# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
from __future__ import annotations
import json
import logging
import re
import uuid
from collections import defaultdict
from typing import (
TYPE_CHECKING,
Any,
Callable,
Dict,
List,
Optional,
Tuple,
Type,
Union,
)
from openai.types.chat import ChatCompletionMessageToolCall
from openai.types.chat.chat_completion_message_tool_call import Function
from pydantic import BaseModel, ValidationError
from camel.agents.base import BaseAgent
from camel.memories import (
AgentMemory,
ChatHistoryMemory,
MemoryRecord,
ScoreBasedContextCreator,
)
from camel.messages import BaseMessage, FunctionCallingMessage, OpenAIMessage
from camel.models import (
BaseModelBackend,
ModelFactory,
ModelManager,
ModelProcessingError,
)
from camel.responses import ChatAgentResponse
from camel.types import (
ChatCompletion,
ChatCompletionChunk,
ModelPlatformType,
ModelType,
OpenAIBackendRole,
RoleType,
)
from camel.utils import (
func_string_to_callable,
generate_prompt_for_structured_output,
get_model_encoding,
get_pydantic_object_schema,
json_to_function_code,
)
if TYPE_CHECKING:
from openai import Stream
from camel.terminators import ResponseTerminator
from camel.toolkits import FunctionTool
logger = logging.getLogger(__name__)
# AgentOps decorator setting
try:
import os
if os.getenv("AGENTOPS_API_KEY") is not None:
from agentops import track_agent
else:
raise ImportError
except (ImportError, AttributeError):
from camel.utils import track_agent
[docs]
class FunctionCallingRecord(BaseModel):
r"""Historical records of functions called in the conversation.
Attributes:
func_name (str): The name of the function being called.
args (Dict[str, Any]): The dictionary of arguments passed to
the function.
result (Any): The execution result of calling this function.
"""
func_name: str
args: Dict[str, Any]
result: Any
def __str__(self) -> str:
r"""Overridden version of the string function.
Returns:
str: Modified string to represent the function calling.
"""
return (
f"Function Execution: {self.func_name}\n"
f"\tArgs: {self.args}\n"
f"\tResult: {self.result}"
)
[docs]
def as_dict(self) -> dict[str, Any]:
r"""Returns the function calling record as a dictionary.
Returns:
dict[str, Any]: The function calling record as a dictionary.
"""
return self.model_dump()
[docs]
@track_agent(name="ChatAgent")
class ChatAgent(BaseAgent):
r"""Class for managing conversations of CAMEL Chat Agents.
Args:
system_message (Union[BaseMessage, str], optional): The system message
for the chat agent.
model (BaseModelBackend, optional): The model backend to use for
generating responses. (default: :obj:`ModelPlatformType.DEFAULT`
with `ModelType.DEFAULT`)
memory (AgentMemory, optional): The agent memory for managing chat
messages. If `None`, a :obj:`ChatHistoryMemory` will be used.
(default: :obj:`None`)
message_window_size (int, optional): The maximum number of previous
messages to include in the context window. If `None`, no windowing
is performed. (default: :obj:`None`)
token_limit (int, optional): The maximum number of tokens in a context.
The context will be automatically pruned to fulfill the limitation.
If `None`, it will be set according to the backend model.
(default: :obj:`None`)
output_language (str, optional): The language to be output by the
agent. (default: :obj:`None`)
tools (Optional[List[Union[FunctionTool, Callable]]], optional): List
of available :obj:`FunctionTool` or :obj:`Callable`. (default:
:obj:`None`)
external_tools (Optional[List[Union[FunctionTool, Callable]]],
optional): List of external tools (:obj:`FunctionTool` or or
:obj:`Callable`) bind to one chat agent. When these tools are
called, the agent will directly return the request instead of
processing it. (default: :obj:`None`)
response_terminators (List[ResponseTerminator], optional): List of
:obj:`ResponseTerminator` bind to one chat agent.
