AIGN.py

import os
import re
import time
from typing import Optional

import agentscope
from agentscope.agents import AgentBase, DialogAgent, UserAgent
from agentscope.message import Msg
from agentscope.prompt import PromptEngine, PromptType

from AIGN_Prompt import *


def Retryer(func, max_retries=10):
    def wrapper(*args, **kwargs):
        for _ in range(max_retries):
            try:
                return func(*args, **kwargs)
            except Exception as e:
                print("-" * 30 + f"\n失败：\n{e}\n" + "-" * 30)
                time.sleep(2.333)
        raise ValueError("失败")

    return wrapper


class MarkdownAgent(AgentBase):
    """专门应对输入输出都是md格式的情况，例如小说生成"""

    def __init__(
        self,
        chatLLM,
        sys_prompt: str,
        user_prompt: str,
        name: str,
        temperature=0.8,
        top_p=0.8,
        use_memory=False,
        first_replay="明白了。",
        # first_replay=None,
        is_speak=True,
    ) -> None:
        super().__init__(name=name, use_memory=False)

        self.chatLLM = chatLLM
        self.sys_prompt = sys_prompt
        self.user_prompt = user_prompt
        self.temperature = temperature
        self.top_p = top_p
        self.use_memory = use_memory
        self.is_speak = is_speak

        self.history = [{"role": "system", "content": self.sys_prompt}, {"role": "user", "content": self.user_prompt}]

        if first_replay:
            self.history.append({"role": "assistant", "content": first_replay})
        else:
            resp = chatLLM(messages=self.history)
            self.history.append({"role": "assistant", "content": resp["content"]})
            if self.is_speak:
                self.speak(Msg(self.name, resp["content"]))

    def query(self, user_input: str) -> str:
        resp = self.chatLLM(
            messages=self.history + [{"role": "user", "content": user_input}],
            temperature=self.temperature,
            top_p=self.top_p,
        )
        if self.use_memory:
            self.history.append({"role": "user", "content": user_input})
            self.history.append({"role": "assistant", "content": resp["content"]})

        return resp

    def getOutput(self, input_content: str, output_keys: list) -> dict:
        """解析类md格式中 # key 的内容"""
        resp = self.query(input_content)
        output = resp["content"]

        lines = output.split("\n")
        sections = self.parse_sections1(lines, output_keys)

        # 检查是否所有需要的键都存在
        for k in output_keys:
            if (k not in sections) or (len(sections[k]) == 0):
                # 单独对k进行重新parse_sections2，此时查找##，并更新sections
                section_content = self.parse_sections2(lines, k)
                if section_content:
                    sections[k] = section_content
                else:
                    raise ValueError(f"fail to parse {k} in output:\n")

        if self.is_speak:
            self.speak(
                Msg(
                    self.name,
                    f"total_tokens: {resp['total_tokens']}\n{resp['content']}\n",
                )
            )
        return sections

    def parse_sections1(self, lines, output_keys):
        sections = {key: "" for key in output_keys}
        current_section = ""
        for line in lines:
            if line.startswith("# ") or line.startswith(" # "):
                # new key
                current_section = line[2:].strip()
                sections[current_section] = []
            else:
                # add content to current key
                if current_section:
                    sections[current_section].append(line.strip())
        for key in sections.keys():
            sections[key] = "\n".join(sections[key]).strip()

        return sections

    def parse_sections2(self, lines, k):
        content = []
        capturing = False
        for line in lines:
            stripped_line = line.strip()
            if stripped_line.startswith(("##", " ##", "###", " ###")) and k.lower() in stripped_line.lower():
                capturing = True
                continue
            elif stripped_line.startswith(("##", " ##", "###", " ###")) and capturing:
                break
            if capturing:
                content.append(stripped_line)
        return "\n".join(content).strip()

    def invoke(self, inputs: dict, output_keys: list) -> dict:
        input_content = ""
        for k, v in inputs.items():
            if isinstance(v, str) and len(v) > 0:
                input_content += f"# {k}\n{v}\n\n"

        #取消重试便于查找错误
        result = Retryer(self.getOutput)(input_content, output_keys)
        #result = self.getOutput(input_content, output_keys)

        return result

    def clear_memory(self):
        if self.use_memory:
            self.history = self.history[:2]


class AIGN:
    def __init__(self, chatLLM):
        agentscope.init()
        self.chatLLM = chatLLM

        self.novel_outline = ""
        self.paragraph_list = []
        self.novel_content = ""
        self.writing_plan = ""
        self.temp_setting = ""
        self.writing_memory = ""
        self.no_memory_paragraph = ""
        self.user_idea = ""
        self.user_requriments = ""
        #self.embellishment_idea = ""
        self.history_states = []  # 用于存储历史状态
        self.chapter_list = [] # 用于存储章节列表

        self.novel_outline_writer = MarkdownAgent(
            chatLLM=self.chatLLM,
            sys_prompt=system_prompt,
            user_prompt=novel_outline_writer_prompt,
            name="NovelOutlineWriter",
            temperature=0.98,
        )
        #self.novel_beginning_writer = MarkdownAgent(
            #chatLLM=self.chatLLM,
            #sys_prompt=system_prompt + "不少于5000字",
            #user_prompt=novel_beginning_writer_prompt,
            #name="NovelBeginningWriter",
            #temperature=0.80,
        #)
        self.novel_writer = MarkdownAgent(
            chatLLM=self.chatLLM,
            sys_prompt=system_prompt,
            user_prompt=novel_writer_prompt,
            name="NovelWriter",
            temperature=0.81,
        )
        self.memory_maker = MarkdownAgent(
            chatLLM=self.chatLLM,
            sys_prompt=system_prompt,
            user_prompt=memory_maker_prompt,
            name="MemoryMaker",
            temperature=0.66,
        )

    def split_chapters(self, novel_content):
        # 使用正则表达式匹配章节标题
        chapter_pattern = re.compile(r'(?:##?|)?\s*第([一二三四五六七八九十百千万亿\d]+)章[:：]?\s*(.+)')
        
