationChunk]:
for chunk in response:
converted_message = self._convert_ai21_chunk_to_chunk(chunk)
yield ChatGenerationChunk(message=converted_message)
def _convert_ai21_chunk_to_chunk(
self,
chunk: ChatCompletionChunk,
) -> BaseMessageChunk:
usage = chunk.usage
content = chunk.choices[0].delta.content or ""
if usage is None:
return AIMessageChunk(
content=content,
)
return AIMessageChunk(
content=content,
usage_metadata=UsageMetadata(
input_tokens=usage.prompt_tokens,
output_tokens=usage.completion_tokens,
total_tokens=usage.total_tokens,
),
)