(default: :obj:`None`)
scheduling_strategy (str): name of function that defines how to select
the next model in ModelManager. (default: :str:`round_robin`)
single_iteration (bool): Whether to let the agent perform only one
model calling at each step. (default: :obj:`False`)
"""
def __init__(
self,
system_message: Optional[Union[BaseMessage, str]] = None,
model: Optional[
Union[BaseModelBackend, List[BaseModelBackend]]
] = None,
memory: Optional[AgentMemory] = None,
message_window_size: Optional[int] = None,
token_limit: Optional[int] = None,
output_language: Optional[str] = None,
tools: Optional[List[Union[FunctionTool, Callable]]] = None,
external_tools: Optional[List[Union[FunctionTool, Callable]]] = None,
response_terminators: Optional[List[ResponseTerminator]] = None,
scheduling_strategy: str = "round_robin",
single_iteration: bool = False,
) -> None:
# Initialize the system message, converting string to BaseMessage if needed
if isinstance(system_message, str):
system_message = BaseMessage.make_assistant_message(
role_name='Assistant', content=system_message
)
self.orig_sys_message: Optional[BaseMessage] = system_message
self._system_message: Optional[BaseMessage] = system_message
self.role_name: str = (
getattr(system_message, 'role_name', None) or "assistant"
)
self.role_type: RoleType = (
getattr(system_message, 'role_type', None) or RoleType.ASSISTANT
)
self.model_backend = ModelManager(
model
if model is not None
else ModelFactory.create(
model_platform=ModelPlatformType.DEFAULT,
model_type=ModelType.DEFAULT,
),
scheduling_strategy=scheduling_strategy,
)
self.model_type = self.model_backend.model_type
# Initialize tools
self.tools: List[FunctionTool] = (
self._initialize_tools(tools) if tools else []
)
self.external_tools: List[FunctionTool] = (
self._initialize_tools(external_tools) if external_tools else []
)
self.external_tool_names: List[str] = [
tool.get_function_name() for tool in self.external_tools
]
self.all_tools = self.tools + self.external_tools or []
# Create tool dictionaries and configure backend tools if necessary
self.tool_dict = {
tool.get_function_name(): tool for tool in self.all_tools
}
# If the user set tools from `ChatAgent`, it will override the
# configured tools in `BaseModelBackend`.
if self.all_tools:
logger.warning(
"Overriding the configured tools in `BaseModelBackend` with the tools from `ChatAgent`."
)
tool_schema_list = [
tool.get_openai_tool_schema() for tool in self.all_tools
]
self.model_backend.model_config_dict['tools'] = tool_schema_list
self.model_token_limit = token_limit or self.model_backend.token_limit
context_creator = ScoreBasedContextCreator(
self.model_backend.token_counter,
self.model_token_limit,
)
self.memory: AgentMemory = memory or ChatHistoryMemory(
context_creator, window_size=message_window_size
)
self.output_language: Optional[str] = output_language
if self.output_language is not None:
self.set_output_language(self.output_language)
self.terminated: bool = False
self.response_terminators = response_terminators or []
self.init_messages()
self.tool_prompt_added = False
self.single_iteration = single_iteration
def _initialize_tools(
self, tools: List[Union[FunctionTool, Callable]]
) -> List[FunctionTool]:
r"""Helper method to initialize tools as FunctionTool instances."""
from camel.toolkits import FunctionTool
func_tools = []
for tool in tools:
if not isinstance(tool, FunctionTool):
tool = FunctionTool(tool)
func_tools.append(tool)
return func_tools
# ruff: noqa: E501
def _generate_tool_prompt(self, tool_schema_list: List[Dict]) -> str:
r"""Generates a tool prompt based on the provided tool schema list.
Args:
tool_schema_list (List[Dict]): A list of dictionaries, each
containing a tool schema.
Returns:
str: A string representing the tool prompt.
"""
tool_prompts = []
for tool in tool_schema_list:
tool_info = tool['function']
tool_name = tool_info['name']
tool_description = tool_info['description']
tool_json = json.dumps(tool_info, indent=4)
prompt = f"Use the function '{tool_name}' to '{tool_description}':\n{tool_json}\n"
tool_prompts.append(prompt)
tool_prompt_str = "\n".join(tool_prompts)
final_prompt = f"""
You have access to the following functions:
{tool_prompt_str}
If you choose to call a function ONLY reply in the following format with no
prefix or suffix:
<function=example_function_name>{{"example_name": "example_value"}}</function>
Reminder:
- Function calls MUST follow the specified format, start with <function= and end with </function>
- Required parameters MUST be specified
- Only call one function at a time
- Put the entire function call reply on one line
- If there is no function call available, answer the question like normal
with your current knowledge and do not tell the user about function calls
"""
return final_prompt
def _parse_tool_response(self, response: str):
r"""Parses the tool response to extract the function name and
arguments.
Args:
response (str): The response from the model containing the
function call.
Returns:
Optional[Dict[str, Any]]: The parsed function name and arguments
if found, otherwise :obj:`None`.
