tmp

zhaoqingang

2025-02-20 6efaa41969ab89c1e1bd468b1b895f6689b7f70d

 app/service/v2/chat.py

@@ -1,28 +1,319 @@
import io
import json

import fitz

from Log import logger
from app.config.agent_base_url import RG_CHAT_DIALOG, DF_CHAT_AGENT, DF_CHAT_PARAMETERS, RG_CHAT_SESSIONS, \
    DF_CHAT_WORKFLOW, DF_UPLOAD_FILE
from app.config.config import settings
from app.config.const import *
from app.models import DialogModel, ApiTokenModel, UserTokenModel
from app.models.v2.session_model import ChatSessionDao, ChatData
from app.service.v2.app_driver.chat_agent import ChatAgent
from app.service.v2.app_driver.chat_data import ChatBaseApply
from app.service.v2.app_driver.chat_dialog import ChatDialog
from app.service.v2.app_driver.chat_workflow import ChatWorkflow
from docx import Document
from dashscope import get_tokenizer  # dashscope版本 >= 1.14.0


async def service_chat_dialog(chat_id:str, question: str, session_id: str):
    token = "ragflow-YzMzE1NDRjYzMyZjExZWY5ZjkxMDI0Mm"
    url = f"/api/v1/chats/{chat_id}/completions"
    chat = ChatDialog(token)
    data = {
        "question": question,
        "stream": True,
        "session_id": session_id
    }
    headers = {
        'Content-Type': 'application/json',
        'Authorization': f"Bearer {token}"
    }
async def update_session_log(db, session_id: str, message: dict, conversation_id: str):
    await ChatSessionDao(db).update_session_by_id(
        session_id=session_id,
        session=None,
        message=message,
        conversation_id=conversation_id
    )


async def add_session_log(db, session_id: str, question: str, chat_id: str, user_id, event_type: str,
                          conversation_id: str):
    try:
        for ans in chat.chat_completions(url, data, headers):

            yield "data:" + json.dumps({"code": 0, "message": "", "data": ans}, ensure_ascii=False) + "\n\n"
        ConversationService.update_by_id(conv.id, conv.to_dict())
        session = await ChatSessionDao(db).update_or_insert_by_id(
            session_id=session_id,
            name=question[:255],
            agent_id=chat_id,
            agent_type=1,
            tenant_id=user_id,
            message={"role": "user", "content": question},
            conversation_id=conversation_id,
            event_type=event_type
        )
        return session
    except Exception as e:
        yield "data:" + json.dumps({"code": 500, "message": str(e),
                                    "data": {"answer": "**ERROR**: " + str(e), "reference": []}},
                                   ensure_ascii=False) + "\n\n"
    yield "data:" + json.dumps({"code": 0, "message": "", "data": True}, ensure_ascii=False) + "\n\n"
        logger.error(e)
    return None

async def get_app_token(db, app_id):
    app_token = db.query(UserTokenModel).filter_by(id=app_id).first()
    if app_token:
        return app_token.access_token
    return ""



async def get_chat_token(db, app_id):
    app_token = db.query(ApiTokenModel).filter_by(app_id=app_id).first()
    if app_token:
        return app_token.token
    return ""


async def add_chat_token(db, data):
    try:
        api_token = ApiTokenModel(**data)
        db.add(api_token)
        db.commit()
    except Exception as e:
        logger.error(e)



async def get_chat_info(db, chat_id: str):
    return db.query(DialogModel).filter_by(id=chat_id, status=Dialog_STATSU_ON).first()


async def get_chat_object(mode):
    if mode == workflow_chat:
        url = settings.dify_base_url + DF_CHAT_WORKFLOW
        return ChatWorkflow(), url
    else:
        url = settings.dify_base_url + DF_CHAT_AGENT
        return ChatAgent(), url


