diff --git a/README.md b/README.md index e18e07dfb..7e3616dfe 100644 --- a/README.md +++ b/README.md @@ -1,10 +1,10 @@ # DB-GPT A Open Database-GPT Experiment, A fully localized project. -一个数据库相关的GPT实验项目, 模型与数据全部本地化部署, 绝对保障数据的隐私安全。 同时此GPT项目可以直接本地部署连接到私有数据库, 进行私有数据处理。 - ![GitHub Repo stars](https://img.shields.io/github/stars/csunny/db-gpt?style=social) +一个数据库相关的GPT实验项目, 模型与数据全部本地化部署, 绝对保障数据的隐私安全。 同时此GPT项目可以直接本地部署连接到私有数据库, 进行私有数据处理。 + [DB-GPT](https://github.com/csunny/DB-GPT) 是一个实验性的开源应用程序，它基于[FastChat](https://github.com/lm-sys/FastChat)，并使用[vicuna-13b](https://huggingface.co/Tribbiani/vicuna-13b)作为基础模型。此外，此程序结合了[langchain](https://github.com/hwchase17/langchain)和[llama-index](https://github.com/jerryjliu/llama_index)基于现有知识库进行[In-Context Learning](https://arxiv.org/abs/2301.00234)来对其进行数据库相关知识的增强。它可以进行SQL生成、SQL诊断、数据库知识问答等一系列的工作。 diff --git a/pilot/configs/model_config.py b/pilot/configs/model_config.py index 149ddb296..df0318e2d 100644 --- a/pilot/configs/model_config.py +++ b/pilot/configs/model_config.py @@ -22,7 +22,7 @@ LLM_MODEL_CONFIG = { } -VECTOR_SEARCH_TOP_K = 5 +VECTOR_SEARCH_TOP_K = 3 LLM_MODEL = "vicuna-13b" LIMIT_MODEL_CONCURRENCY = 5 MAX_POSITION_EMBEDDINGS = 2048 diff --git a/pilot/conversation.py b/pilot/conversation.py index 688a5c70d..2dc8df2b9 100644 --- a/pilot/conversation.py +++ b/pilot/conversation.py @@ -147,7 +147,7 @@ conv_vicuna_v1 = Conversation( ) -conv_qk_prompt_template = """ 基于以下已知的信息, 专业、详细的回答用户的问题。 +conv_qa_prompt_template = """ 基于以下已知的信息, 专业、详细的回答用户的问题。 如果无法从提供的恶内容中获取答案, 请说: "知识库中提供的内容不足以回答此问题", 但是你可以给出一些与问题相关答案的建议: 已知内容: @@ -158,6 +158,12 @@ conv_qk_prompt_template = """ 基于以下已知的信息, 专业、详细的回 default_conversation = conv_one_shot +conversation_types = { + "native": "LLM原生对话", + "default_knownledge": "默认知识库对话", + "custome": "新增知识库对话", +} + conv_templates = { "conv_one_shot": conv_one_shot, "vicuna_v1": conv_vicuna_v1, diff --git a/pilot/model/vicuna_llm.py b/pilot/model/vicuna_llm.py index f17a17a00..2337a3bbf 100644 --- a/pilot/model/vicuna_llm.py +++ b/pilot/model/vicuna_llm.py @@ -10,33 +10,29 @@ from typing import Any, Mapping, Optional, List from langchain.llms.base import LLM from pilot.configs.model_config import * -class VicunaRequestLLM(LLM): +class VicunaLLM(LLM): + + vicuna_generate_path = "generate_stream" + def _call(self, prompt: str, temperature: float, max_new_tokens: int, stop: Optional[List[str]] = None) -> str: - vicuna_generate_path = "generate" - def _call(self, prompt: str, stop: Optional[List[str]] = None) -> str: - if isinstance(stop, list): - stop = stop + ["Observation:"] - - skip_echo_len = len(prompt.replace("", " ")) + 1 params = { "prompt": prompt, - "temperature": 0.7, - "max_new_tokens": 1024, + "temperature": temperature, + "max_new_tokens": max_new_tokens, "stop": stop } response = requests.post( url=urljoin(VICUNA_MODEL_SERVER, self.vicuna_generate_path), data=json.dumps(params), ) - response.raise_for_status() - # for chunk in response.iter_lines(decode_unicode=False, delimiter=b"\0"): - # if chunk: - # data = json.loads(chunk.decode()) - # if data["error_code"] == 0: - # output = data["text"][skip_echo_len:].strip() - # output = self.post_process_code(output) - # yield output - return response.json()["response"] + + skip_echo_len = len(params["prompt"]) + 1 - params["prompt"].count("") * 3 + for chunk in response.iter_lines(decode_unicode=False, delimiter=b"\0"): + if chunk: + data = json.loads(chunk.decode()) + if data["error_code"] == 0: + output = data["text"][skip_echo_len:].strip() + yield output @property def _llm_type(self) -> str: diff --git a/pilot/server/embdserver.py b/pilot/server/embdserver.py index 5e0ad9294..6599a18ad 100644 --- a/pilot/server/embdserver.py +++ b/pilot/server/embdserver.py @@ -4,29 +4,44 @@ import requests import json import time +import uuid from urllib.parse import urljoin import