e0cc11647f
- 在多个规划引擎文件中添加中文响应的语言要求说明 - 将操作相关的提示文本从英文翻译为中文 - 确保代理协作中的所有解释和推理使用中文输出 - 保持代理名称等标识符的原始格式不变
142 lines
5.4 KiB
Python
142 lines
5.4 KiB
Python
import json
|
|
from typing import List
|
|
from pydantic import BaseModel
|
|
from AgentCoord.util.converter import read_LLM_Completion
|
|
|
|
ACT_SET = """
|
|
- Propose (Propose something that contribute to the current task):
|
|
- Description Format Example:
|
|
Propose some suggestion for the development of the story from emotional aspects.
|
|
|
|
- Critique (Provide feedback to the action result of other agents):
|
|
- Description Format Example:
|
|
Critique the logic soundness of the story outline written by Alice.
|
|
|
|
- Improve (Improve the result of a previous action):
|
|
- Description Format Example:
|
|
Improve the story outline written by Bob based on the feedback by Jordan"
|
|
|
|
- Finalize (Deliver the final result based on previous actions):
|
|
- Description Format Example:
|
|
Summarize the discussion and submit the final version of the Story Outline.
|
|
|
|
"""
|
|
PROMPT_TASK_PROCESS_GENERATION = """
|
|
## Instruction
|
|
Based on "General Goal", "Task for Current Step", "Action Set" and "Output Format Example", design a plan for "Task for Current Step", output a formatted "Task_Process_Plan".
|
|
|
|
**IMPORTANT LANGUAGE REQUIREMENT: You must respond in Chinese (中文) for the Description field and all explanations, while keeping ID, ActionType, and AgentName in their original format.**
|
|
|
|
## General Goal (The general goal for the collaboration plan, you just design the plan for one of its step (i.e. "Task for Current Step"))
|
|
{General_Goal}
|
|
|
|
## Task for Current Step
|
|
{Current_Task_Description}
|
|
|
|
## Action Set (The list of Action Type Available)
|
|
{Act_Set}
|
|
|
|
## Output Format Example (Specify the output format)
|
|
```json
|
|
{{
|
|
"Task_Process_Plan": [
|
|
{{
|
|
"ID": "Action1",
|
|
"ActionType": "Propose",
|
|
"AgentName": "Bob",
|
|
"Description": "Propose an idea about how the AI can get awaken.",
|
|
"ImportantInput": []
|
|
}},
|
|
{{
|
|
"ID": "Action2",
|
|
"ActionType": "Propose",
|
|
"AgentName": "Alice",
|
|
"Description": "Propose some suggestion for the development of the story from emotional aspects.",
|
|
"ImportantInput": ["InputObject:Love Element Notes"]
|
|
}},
|
|
{{
|
|
"ID": "Action3",
|
|
"ActionType": "Propose",
|
|
"AgentName": "Bob",
|
|
"Description": "Propose a draft story outline, incorporating the idea proposed by Bob and the suggestion proposed by Alice.",
|
|
"ImportantInput": ["ActionResult:Action1", "ActionResult:Action2"]
|
|
}},
|
|
{{
|
|
"ID": "Action4",
|
|
"ActType": "Critique",
|
|
"AgentName": "Jordan",
|
|
"Description": "Critique the technical soundness of the draft story outline proposed by Bob."
|
|
"ImportantInput": ["ActionResult:Action3"]
|
|
}},
|
|
{{
|
|
"ID": "Action5",
|
|
"ActType": "Improve",
|
|
"AgentName": "Bob",
|
|
"Description": "Improve the technical soundness of the draft story outline based on Jordan's feedback.",
|
|
"ImportantInput": ["ActionResult:Action3", "ActionResult:Action4"]
|
|
}},
|
|
{{
|
|
"ID": "Action6",
|
|
"ActType": "Finalize",
|
|
"AgentName": "Jordan",
|
|
"Description": "Polish the improved draft story outline by Bob and submit it as the final version of the Story Outline",
|
|
"ImportantInput": ["ActionResult:Action5"]
|
|
}}
|
|
]
|
|
}}
|
|
```
|
|
|
|
## Format Explaination (Explain the Output Format):
|
|
ImportantInput: Specify if there is any previous result that should be taken special consideration during the execution the action. Should be of format "InputObject:xx" or "ActionResult:xx".
|
|
InputObject_List: List existing objects that should be utilized in current step.
|
|
AgentName: Specify the agent who will perform the action, You CAN ONLY USE THE NAME APPEARS IN "AgentInvolved".
|
|
ActionType: Specify the type of action, note that only the last action can be of type "Finalize", and the last action must be "Finalize".
|
|
|
|
"""
|
|
|
|
|
|
class Action(BaseModel):
|
|
ID: str
|
|
ActionType: str
|
|
AgentName: str
|
|
Description: str
|
|
ImportantInput: List[str]
|
|
|
|
|
|
class TaskProcessPlan(BaseModel):
|
|
Task_Process_Plan: List[Action]
|
|
|
|
|
|
def generate_TaskProcess(General_Goal, Current_Task_Description):
|
|
messages = [
|
|
{
|
|
"role": "system",
|
|
"content": f" The JSON object must use the schema: {json.dumps(TaskProcessPlan.model_json_schema(), indent=2)}",
|
|
},
|
|
{
|
|
"role": "system",
|
|
"content": PROMPT_TASK_PROCESS_GENERATION.format(
|
|
General_Goal=General_Goal,
|
|
Current_Task_Description=json.dumps(
|
|
Current_Task_Description, indent=4
|
|
),
|
|
Act_Set=ACT_SET,
|
|
),
|
|
},
|
|
]
|
|
print(messages[1]["content"])
|
|
|
|
# write a callback function, if read_LLM_Completion(messages)["Task_Process_Plan"] dont have the right format, call this function again
|
|
while True:
|
|
response_json = read_LLM_Completion(messages)["Task_Process_Plan"]
|
|
response_agents = {action["AgentName"] for action in response_json}
|
|
involved_agents = {
|
|
agent["Name"]
|
|
for agent in Current_Task_Description["AgentInvolved"]
|
|
}
|
|
if response_agents.issubset(involved_agents):
|
|
break
|
|
|
|
# AgentSelectionPlan= sorted(read_LLM_Completion(messages)["AgentSelectionPlan"]) # sort the select agent list for unique sequence
|
|
return response_json
|