async def service_chat_dialog(db, chat_id: str, question: str, session_id: str, user_id, mode: str):
    conversation_id = ""
    token = await get_chat_token(db, rg_api_token)
    url = settings.fwr_base_url + RG_CHAT_DIALOG.format(chat_id)
    chat = ChatDialog()
    session = await add_session_log(db, session_id, question, chat_id, user_id, mode, session_id)
    if session:
        conversation_id = session.conversation_id
    message = {"role": "assistant", "answer": "", "reference": {}}
    try:
        async for ans in chat.chat_completions(url, await chat.request_data(question, conversation_id),
                                               await chat.get_headers(token)):
            data = {}
            error = ""
            status = http_200
            if ans.get("code", None) == 102:
                error = ans.get("message", "error！")
                status = http_400
                event = smart_message_error
            else:
                if isinstance(ans.get("data"), bool) and ans.get("data") is True:
                    event = smart_message_end
                else:
                    data = ans.get("data", {})
                    # conversation_id = data.get("session_id", "")
                    if "session_id" in data:
                        del data["session_id"]
                    message = data
                    event = smart_message_cover
            message_str = "data: " + json.dumps(
                {"event": event, "data": data, "error": error, "status": status, "session_id": session_id},
                ensure_ascii=False) + "\n\n"
            for i in range(0, len(message_str), max_chunk_size):
                chunk = message_str[i:i + max_chunk_size]
                # print(chunk)
                yield chunk  # 发送分块消息
    except Exception as e:

        logger.error(e)
        try:
            yield "data: " + json.dumps({"message": smart_message_error,
                                         "error": "\n**ERROR**: " + str(e), "status": http_500},
                                        ensure_ascii=False) + "\n\n"
        except:
            ...
    finally:
        await update_session_log(db, session_id, message, conversation_id)


async def service_chat_workflow(db, chat_id: str, chat_data: ChatData, session_id: str, user_id, mode: str):
    conversation_id = ""
    answer_event = ""
    answer_agent = ""
    message_id = ""
    task_id = ""
    error = ""
    files = []
    node_list = []
    token = await get_chat_token(db, chat_id)
    chat, url = await get_chat_object(mode)
    if hasattr(chat_data, "query"):
        query = chat_data.query
    else:
        query = "start new workflow"
    session = await add_session_log(db, session_id, query, chat_id, user_id, mode, conversation_id)
    if session:
        conversation_id = session.conversation_id
    try:
        async for ans in chat.chat_completions(url,
                                               await chat.request_data(query, conversation_id, str(user_id), chat_data),
                                               await chat.get_headers(token)):
            data = {}
            status = http_200
            conversation_id = ans.get("conversation_id")
            task_id = ans.get("task_id")
            if ans.get("event") == message_error:
                error = ans.get("message", "参数异常！")
                status = http_400
                event = smart_message_error
            elif ans.get("event") == message_agent:
                data = {"answer": ans.get("answer", ""), "id": ans.get("message_id", "")}
                answer_agent += ans.get("answer", "")
                message_id = ans.get("message_id", "")
                event = smart_message_stream
            elif ans.get("event") == message_event:
                data = {"answer": ans.get("answer", ""), "id": ans.get("message_id", "")}
                answer_event += ans.get("answer", "")
                message_id = ans.get("message_id", "")
                event = smart_message_stream
            elif ans.get("event") == message_file:
                data = {"url": ans.get("url", ""), "id": ans.get("id", ""),
                        "type": ans.get("type", "")}
                files.append(data)
                event = smart_message_file
            elif ans.get("event") in [workflow_started, node_started, node_finished]:
                data = ans.get("data", {})
                data["inputs"] = []
                data["outputs"] = []
                data["process_data"] = ""
                node_list.append(ans)
                event = [smart_workflow_started, smart_node_started, smart_node_finished][
                    [workflow_started, node_started, node_finished].index(ans.get("event"))]
            elif ans.get("event") == workflow_finished:
                data = ans.get("data", {})
                event = smart_workflow_finished
                node_list.append(ans)

            elif ans.get("event") == message_end:
                event = smart_message_end
            else:
                continue

            yield "data: " + json.dumps(
                {"event": event, "data": data, "error": error, "status": status, "task_id": task_id,
                 "session_id": session_id},
                ensure_ascii=False) + "\n\n"

    except Exception as e:
        logger.error(e)
        try:
            yield "data: " + json.dumps({"message": smart_message_error,
                                         "error": "\n**ERROR**: " + str(e), "status": http_500},
                                        ensure_ascii=False) + "\n\n"
        except:
            ...
    finally:
        await update_session_log(db, session_id, {"role": "assistant", "answer": answer_event or answer_agent,
                                                  "node_list": node_list, "task_id": task_id, "id": message_id,
                                                  "error": error}, conversation_id)


async def service_chat_basic(db, chat_id: str, chat_data: ChatData, session_id: str, user_id, mode: str):
    ...