        # 将小说正文按章节标题分割
        chapters = chapter_pattern.split(novel_content)
        
        # 移除第一个空字符串(如果存在)
        if chapters[0] == '':
            chapters = chapters[1:]
        
        # 将章节标题和内容组合成元组
        chapter_tuples = []
        for i in range(0, len(chapters), 3):
            if i + 2 < len(chapters):
                chapter_num = chapters[i]
                chapter_title = chapters[i + 1]
                chapter_content = chapters[i + 2]
                chapter_tuples.append((f"第{chapter_num}章 {chapter_title}", chapter_content))
        
        return chapter_tuples

    def update_chapter_list(self):
        self.chapter_list = self.split_chapters(self.novel_content)


    def updateNovelContent(self):
        self.novel_content = ""
        for paragraph in self.paragraph_list:
            self.novel_content += f"{paragraph}\n\n"

        self.update_chapter_list()
        return self.novel_content

    def genNovelOutline(self, user_idea=None):
        if user_idea:
            self.user_idea = user_idea
        resp = self.novel_outline_writer.invoke(
            inputs={"用户想法": self.user_idea},
            output_keys=["大纲"],
        )
        self.novel_outline = resp["大纲"]
        return self.novel_outline

    # 添加分隔符以便于二次编辑
    def add_separator(self):
        separator = "\n---------------------下一段---------------------\n"
        self.novel_content += separator
        if self.paragraph_list:
            self.paragraph_list[-1] += separator

    def genBeginning(self, user_requriments=None):
        if user_requriments:
            self.user_requriments = user_requriments

        resp = self.novel_beginning_writer.invoke(
            inputs={
                "用户想法": self.user_idea,
                "小说大纲": self.novel_outline,
                "用户要求": self.user_requriments,
            },
            output_keys=["开头", "计划", "临时设定"],
        )
        beginning = resp["开头"]
        self.writing_plan = resp["计划"]
        self.temp_setting = resp["临时设定"]

        self.paragraph_list.append(beginning)
        self.updateNovelContent()

        self.update_chapter_list()

        #self.add_separator()  # 添加分隔符便于二次编辑
        return beginning

    def getLastParagraph(self, max_length=2000):
        last_paragraph = ""

        for i in range(0, len(self.paragraph_list)):
            if (len(last_paragraph) + len(self.paragraph_list[-1 - i])) < max_length:
                last_paragraph = self.paragraph_list[-1 - i] + "\n" + last_paragraph
            else:
                break
        return last_paragraph

    def recordNovel(self):
        record_content = ""
        record_content += f"# 大纲\n\n{self.novel_outline}\n\n"
        record_content += f"# 正文\n\n"
        record_content += self.novel_content
        record_content += f"# 记忆\n\n{self.writing_memory}\n\n"
        record_content += f"# 计划\n\n{self.writing_plan}\n\n"
        record_content += f"# 临时设定\n\n{self.temp_setting}\n\n"

        with open("novel_record.md", "w", encoding="utf-8") as f:
            f.write(record_content)

    def updateMemory(self):
        if (len(self.no_memory_paragraph)) > 2000:
            resp = self.memory_maker.invoke(
                inputs={
                    "前文记忆": self.writing_memory,
                    "正文内容": self.no_memory_paragraph,
                },
                output_keys=["新的记忆"],
            )
            self.writing_memory = resp["新的记忆"]
            self.no_memory_paragraph = ""


    def save_state(self):
        state = {
            "novel_outline": self.novel_outline,
            "paragraph_list": self.paragraph_list,
            "novel_content": self.novel_content,
            "writing_plan": self.writing_plan,
            "temp_setting": self.temp_setting,
            "writing_memory": self.writing_memory
        }
        self.history_states.append(state)
    
    def undo(self):
        if self.history_states:
            previous_state = self.history_states.pop()
            self.novel_outline = previous_state["novel_outline"]
            self.paragraph_list = previous_state["paragraph_list"]
            self.novel_content = previous_state["novel_content"]
            self.writing_plan = previous_state["writing_plan"]
            self.temp_setting = previous_state["temp_setting"]
            self.writing_memory = previous_state["writing_memory"]
            return True
        return False

    def genNextParagraph(self, user_requriments=None):
        self.save_state()  # 保存当前状态
        if user_requriments:
            self.user_requriments = user_requriments

        resp = self.novel_writer.invoke(
            inputs={
                "用户想法": self.user_idea,
                "大纲": self.novel_outline,
                "前文记忆": self.writing_memory,
                "临时设定": self.temp_setting,
                "计划": self.writing_plan,
                "用户要求": self.user_requriments,
                "上文内容": self.getLastParagraph(),
            },
            output_keys=["段落", "计划", "临时设定"],
        )
        next_paragraph = resp["段落"]
        next_writing_plan = resp["计划"]
        next_temp_setting = resp["临时设定"]

        self.paragraph_list.append(next_paragraph)
        self.writing_plan = next_writing_plan
        self.temp_setting = next_temp_setting

        self.no_memory_paragraph += f"\n{next_paragraph}"

        self.updateMemory()
        self.updateNovelContent()
        self.recordNovel()

        self.update_chapter_list()

        #self.add_separator()  # 添加分隔符便于二次编辑
        return next_paragraph