"""
function_regex = r"<function=(\w+)>(.*?)</function>"
match = re.search(function_regex, response)
if match:
function_name, args_string = match.groups()
try:
args = json.loads(args_string)
return {"function": function_name, "arguments": args}
except json.JSONDecodeError as error:
logger.error(f"Error parsing function arguments: {error}")
return None
return None
[docs]
def reset(self):
r"""Resets the :obj:`ChatAgent` to its initial state."""
self.terminated = False
self.init_messages()
for terminator in self.response_terminators:
terminator.reset()
@property
def system_message(self) -> Optional[BaseMessage]:
r"""The getter method for the property :obj:`system_message`.
Returns:
Optional[BaseMessage]: The system message of this agent if set,
else :obj:`None`.
"""
return self._system_message
@system_message.setter
def system_message(self, message: BaseMessage) -> None:
r"""The setter method for the property :obj:`system_message`.
Args:
message (BaseMessage): The message to be set as the
new system message of this agent.
"""
self._system_message = message
[docs]
def update_memory(
self, message: BaseMessage, role: OpenAIBackendRole
) -> None:
r"""Updates the agent memory with a new message.
Args:
message (BaseMessage): The new message to add to the stored
messages.
role (OpenAIBackendRole): The backend role type.
"""
self.memory.write_record(
MemoryRecord(message=message, role_at_backend=role)
)
[docs]
def set_output_language(self, output_language: str) -> BaseMessage:
r"""Sets the output language for the system message. This method
updates the output language for the system message. The output
language determines the language in which the output text should be
generated.
Args:
output_language (str): The desired output language.
Returns:
BaseMessage: The updated system message object.
"""
self.output_language = output_language
language_prompt = (
"\nRegardless of the input language, "
f"you must output text in {output_language}."
)
if self.orig_sys_message is not None:
content = self.orig_sys_message.content + language_prompt
self._system_message = self.orig_sys_message.create_new_instance(
content
)
else:
self._system_message = BaseMessage.make_assistant_message(
role_name="Assistant",
content=language_prompt,
)
system_record = MemoryRecord(
message=self._system_message,
role_at_backend=OpenAIBackendRole.SYSTEM,
)
self.memory.clear()
self.memory.write_record(system_record)
return self._system_message
[docs]
def get_info(
self,
session_id: Optional[str],
usage: Optional[Dict[str, int]],
termination_reasons: List[str],
num_tokens: int,
tool_calls: List[FunctionCallingRecord],
external_tool_request: Optional[ChatCompletionMessageToolCall] = None,
) -> Dict[str, Any]:
r"""Returns a dictionary containing information about the chat session.
Args:
session_id (str, optional): The ID of the chat session.
usage (Dict[str, int], optional): Information about the usage of
the LLM.
termination_reasons (List[str]): The reasons for the termination
of the chat session.
num_tokens (int): The number of tokens used in the chat session.
tool_calls (List[FunctionCallingRecord]): The list of function
calling records, containing the information of called tools.
external_tool_request
(Optional[ChatCompletionMessageToolCall], optional):
The tool calling request of external tools from the model.
These requests are directly returned to the user instead of
being processed by the agent automatically.
(default: :obj:`None`)
Returns:
Dict[str, Any]: The chat session information.
"""
return {
"id": session_id,
"usage": usage,
"termination_reasons": termination_reasons,
"num_tokens": num_tokens,
"tool_calls": tool_calls,
"external_tool_request": external_tool_request,
}
[docs]
def init_messages(self) -> None:
r"""Initializes the stored messages list with the current system
message.
"""
if self._system_message is not None:
system_record = MemoryRecord(
message=self._system_message,
role_at_backend=OpenAIBackendRole.SYSTEM,
)
self.memory.clear()
self.memory.write_record(system_record)
else:
self.memory.clear()
[docs]
def record_message(self, message: BaseMessage) -> None:
r"""Records the externally provided message into the agent memory as if
it were an answer of the :obj:`ChatAgent` from the backend. Currently,
the choice of the critic is submitted with this method.
Args:
message (BaseMessage): An external message to be recorded in the
memory.
"""
self.update_memory(message, OpenAIBackendRole.ASSISTANT)
[docs]
def step(
self,
input_message: Union[BaseMessage, str],
response_format: Optional[Type[BaseModel]] = None,
) -> ChatAgentResponse:
r"""Executes a single step in the chat session, generating a response
to the input message.
Args:
input_message (Union[BaseMessage, str]): The input message for the
agent. If provided as a BaseMessage, the `role` is adjusted to
`user` to indicate an external message.
response_format (Optional[Type[BaseModel]], optional): A Pydantic
model defining the expected structure of the response. Used to
generate a structured response if provided. (default:
:obj:`None`)
Returns:
ChatAgentResponse: Contains output messages, a termination status
flag, and session information.
