diff --git a/src/blackbox/blackbox_factory.py b/src/blackbox/blackbox_factory.py index edb0579..7dd8754 100644 --- a/src/blackbox/blackbox_factory.py +++ b/src/blackbox/blackbox_factory.py @@ -1,21 +1,19 @@ -from . import melotts +from blackbox.emotion import Emotion +from .chat import Chat from .audio_chat import AudioChat from .sentiment import Sentiment from .tts import TTS from .asr import ASR from .audio_to_text import AudioToText -#from .emotion import Emotion from .blackbox import Blackbox -from .text_to_audio import TextToAudio -from .tesou import Tesou from .fastchat import Fastchat from .g2e import G2E from .text_and_image import TextAndImage -# from .chroma_query import ChromaQuery -# from .chroma_upsert import ChromaUpsert -# from .chroma_chat import ChromaChat from .melotts import MeloTTS from .vlms import VLMS +from .chroma_query import ChromaQuery +from .chroma_upsert import ChromaUpsert +from .chroma_chat import ChromaChat from injector import inject, singleton @singleton @@ -25,37 +23,35 @@ class BlackboxFactory: @inject def __init__(self, audio_to_text: AudioToText, - text_to_audio: TextToAudio, asr: ASR, tts: TTS, sentiment_engine: Sentiment, - #emotion: Emotion, - tesou: Tesou, + emotion: Emotion, fastchat: Fastchat, audio_chat: AudioChat, g2e: G2E, text_and_image: TextAndImage, - #chroma_query: ChromaQuery, - #chroma_upsert: ChromaUpsert, - #chroma_chat: ChromaChat, melotts: MeloTTS, - vlms: VLMS) -> None: + vlms: VLMS, + chroma_query: ChromaQuery, + chroma_upsert: ChromaUpsert, + chroma_chat: ChromaChat, + chat: Chat) -> None: self.models["audio_to_text"] = audio_to_text - self.models["text_to_audio"] = text_to_audio self.models["asr"] = asr self.models["tts"] = tts self.models["sentiment_engine"] = sentiment_engine - self.models["tesou"] = tesou - #self.models["emotion"] = emotion + self.models["emotion"] = emotion self.models["fastchat"] = fastchat self.models["audio_chat"] = audio_chat self.models["g2e"] = g2e self.models["text_and_image"] = text_and_image - #self.models["chroma_query"] = chroma_query - #self.models["chroma_upsert"] = chroma_upsert - #self.models["chroma_chat"] = chroma_chat + self.models["chroma_query"] = chroma_query + self.models["chroma_upsert"] = chroma_upsert + self.models["chroma_chat"] = chroma_chat self.models["melotts"] = melotts self.models["vlms"] = vlms + self.models["chat"] = chat def __call__(self, *args, **kwargs): return self.processing(*args, **kwargs) diff --git a/src/blackbox/chat.py b/src/blackbox/chat.py new file mode 100644 index 0000000..0ba8c68 --- /dev/null +++ b/src/blackbox/chat.py @@ -0,0 +1,128 @@ +import logging +from typing import Any, Coroutine + +from fastapi import Request, Response, status +from fastapi.responses import JSONResponse + +from ..log.logging_time import logging_time +from .blackbox import Blackbox + +import requests +import json +from openai import OpenAI +import re + +from injector import singleton + +logger = logging.getLogger + +@singleton +class Chat(Blackbox): + + def __call__(self, *args, **kwargs): + return self.processing(*args, **kwargs) + + def valid(self, *args, **kwargs) -> bool: + data = args[0] + return isinstance(data, list) + + # model_name有 Qwen1.5-14B-Chat , internlm2-chat-20b + @logging_time(logger=logger) + def processing(self, *args, **kwargs) -> str: + + settings: dict = args[0] + if settings is None: + settings = {} + user_model_name = settings.get("model_name") + user_context = settings.get("context") + user_question = settings.get("question") + user_template = settings.get("template") + user_temperature = settings.get("temperature") + user_top_p = settings.get("top_p") + user_n = settings.get("n") + user_max_tokens = settings.get("max_tokens") + user_stop = settings.get("stop") + user_frequency_penalty = settings.get("frequency_penalty") + user_presence_penalty = settings.get("presence_penalty") + + if user_context == None: + user_context = [] + + if user_question is None: + return JSONResponse(content={"error": "question is required"}, status_code=status.HTTP_400_BAD_REQUEST) + + if user_model_name is None or user_model_name.isspace() or user_model_name == "": + user_model_name = "Qwen1.5-14B-Chat" + + if user_template is None or user_template.isspace(): + user_template = "" + + if user_temperature is None or user_temperature == "": + user_temperature = 0.8 + + if user_top_p is None or user_top_p == "": + user_top_p = 0.8 + + if user_n is None or user_n == "": + user_n = 1 + + if user_max_tokens is None or user_max_tokens == "": + user_max_tokens = 1024 + + if user_stop is None or user_stop == "": + user_stop = 100 + + if user_frequency_penalty is None or user_frequency_penalty == "": + user_frequency_penalty = 0.5 + + if user_presence_penalty is None or user_presence_penalty == "": + user_presence_penalty = 0.8 + + + # gpt-4， gpt-3.5-turbo + if re.search(r"gpt", user_model_name): + url = 'https://api.openai.com/v1/completions' + key = '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' + else: + url = 'http://120.196.116.194:48892/v1/chat/completions' + key = 'YOUR_API_KEY' + + prompt_template = [ + {"role": "system", "content": user_template}, + ] + + chat_inputs={ + "model": user_model_name, + "messages": prompt_template + user_context + [ + { + "role": "user", + "content": user_question + } + ], + "temperature": user_temperature, + "top_p": user_top_p, + "n": user_n, + "max_tokens": user_max_tokens, + "frequency_penalty": user_frequency_penalty, + "presence_penalty": user_presence_penalty, + "stop": user_stop + } + + header = { + 'Content-Type': 'application/json', + 'Authorization': "Bearer " + key + } + + fastchat_response = requests.post(url, json=chat_inputs, headers=header) + + return fastchat_response.json()["choices"][0]["message"]["content"] + + async def fast_api_handler(self, request: Request) -> Response: + try: + data = await request.json() + except: + return JSONResponse(content={"error": "json parse error"}, status_code=status.HTTP_400_BAD_REQUEST) + + setting: dict = data.get("settings") + + return JSONResponse(content={"response": self.processing(setting)}, status_code=status.HTTP_200_OK) \ No newline at end of file diff --git a/src/blackbox/chroma_chat.py b/src/blackbox/chroma_chat.py index 560712c..01685f3 100755 --- a/src/blackbox/chroma_chat.py +++ b/src/blackbox/chroma_chat.py @@ -4,15 +4,18 @@ from fastapi import Request, Response, status from fastapi.responses import JSONResponse from .blackbox import Blackbox -from ..utils import chroma_setting +from .chat import Chat +from .chroma_query import ChromaQuery + DEFAULT_COLLECTION_ID = "123" -from injector import singleton +from injector import singleton,inject @singleton class ChromaChat(Blackbox): - def __init__(self, fastchat, chroma_query): - self.fastchat = fastchat + @inject + def __init__(self, chat: Chat, chroma_query: ChromaQuery): + self.chat = chat self.chroma_query = chroma_query def __call__(self, *args, **kwargs): @@ -22,18 +25,18 @@ class ChromaChat(Blackbox): data = args[0] return isinstance(data, list) - def processing(self, question, setting: chroma_setting) -> str: + def processing(self, question, context: list) -> str: + if context == None: + context = [] # load or create collection - if setting is None: - collection_id = DEFAULT_COLLECTION_ID - else: - collection_id = setting.ChromaSetting.collection_ids[0] + collection_id = DEFAULT_COLLECTION_ID + # query it chroma_result = self.chroma_query(question, collection_id) - fast_question = "问题: "+ question + "。根据问题，总结以下内容：" + chroma_result - response = self.fastchat(fast_question) + fast_question = "问题: "+ question + "。根据问题，总结以下内容和来源：" + chroma_result + response = self.chat(model_name="Qwen1.5-14B-Chat", prompt=fast_question, template='', context=context, temperature=0.8, top_p=0.8, n=1, max_tokens=1024, stop=100,frequency_penalty=0.5,presence_penalty=0.8) return response diff --git a/src/blackbox/chroma_query.py b/src/blackbox/chroma_query.py index 8597677..6e39eae 100755 --- a/src/blackbox/chroma_query.py +++ b/src/blackbox/chroma_query.py @@ -31,7 +31,7 @@ class ChromaQuery(Blackbox): def processing(self, question: str, collection_id) -> str: # load or create collection - collection = self.client.get_or_create_collection(collection_id, embedding_function=self.embedding_model) + collection = self.client.get_collection(collection_id, embedding_function=self.embedding_model) # query it results = collection.query( @@ -39,7 +39,7 @@ class ChromaQuery(Blackbox): n_results=3, ) - response = results["documents"] + results["metadatas"] + response = str(results["documents"] + results["metadatas"]) return response diff --git a/src/blackbox/chroma_upsert.py b/src/blackbox/chroma_upsert.py index 60607d5..c6e71f9 100755 --- a/src/blackbox/chroma_upsert.py +++ b/src/blackbox/chroma_upsert.py @@ -13,6 +13,9 @@ from langchain_community.vectorstores import Chroma from langchain.text_splitter import RecursiveCharacterTextSplitter, CharacterTextSplitter from langchain_community.embeddings.sentence_transformer import SentenceTransformerEmbeddings import chromadb + +import os +import tempfile from ..utils import chroma_setting @@ -50,7 +53,7 @@ class ChromaUpsert(Blackbox): "collection_id": "123", "action": "upsert", "content": "file_name or string", - "answer": "success, collection has 100 documents.", + "answer": "collection 123 has 12472 documents. /tmp/Cheap and Quick：Efficient Vision-Language Instruction Tuning for Large Language Models.pdf ids is 0~111", }, ] @@ -61,6 +64,7 @@ class ChromaUpsert(Blackbox): if file is not None: file_type = file.split(".")[-1] + print("file_type: ",file_type) if file_type == "pdf": loader = PyPDFLoader(file) elif file_type == "txt": @@ -77,7 +81,6 @@ class ChromaUpsert(Blackbox): loader = UnstructuredExcelLoader(file) - loader = PyPDFLoader(file) documents = loader.load() text_splitter = RecursiveCharacterTextSplitter(chunk_size=512, chunk_overlap=0) @@ -86,39 +89,58 @@ class ChromaUpsert(Blackbox): ids = [str(file)+str(i) for i in range(len(docs))] Chroma.from_documents(documents=docs, embedding=self.embedding_model, ids=ids, collection_name=collection_id, client=self.client) + + collection_number = self.client.get_collection(collection_id).count() + response_file = f"collection {collection_id} has {collection_number} documents. {file} ids is 0~{len(docs)-1}" if string is not None: # 生成一个新的id ids_string: 1 - ids = setting.ChromaSetting.string_ids[0] + 1 + # ids = setting.ChromaSetting.string_ids[0] + 1 + ids = "1" Chroma.from_texts(texts=[string], embedding=self.embedding_model, ids=[ids], collection_name=collection_id, client=self.client) - collection_number = self.client.get_collection(collection_id).count() - response = f"collection {collection_id} has {collection_number} documents." + collection_number = self.client.get_collection(collection_id).count() + response_string = f"collection {collection_id} has {collection_number} documents. {string} ids is {ids}" - return response + + if file is not None and string is not None: + return response_file + " \n and " + response_string + elif file is not None and string is None: + return response_file + elif file is None and string is not None: + return response_string async def fast_api_handler(self, request: Request) -> Response: - try: - data = await request.json() - except: - return JSONResponse(content={"error": "json parse error"}, status_code=status.HTTP_400_BAD_REQUEST) - - user_collection_id = data.get("collection_id") - user_file = data.get("file") - user_string = data.get("string") - user_context = data.get("context") - user_setting = data.get("setting") + + user_collection_id = (await request.form()).get("collection_id") + user_file = (await request.form()).get("file") + user_string = (await request.form()).get("string") + user_context = (await request.form()).get("context") + user_setting = (await request.form()).get("setting") if user_collection_id is None and user_setting["collections"] == []: return JSONResponse(content={"error": "The first creation requires a collection id"}, status_code=status.HTTP_400_BAD_REQUEST) if user_file is None and user_string is None: return JSONResponse(content={"error": "file or string is required"}, status_code=status.HTTP_400_BAD_REQUEST) + + if user_file is not None: + pdf_bytes = await user_file.read() + + custom_filename = user_file.filename + # 获取系统的临时目录路径 + safe_filename = os.path.join(tempfile.gettempdir(), os.path.basename(custom_filename)) + with open(safe_filename, "wb") as f: + f.write(pdf_bytes) + else: + safe_filename = None + + return JSONResponse( - content={"response": self.processing(user_collection_id, user_file, user_string, user_context, user_setting)}, + content={"response": self.processing(user_collection_id, safe_filename, user_string, user_context, user_setting)}, status_code=status.HTTP_200_OK) \ No newline at end of file diff --git a/src/blackbox/fastchat.py b/src/blackbox/fastchat.py index 27c95c1..3144d5e 100755 --- a/src/blackbox/fastchat.py +++ b/src/blackbox/fastchat.py @@ -19,40 +19,68 @@ class Fastchat(Blackbox): return isinstance(data, list) # model_name有 Qwen1.5-14B-Chat , internlm2-chat-20b - def processing(self, model_name, prompt, template, context: list) -> str: + def processing(self, model_name, prompt, template, context: list, temperature, top_p, top_k, n, max_tokens) -> str: if context == None: context = [] url = 'http://120.196.116.194:48892/v1/chat/completions' - # history可以为空列表，也可以是用户的对话历史 - # history = [ + # context可以为空列表，也可以是用户的对话历史 + # context = [ # { # "role": "user", - # "content": "你吃饭了吗" + # "content": "智能体核心思想" # }, # { # "role": "assistant", - # "content": "作为一个AI模型，我没有吃饭的需要，因为我并不具备实体形态。