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 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: 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: 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)