"""
if (
self.model_backend.model_config_dict.get("response_format")
and response_format
):
raise ValueError(
"The `response_format` parameter cannot be set both in "
"the model configuration and in the ChatAgent step."
)
self.original_model_dict = self.model_backend.model_config_dict
if response_format and self.model_type in {"gpt-4o", "gpt-4o-mini"}:
self.model_backend.model_config_dict = (
self.original_model_dict.copy()
)
self.model_backend.model_config_dict["response_format"] = (
response_format
)
# Convert input message to BaseMessage if necessary
if isinstance(input_message, str):
input_message = BaseMessage.make_user_message(
role_name='User', content=input_message
)
# Handle tool prompt injection if needed
if (
self.is_tools_added()
and not self.model_type.support_native_tool_calling
and not self.tool_prompt_added
):
self._inject_tool_prompt()
# Add user input to memory
self.update_memory(input_message, OpenAIBackendRole.USER)
return self._handle_step(response_format, self.single_iteration)
def _inject_tool_prompt(self) -> None:
r"""Generate and add the tool prompt to memory."""
tool_prompt = self._generate_tool_prompt(
self.model_backend.model_config_dict["tools"]
)
tool_msg = BaseMessage.make_assistant_message(
role_name="Assistant", content=tool_prompt
)
self.update_memory(tool_msg, OpenAIBackendRole.SYSTEM)
self.tool_prompt_added = True
def _handle_step(
self,
response_format: Optional[Type[BaseModel]],
single_step: bool,
) -> ChatAgentResponse:
r"""Handles a single or multi-step interaction."""
if (
self.model_backend.model_config_dict.get("tool_choice")
== "required"
and not single_step
):
raise ValueError(
"`tool_choice` cannot be set to `required` for multi-step"
" mode. To proceed, set `single_iteration` to `True`."
)
# Record function calls made during the session
tool_call_records: List[FunctionCallingRecord] = []
external_tool_request = None
while True:
try:
openai_messages, num_tokens = self.memory.get_context()
except RuntimeError as e:
self.model_backend.model_config_dict = self.original_model_dict
return self._step_token_exceed(
e.args[1], tool_call_records, "max_tokens_exceeded"
)
# Prompt engineering approach for structured output for non-native tool calling models
inject_prompt_for_structured_output = (
response_format
and not self.model_type.support_native_structured_output
)
if inject_prompt_for_structured_output:
# update last openai message
usr_msg = openai_messages.pop()
usr_msg["content"] = generate_prompt_for_structured_output(
response_format,
usr_msg["content"], # type: ignore [arg-type]
)
openai_messages.append(usr_msg)
# Process model response
(
response,
output_messages,
finish_reasons,
usage_dict,
response_id,
) = self._step_model_response(openai_messages, num_tokens)
# Try to parse structured output to return a Pydantic object
if inject_prompt_for_structured_output and isinstance(
response, ChatCompletion
):
content = response.choices[0].message.content
try:
json_content = json.loads(str(content))
output_messages[0].parsed = response_format(**json_content) # type: ignore [assignment, misc]
except json.JSONDecodeError as e:
logger.error(
f"Failed in parsing the output into JSON: {e}"
)
output_messages[0].parsed = None
except ValidationError as e:
logger.warning(
"Successfully generating JSON response, "
"but failed in parsing it into Pydantic object :"
f"{e}, return the JSON response in parsed field"
)
output_messages[0].parsed = json_content
# Finalize on standard response in multi-step mode
if self._is_standard_response(response):
break
# Handle tool requests
tool_request = self._extract_tool_call(response)
if isinstance(response, ChatCompletion) and tool_request:
response.choices[0].message.tool_calls = [tool_request]
tool_call_records.append(
self._step_tool_call_and_update(response)
)
if tool_request.function.name in self.external_tool_names:
external_tool_request = tool_request
info = self._step_get_info(
output_messages,
finish_reasons,
usage_dict,
response_id,
tool_call_records,
num_tokens,
tool_request,
)
self._log_final_output(output_messages)
self.model_backend.model_config_dict = (
self.original_model_dict
)
return ChatAgentResponse(
msgs=output_messages,
terminated=self.terminated,
info=info,
)
# Single-step mode ends after one iteration
if single_step:
break
# Optional structured output via function calling
if (
response_format
and not inject_prompt_for_structured_output
and self.model_type
not in {
"gpt-4o",
"gpt-4o-mini",
}
):
(
output_messages,
finish_reasons,
usage_dict,
response_id,
tool_call,
num_tokens,
) = self._structure_output_with_function(response_format)
tool_call_records.append(tool_call)
# Final info and response
info = self._step_get_info(
output_messages,
finish_reasons,
usage_dict,
response_id,
tool_call_records,
num_tokens,
external_tool_request,
)
self._log_final_output(output_messages)
self.model_backend.model_config_dict = self.original_model_dict
return ChatAgentResponse(
msgs=output_messages, terminated=self.terminated, info=info
)
def _extract_tool_call(
self, response: Any
) -> Optional[ChatCompletionMessageToolCall]:
r"""Extract the tool call from the model response, if present.