async def service_chat_parameters(db, chat_id, user_id):
    chat_info = db.query(DialogModel).filter_by(id=chat_id).first()
    if not chat_info:
        return {}
    return chat_info.parameters
    # if chat_info.dialog_type == RG_TYPE:
    #     return {"retriever_resource":
    #         {
    #             "enabled": True
    #         }
    #     }
    # elif chat_info.dialog_type == BASIC_TYPE:
    #     ...
    # elif chat_info.dialog_type == DF_TYPE:
    #     token = await get_chat_token(db, chat_id)
    #     if not token:
    #         return {}
    #     url = settings.dify_base_url + DF_CHAT_PARAMETERS
    #     chat = ChatBaseApply()
    #     return await chat.chat_get(url, {"user": str(user_id)}, await chat.get_headers(token))


async def service_chat_sessions(db, chat_id, name):
    token = await get_chat_token(db, rg_api_token)
    if not token:
        return {}
    url = settings.fwr_base_url + RG_CHAT_SESSIONS.format(chat_id)
    chat = ChatDialog()
    return await chat.chat_sessions(url, {"name": name}, await chat.get_headers(token))


async def service_chat_upload(db, chat_id, file, user_id):
    files = []
    token = await get_chat_token(db, chat_id)
    if not token:
        return files
    url = settings.dify_base_url + DF_UPLOAD_FILE
    chat = ChatBaseApply()
    for f in file:
        try:
            file_content = await f.read()
            file_upload = await chat.chat_upload(url, {"file": (f.filename, file_content)}, {"user": str(user_id)},
                                                 {'Authorization': f'Bearer {token}'})
            try:
                tokens = await read_file(file_content, f.filename, f.content_type)
                file_upload["tokens"] = tokens
            except:
                ...
            files.append(file_upload)
        except Exception as e:
            logger.error(e)
    return json.dumps(files) if files else ""


async def get_str_token(input_str):
    # 获取tokenizer对象，目前只支持通义千问系列模型
    tokenizer = get_tokenizer('qwen-turbo')
    # 将字符串切分成token并转换为token id
    tokens = tokenizer.encode(input_str)
    # print(f"经过切分后的token id为：{tokens}。")
    # # 经过切分后的token id为： [31935, 64559, 99320, 56007, 100629, 104795, 99788, 1773]
    # print(f"经过切分后共有{len(tokens)}个token")
    # # 经过切分后共有8个token
    #
    # # 将token id转化为字符串并打印出来
    # for i in range(len(tokens)):
    #     print(f"token id为{tokens[i]}对应的字符串为：{tokenizer.decode(tokens[i])}")
    return len(tokens)

async def read_pdf(pdf_stream):
    text = ""
    with fitz.open(stream=pdf_stream, filetype="pdf") as pdf_document:
        for page in pdf_document:
            text += page.get_text()
    return text


async def read_word(word_stream):
    # 使用 python-docx 打开 Word 文件流
    doc = Document(io.BytesIO(word_stream))

    # 提取每个段落的文本
    text = ""
    for para in doc.paragraphs:
        text += para.text

    return text

async def read_file(file, filename, content_type):
    text = ""
    if content_type == "application/pdf" or filename.endswith('.pdf'):

        # 提取 PDF 内容
        text = await read_pdf(file)
    elif content_type == "application/vnd.openxmlformats-officedocument.wordprocessingml.document" or filename.endswith(
            '.docx'):
        text = await read_word(file)

    return await get_str_token(text)