gradio as gr from pilot.configs.model_config import * -vicuna_base_uri = "http://192.168.31.114:21002/" -vicuna_stream_path = "worker_generate_stream" -vicuna_status_path = "worker_get_status" +from pilot.conversation import conv_qa_prompt_template, conv_templates +from langchain.prompts import PromptTemplate -def generate(prompt): +vicuna_stream_path = "generate_stream" + +def generate(query): + + template_name = "conv_one_shot" + state = conv_templates[template_name].copy() + + pt = PromptTemplate( + template=conv_qa_prompt_template, + input_variables=["context", "question"] + ) + + result = pt.format(context="This page covers how to use the Chroma ecosystem within LangChain. It is broken into two parts: installation and setup, and then references to specific Chroma wrappers.", + question=query) + + print(result) + + state.append_message(state.roles[0], result) + state.append_message(state.roles[1], None) + + prompt = state.get_prompt() params = { "model": "vicuna-13b", "prompt": prompt, "temperature": 0.7, - "max_new_tokens": 512, + "max_new_tokens": 1024, "stop": "###" } - sts_response = requests.post( - url=urljoin(vicuna_base_uri, vicuna_status_path) - ) - print(sts_response.text) - response = requests.post( - url=urljoin(vicuna_base_uri, vicuna_stream_path), data=json.dumps(params) + url=urljoin(VICUNA_MODEL_SERVER, vicuna_stream_path), data=json.dumps(params) ) skip_echo_len = len(params["prompt"]) + 1 - params["prompt"].count("") * 3 @@ -34,11 +49,10 @@ def generate(prompt): if chunk: data = json.loads(chunk.decode()) if data["error_code"] == 0: - output = data["text"] + output = data["text"][skip_echo_len:].strip() + state.messages[-1][-1] = output + "▌" yield(output) - - time.sleep(0.02) - + if __name__ == "__main__": print(LLM_MODEL) with gr.Blocks() as demo: diff --git a/pilot/server/vectordb_qa.py b/pilot/server/vectordb_qa.py index 083ce20cd..71a9b881d 100644 --- a/pilot/server/vectordb_qa.py +++ b/pilot/server/vectordb_qa.py @@ -3,31 +3,27 @@ from pilot.vector_store.file_loader import KnownLedge2Vector from langchain.prompts import PromptTemplate -from pilot.conversation import conv_qk_prompt_template -from langchain.chains import RetrievalQA +from pilot.conversation import conv_qa_prompt_template from pilot.configs.model_config import VECTOR_SEARCH_TOP_K +from pilot.model.vicuna_llm import VicunaLLM class KnownLedgeBaseQA: - llm: object = None - def __init__(self) -> None: k2v = KnownLedge2Vector() self.vector_store = k2v.init_vector_store() + self.llm = VicunaLLM() - def get_answer(self, query): - prompt_template = conv_qk_prompt_template + def get_similar_answer(self, query): prompt = PromptTemplate( - template=prompt_template, + template=conv_qa_prompt_template, input_variables=["context", "question"] ) - knownledge_chain = RetrievalQA.from_llm( - llm=self.llm, - retriever=self.vector_store.as_retriever(search_kwargs={"k", VECTOR_SEARCH_TOP_K}), - prompt=prompt - ) - knownledge_chain.return_source_documents = True - result = knownledge_chain({"query": query}) - yield result + retriever = self.vector_store.as_retriever(search_kwargs={"k": VECTOR_SEARCH_TOP_K}) + docs = retriever.get_relevant_documents(query=query) + + context = [d.page_content for d in docs] + result = prompt.format(context="\n".join(context), question=query) + return result diff --git a/pilot/server/webserver.py b/pilot/server/webserver.py index dd6753a3b..b31bfde7f 100644 --- a/pilot/server/webserver.py +++ b/pilot/server/webserver.py @@ -11,6 +11,7 @@ import datetime import requests from urllib.parse import urljoin from pilot.configs.model_config import DB_SETTINGS +from pilot.server.vectordb_qa import KnownLedgeBaseQA from pilot.connections.mysql_conn import MySQLOperator from pilot.vector_store.extract_tovec import get_vector_storelist, load_knownledge_from_doc, knownledge_tovec_st @@ -19,6 +20,7 @@ from pilot.configs.model_config import LOGDIR, VICUNA_MODEL_SERVER, LLM_MODEL, D from pilot.conversation import ( default_conversation, conv_templates, + conversation_types, SeparatorStyle ) @@ -149,7 +151,7 @@ def post_process_code(code): code = sep.join(blocks) return code -def http_bot(state, db_selector, temperature, max_new_tokens, request: gr.Request): +def http_bot(state, mode, db_selector, temperature, max_new_tokens, request: gr.Request): start_tstamp = time.time() model_name = LLM_MODEL @@ -170,7 +172,8 @@ def http_bot(state, db_selector, temperature, max_new_tokens, request: gr.Reques query = state.messages[-2][1] - # prompt 中添加上下文提示 + # prompt 中添加上下文提示, 根据已有知识对话, 上下文提示是否也应该放在第一轮, 还是每一轮都添加上下文? + # 如果用户侧的问题跨度很大, 应该每一轮都加提示。 if db_selector: new_state.append_message(new_state.roles[0], gen_sqlgen_conversation(dbname) + query) new_state.append_message(new_state.roles[1], None) @@ -180,13 +183,11 @@ def http_bot(state, db_selector, temperature, max_new_tokens, request: gr.Reques new_state.append_message(new_state.roles[1], None) state = new_state - # try: - # if not db_selector: - # sim_q = get_simlar(query) - # print("********vector similar info*************: ", sim_q) - # state.append_message(new_state.roles[0], sim_q + query) - # except Exception as e: - # print(e) + if mode == conversation_types["default_knownledge"] and not db_selector: + query = state.messages[-2][1] + knqa = KnownLedgeBaseQA() + state.messages[-2][1] = knqa.get_similar_answer(query) + prompt = state.get_prompt() @@ -222,7 +223,7 @@ def http_bot(state, db_selector, temperature, max_new_tokens, request: gr.Reques state.messages[-1][-1] = output yield (state, state.to_gradio_chatbot()) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn) return - time.sleep(0.02) + except requests.exceptions.RequestException as e: state.messages[-1][-1] = server_error_msg + f" (error_code: 4)" yield (state, state.to_gradio_chatbot()) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn) @@ -231,6 +232,7 @@ def http_bot(state, db_selector, temperature, max_new_tokens, request: gr.Reques state.messages[-1][-1] = state.messages[-1][-1][:-1] yield (state, state.to_gradio_chatbot()) + (enable_btn,) * 5 + # 记录运行日志 finish_tstamp = time.time() logger.info(f"{output}") @@ -266,7 +268,7 @@ def change_tab(tab): pass def change_mode(mode): - if mode == "默认知识库对话": + if mode in ["默认知识库对话", "LLM原生对话"]: return gr.update(visible=False) else: return gr.update(visible=True) @@ -318,7 +320,8 @@ def build_single_model_ui(): show_label=True).style(container=False) with gr.TabItem("知识问答", elem_id="QA"): - mode = gr.Radio(["默认知识库对话", "新增知识库"], show_label=False, value="默认知识库对话") + + mode = gr.Radio(["LLM原生对话", "默认知识库对话", "新增知识库对话"], show_label=False, value="LLM原生对话") vs_setting = gr.Accordion("配置知识库", open=False) mode.change(fn=change_mode, inputs=mode, outputs=vs_setting) with vs_setting: @@ -363,7 +366,7 @@ def build_single_model_ui(): btn_list = [regenerate_btn, clear_btn] regenerate_btn.click(regenerate, state, [state, chatbot, textbox] + btn_list).then( http_bot, - [state, db_selector, temperature, max_output_tokens], + [state, mode, db_selector, temperature, max_output_tokens], [state, chatbot] + btn_list, ) clear_btn.click(clear_history, None, [state, chatbot, textbox] + btn_list) @@ -372,7 +375,7 @@ def build_single_model_ui(): add_text, [state, textbox], [state, chatbot, textbox] + btn_list ).then( http_bot, - [state, db_selector, temperature, max_output_tokens], + [state, mode, db_selector, temperature, max_output_tokens], [state, chatbot] + btn_list, ) @@ -380,7 +383,7 @@ def build_single_model_ui(): add_text, [state, textbox], [state, chatbot, textbox] + btn_list ).then( http_bot, - [state, db_selector, temperature, max_output_tokens], + [state, mode, db_selector, temperature, max_output_tokens], [state, chatbot] + btn_list )