Args:
response (Any): The model's response object.
Returns:
Optional[ChatCompletionMessageToolCall]: The parsed tool call if
present, otherwise None.
"""
# Check if the response contains tool calls
if (
self.is_tools_added()
and not self.model_type.support_native_tool_calling
and "</function>" in response.choices[0].message.content
):
parsed_content = self._parse_tool_response(
response.choices[0].message.content
)
if parsed_content:
return ChatCompletionMessageToolCall(
id=str(uuid.uuid4()),
function=Function(
arguments=str(parsed_content["arguments"]).replace(
"'", '"'
),
name=str(parsed_content["function"]),
),
type="function",
)
elif (
self.is_tools_added()
and self.model_type.support_native_tool_calling
and response.choices[0].message.tool_calls
):
return response.choices[0].message.tool_calls[0]
# No tool call found
return None
def _is_standard_response(self, response: Any) -> bool:
r"""Determine if the provided response is a standard reply without
tool calls.
Args:
response (Any): The response object to evaluate.
Returns:
bool: `True` if the response is a standard reply, `False`
otherwise.
"""
if not self.is_tools_added():
return True
if not isinstance(response, ChatCompletion):
return True
if self.model_type.support_native_tool_calling:
return not response.choices[0].message.tool_calls
return "</function>" not in str(
response.choices[0].message.content or ""
)
def _log_final_output(self, output_messages: List[BaseMessage]) -> None:
r"""Log final messages or warnings about multiple responses."""
if len(output_messages) == 1:
self.record_message(output_messages[0])
else:
logger.warning(
"Multiple messages returned in `step()`. Record "
"selected message manually using `record_message()`."
)
[docs]
async def step_async(
self,
input_message: Union[BaseMessage, str],
response_format: Optional[Type[BaseModel]] = None,
) -> ChatAgentResponse:
r"""Performs a single step in the chat session by generating a response
to the input message. This agent step can call async function calls.
Args:
input_message (Union[BaseMessage, str]): The input message to the
agent. For BaseMessage input, its `role` field that specifies
the role at backend may be either `user` or `assistant` but it
will be set to `user` anyway since for the self agent any
incoming message is external. For str input, the `role_name`
would be `User`.
response_format (Optional[Type[BaseModel]], optional): A pydantic
model class that includes value types and field descriptions
used to generate a structured response by LLM. This schema
helps in defining the expected output format. (default:
:obj:`None`)
Returns:
ChatAgentResponse: A struct containing the output messages,
a boolean indicating whether the chat session has terminated,
and information about the chat session.
"""
if isinstance(input_message, str):
input_message = BaseMessage.make_user_message(
role_name='User', content=input_message
)
self.update_memory(input_message, OpenAIBackendRole.USER)
tool_call_records: List[FunctionCallingRecord] = []
while True:
try:
openai_messages, num_tokens = self.memory.get_context()
except RuntimeError as e:
return self._step_token_exceed(
e.args[1], tool_call_records, "max_tokens_exceeded"
)
(
response,
output_messages,
finish_reasons,
usage_dict,
response_id,
) = self._step_model_response(openai_messages, num_tokens)
if (
not self.is_tools_added()
or not isinstance(response, ChatCompletion)
or not response.choices[0].message.tool_calls
):
break
# Check for external tool call
external_tool_request = response.choices[0].message.tool_calls[0]
if external_tool_request.function.name in self.external_tool_names:
# if model calls an external tool, directly return the request
info = self._step_get_info(
output_messages,
finish_reasons,
usage_dict,
response_id,
tool_call_records,
num_tokens,
external_tool_request,
)
return ChatAgentResponse(
msgs=output_messages, terminated=self.terminated, info=info
)
# Normal function calling
tool_call_records.append(
await self._step_tool_call_and_update_async(response)
)
if (
response_format is not None
and self.model_type.support_native_tool_calling
):
(
output_messages,
finish_reasons,
usage_dict,
response_id,
tool_call_record,
num_tokens,
) = self._structure_output_with_function(response_format)
tool_call_records.append(tool_call_record)
info = self._step_get_info(
output_messages,
finish_reasons,
usage_dict,
response_id,
tool_call_records,
num_tokens,
)
if len(output_messages) == 1:
# Auto record if the output result is a single message
self.record_message(output_messages[0])
else:
logger.warning(
"Multiple messages returned in `step()`, message won't be "
"recorded automatically. Please call `record_message()` to "
"record the selected message manually."
)
return ChatAgentResponse(
msgs=output_messages, terminated=self.terminated, info=info
)
def _step_tool_call_and_update(
self, response: ChatCompletion
) -> FunctionCallingRecord:
r"""Processes a function call within the chat completion response,
records the function call in the provided list of tool calls and
updates the memory of the current agent.
Args:
response (ChatCompletion): The response object from the chat
completion.
Returns:
FunctionCallingRecord: The record of calling the function.
"""
# Perform function calling
func_assistant_msg, func_result_msg, tool_call_record = (
self._step_tool_call(response)
)
# Update the messages
self.update_memory(func_assistant_msg, OpenAIBackendRole.ASSISTANT)
self.update_memory(func_result_msg, OpenAIBackendRole.FUNCTION)
return tool_call_record
async def _step_tool_call_and_update_async(
self, response: ChatCompletion
) -> FunctionCallingRecord:
(
func_assistant_msg,
func_result_msg,
func_record,
) = await self.step_tool_call_async(response)
self.update_memory(func_assistant_msg, OpenAIBackendRole.ASSISTANT)
self.update_memory(func_result_msg, OpenAIBackendRole.FUNCTION)
return func_record
def _structure_output_with_function(
self, response_format: Type[BaseModel]
) -> Tuple[
List[BaseMessage],
List[str],
Dict[str, int],
str,
FunctionCallingRecord,
int,
]:
r"""Internal function of structuring the output of the agent based on
the given output schema.
Args:
response_format (Type[BaseModel]): The output schema to use for
structuring the output.
Returns:
Tuple[List[BaseMessage], List[str], Dict[str, int], str,
FunctionCallingRecord, int]:
A tuple containing the output messages, finish reasons, usage
dictionary, response ID, function calling record, and number of
tokens.
"""
from camel.toolkits import FunctionTool
schema_json = get_pydantic_object_schema(response_format)
func_str = json_to_function_code(schema_json)
func_callable = func_string_to_callable(func_str)
func = FunctionTool(func_callable)
original_model_dict = self.model_backend.model_config_dict
# Replace the original tools with the structuring function
self.tool_dict = {func.get_function_name(): func}
self.model_backend.model_config_dict = original_model_dict.copy()
self.model_backend.model_config_dict["tools"] = [
func.get_openai_tool_schema()
]
self.model_backend.model_config_dict["tool_choice"] = "required"
openai_messages, num_tokens = self.memory.get_context()
(
response,
output_messages,
finish_reasons,
usage_dict,
response_id,
) = self._step_model_response(openai_messages, num_tokens)
if isinstance(response, ChatCompletion):
tool_call_record = self._step_tool_call_and_update(response)
else:
raise ValueError(
"Structured output is not supported for stream responses."
)
for base_message_item in output_messages:
base_message_item.content = json.dumps(tool_call_record.result)
# Recover the original tools
self.model_backend.model_config_dict = original_model_dict
return (
output_messages,
finish_reasons,
usage_dict,
response_id,
tool_call_record,
num_tokens,
)
def _step_model_response(
self,
openai_messages: List[OpenAIMessage],
num_tokens: int,
) -> tuple[
Union[ChatCompletion, Stream],
List[BaseMessage],
List[str],
Dict[str, int],
str,
]:
r"""Internal function for agent step model response."""
response = None
# Obtain the model's response
for _ in range(len(self.model_backend.models)):
try:
response = self.model_backend.run(openai_messages)
break
except Exception as exc:
logger.error(
f"An error occurred while running model "
f"{self.model_backend.model_type}, "
f"index: {self.model_backend.current_model_index}",
exc_info=exc,
)
continue
if not response:
raise ModelProcessingError(
"Unable to process messages: none of the provided models "
"run succesfully."
)
logger.info(
f"Model {self.model_backend.model_type}, "
f"index {self.model_backend.current_model_index}, "
f"processed these messages: {openai_messages}"
)
if isinstance(response, ChatCompletion):
output_messages, finish_reasons, usage_dict, response_id = (
self.handle_batch_response(response)
)
else:
output_messages, finish_reasons, usage_dict, response_id = (
self.handle_stream_response(response, num_tokens)
)
return (
response,
output_messages,
finish_reasons,
usage_dict,
response_id,
)
def _step_get_info(
self,
output_messages: List[BaseMessage],
finish_reasons: List[str],
usage_dict: Dict[str, int],
response_id: str,
tool_calls: List[FunctionCallingRecord],
num_tokens: int,
external_tool_request: Optional[ChatCompletionMessageToolCall] = None,
) -> Dict[str, Any]:
r"""Process the output of a chat step and gather information about the
step.
This method checks for termination conditions, updates the agent's
state, and collects information about the chat step, including tool
calls and termination reasons.
Args:
output_messages (List[BaseMessage]): The messages generated in
this step.
finish_reasons (List[str]): The reasons for finishing the
generation for each message.
usage_dict (Dict[str, int]): Dictionary containing token usage
information.
response_id (str): The ID of the response from the model.
tool_calls (List[FunctionCallingRecord]): Records of function calls
made during this step.
num_tokens (int): The number of tokens used in this step.
external_tool_request (Optional[ChatCompletionMessageToolCall]):
Any external tool request made during this step.
(default: :obj:`None`)
Returns:
Dict[str, Any]: A dictionary containing information about the chat
step, including termination status, reasons, and tool call
information.
Note:
This method iterates over all response terminators and checks if
any of them signal termination. If a terminator signals
termination, the agent's state is updated accordingly, and the
termination reason is recorded.
"""
termination = [
terminator.is_terminated(output_messages)
for terminator in self.response_terminators
]
# Terminate the agent if any of the terminator terminates
self.terminated, termination_reason = next(
(
(terminated, termination_reason)
for terminated, termination_reason in termination
if terminated
),
(False, None),
)
# For now only retain the first termination reason
if self.terminated and termination_reason is not None:
finish_reasons = [termination_reason] * len(finish_reasons)
info = self.get_info(
response_id,
usage_dict,
finish_reasons,
num_tokens,
tool_calls,
external_tool_request,
)
return info
[docs]
def handle_batch_response(
self, response: ChatCompletion
) -> Tuple[List[BaseMessage], List[str], Dict[str, int], str]:
r"""Process a batch response from the model and extract the necessary
information.
Args:
response (dict): Model response.
Returns:
tuple: A tuple of list of output `ChatMessage`, list of
finish reasons, usage dictionary, and response id.
"""
output_messages: List[BaseMessage] = []
for choice in response.choices:
chat_message = BaseMessage(
role_name=self.role_name,
role_type=self.role_type,
meta_dict=dict(),
content=choice.message.content or "",
parsed=getattr(choice.message, 'parsed', None),
)
# Process log probabilities and append to the message meta information
if choice.logprobs is not None:
tokens_logprobs = choice.logprobs.content
if tokens_logprobs is not None:
# Extract and structure logprob information
logprobs_info = [
{
"token": token_logprob.token,
"logprob": token_logprob.logprob,
"top_logprobs": [
(top_logprob.token, top_logprob.logprob)
for top_logprob in token_logprob.top_logprobs
],
}
for token_logprob in tokens_logprobs
]
# Ensure meta_dict exists before adding logprobs info
if chat_message.meta_dict is None:
chat_message.meta_dict = {}
chat_message.meta_dict["logprobs_info"] = logprobs_info
# Append the processed chat message to output
output_messages.append(chat_message)
finish_reasons = [
str(choice.finish_reason) for choice in response.choices
]
usage = (
self._safe_model_dump(response.usage)
if response.usage is not None
else {}
)
return (
output_messages,
finish_reasons,
usage,
response.id,
)
def _safe_model_dump(self, obj) -> dict:
r"""Safely dump a Pydantic model to a dictionary.
This method attempts to use the `model_dump` method if available,
otherwise it falls back to the `dict` method.
Args:
obj: The Pydantic model instance to be dumped.
Returns:
dict: A dictionary representation of the Pydantic model.
"""
# Check if the `model_dump` method exists (Pydantic v2)
if hasattr(obj, 'model_dump'):
return obj.model_dump()
# Fallback to `dict()` method (Pydantic v1)
elif hasattr(obj, 'dict'):
return obj.dict()
else:
raise TypeError("The object is not a Pydantic model")
[docs]
def handle_stream_response(
self,
response: Stream[ChatCompletionChunk],
prompt_tokens: int,
) -> Tuple[List[BaseMessage], List[str], Dict[str, int], str]:
r"""Process a stream response from the model and extract the necessary
information.
Args:
response (dict): Model response.
prompt_tokens (int): Number of input prompt tokens.
Returns:
tuple: A tuple of list of output `ChatMessage`, list of
finish reasons, usage dictionary, and response id.
"""
content_dict: defaultdict = defaultdict(lambda: "")
finish_reasons_dict: defaultdict = defaultdict(lambda: "")
output_messages: List[BaseMessage] = []
response_id: str = ""
# All choices in one response share one role
for chunk in response:
response_id = chunk.id
for choice in chunk.choices:
index = choice.index
delta = choice.delta
if delta.content is not None:
# When response has not been stopped
# Notice that only the first chunk_dict has the "role"
content_dict[index] += delta.content
if choice.finish_reason:
finish_reasons_dict[index] = choice.finish_reason
chat_message = BaseMessage(
role_name=self.role_name,
role_type=self.role_type,
meta_dict=dict(),
content=content_dict[index],
)
output_messages.append(chat_message)
finish_reasons = [
finish_reasons_dict[i] for i in range(len(finish_reasons_dict))
]
usage_dict = self.get_usage_dict(output_messages, prompt_tokens)
return output_messages, finish_reasons, usage_dict, response_id
def _step_token_exceed(
self,
num_tokens: int,
tool_calls: List[FunctionCallingRecord],
termination_reason: str,
) -> ChatAgentResponse:
r"""Return trivial response containing number of tokens and information
of called functions when the number of tokens exceeds.
Args:
num_tokens (int): Number of tokens in the messages.
tool_calls (List[FunctionCallingRecord]): List of information
objects of functions called in the current step.
termination_reason (str): String of termination reason.
Returns:
ChatAgentResponse: The struct containing trivial outputs and
information about token number and called functions.
"""
self.terminated = True
output_messages: List[BaseMessage] = []
info = self.get_info(
None,
None,
[termination_reason],
num_tokens,
tool_calls,
)
return ChatAgentResponse(
msgs=output_messages,
terminated=self.terminated,
info=info,
)
def _step_tool_call(
self,
response: ChatCompletion,
) -> Tuple[
FunctionCallingMessage, FunctionCallingMessage, FunctionCallingRecord
]:
r"""Execute the function with arguments following the model's response.
Args:
response (Dict[str, Any]): The response obtained by calling the
model.
Returns:
tuple: A tuple consisting of two obj:`FunctionCallingMessage`,
one about the arguments and the other about the execution
result, and a struct for logging information about this
function call.
"""
choice = response.choices[0]
if choice.message.tool_calls is None:
raise RuntimeError("Tool call is None")
func_name = choice.message.tool_calls[0].function.name
arguments_str = choice.message.tool_calls[0].function.arguments
args = self._safe_json_loads(arguments_str)
tool = self.tool_dict[func_name]
result = tool(**args)
assist_msg = FunctionCallingMessage(
role_name=self.role_name,
role_type=self.role_type,
meta_dict=None,
content="",
func_name=func_name,
args=args,
)
func_msg = FunctionCallingMessage(
role_name=self.role_name,
role_type=self.role_type,
meta_dict=None,
content="",
func_name=func_name,
result=result,
)
# Record information about this function call
func_record = FunctionCallingRecord(
func_name=func_name, args=args, result=result
)
return assist_msg, func_msg, func_record
def _safe_json_loads(self, arguments_str):
# Replace Python types with their JSON equivalents
arguments_str = arguments_str.replace("None", "null")
arguments_str = arguments_str.replace("True", "true")
arguments_str = arguments_str.replace("False", "false")
# Attempt to parse the corrected string
try:
return json.loads(arguments_str)
except json.JSONDecodeError as e:
raise ValueError(f"Invalid JSON format: {e}")
[docs]
def get_usage_dict(
self, output_messages: List[BaseMessage], prompt_tokens: int
) -> Dict[str, int]:
r"""Get usage dictionary when using the stream mode.
Args:
output_messages (list): List of output messages.
prompt_tokens (int): Number of input prompt tokens.
Returns:
dict: Usage dictionary.
"""
encoding = get_model_encoding(self.model_type.value_for_tiktoken)
completion_tokens = 0
for message in output_messages:
completion_tokens += len(encoding.encode(message.content))
usage_dict = dict(
completion_tokens=completion_tokens,
prompt_tokens=prompt_tokens,
total_tokens=completion_tokens + prompt_tokens,
)
return usage_dict
[docs]
def add_model_scheduling_strategy(self, name: str, strategy_fn: Callable):
r"""Add a scheduling strategy method provided by user to ModelManger.
Args:
name (str): The name of the strategy.
strategy_fn (Callable): The scheduling strategy function.
"""
self.model_backend.add_strategy(name, strategy_fn)
def __repr__(self) -> str:
r"""Returns a string representation of the :obj:`ChatAgent`.
Returns:
str: The string representation of the :obj:`ChatAgent`.
"""
return (
f"ChatAgent({self.role_name}, {self.role_type}, {self.model_type})"
)