我专注于提供信息和帮助回答你的问题。你有什么需要帮助的吗？" + # "content": "智能体的核心思想是将人工智能应用于问题求解者角色，它通过算法模拟人类决策过程，通过感知环境、学习、规划和执行行动，以实现特定任务或目标。其目标是通过自我适应和优化，实现高效问题解决。" # }, # ] + prompt_template = [ + {"role": "system", "content": template}, + ] + fastchat_inputs={ "model": model_name, - "messages": context + [ + "messages": prompt_template + context + [ { "role": "user", - "content": template + prompt + "content": prompt } - ] + ], + "temperature": temperature, + "top_p": top_p, + "top_k": top_k, + "n": n, + "max_tokens": max_tokens, + "stream": False, } + + # { + # "model": "string", + # "messages": "string", + # "temperature": 0.7, # between 0 and 2. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic. + # "top_p": 1, # 控制生成下一个单词的概率分布，即从所有可能的单词中，只选择概率最高的一部分作为候选单词 + # "top_k": -1, # top-k 参数设置为 3意味着选择前三个tokens。 + # "n": 1, # How many chat completion choices to generate for each input message. + # "max_tokens": 1024, # The maximum number of tokens to generate in the chat completion. + # "stop": [ + # "string" + # ], + # "stream": False, + # "presence_penalty": 0, # Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. + # "frequency_penalty": 0, # Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model’s likelihood to repeat the same line verbatim. + # "user": "string" + # } + fastchat_response = requests.post(url, json=fastchat_inputs) - user_message = fastchat_inputs["messages"] - context.append(user_message) + # user_message = fastchat_inputs["messages"] + # context.append(user_message) assistant_message = fastchat_response.json()["choices"][0]["message"] - context.append(assistant_message) + # context.append(assistant_message) fastchat_content = assistant_message["content"] @@ -66,19 +94,40 @@ class Fastchat(Blackbox): user_model_name = data.get("model_name") user_context = data.get("context") - user_prompt = data.get("prompt") + user_question = data.get("question") user_template = data.get("template") + user_temperature = data.get("temperature") + user_top_p = data.get("top_p") + user_top_k = data.get("top_k") + user_n = data.get("n") + user_max_tokens = data.get("max_tokens") + - if user_prompt is None: + if user_question is None: return JSONResponse(content={"error": "question is required"}, status_code=status.HTTP_400_BAD_REQUEST) - if user_model_name is None or user_model_name.isspace(): + if user_model_name is None or user_model_name.isspace() or user_model_name == "": user_model_name = "Qwen1.5-14B-Chat" if user_template is None or user_template.isspace(): - # user_template 是定义LLM的语气，例如template = "使用小丑的语气说话。"，user_template可以为空字串，或者是用户自定义的语气，或者是使用我们提供的语气 + # user_template 是定义LLM的语气，例如template = "使用小丑的语气说话。"，user_template可以为空字串，或者是用户自定义的语气 user_template = "" - else: - user_template = f"使用{user_template}的语气说话。" + + if user_temperature is None or user_temperature == "": + user_temperature = 0.7 - return JSONResponse(content={"response": self.processing(user_model_name, user_prompt, user_template, user_context)}, status_code=status.HTTP_200_OK) \ No newline at end of file + if user_top_p is None or user_top_p == "": + user_top_p = 1 + + if user_top_k is None or user_top_k == "": + user_top_k = -1 + + if user_n is None or user_n == "": + user_n = 1 + + if user_max_tokens is None or user_max_tokens == "": + user_max_tokens = 1024 + + + return JSONResponse(content={"response": self.processing(user_model_name, user_question, user_template, user_context, + user_temperature, user_top_p, user_top_k, user_n, user_max_tokens)}, status_code=status.HTTP_200_OK) \ No newline at end of file diff --git a/src/blackbox/modelscope.py b/src/blackbox/modelscope.py new file mode 100755 index 0000000..9e5e61d --- /dev/null +++ b/src/blackbox/modelscope.py @@ -0,0 +1,145 @@ +from typing import Any, Coroutine + +from fastapi import Request, Response, status +from fastapi.responses import JSONResponse +from .blackbox import Blackbox + +import requests +import json + +from modelscope_agent.agents import RolePlay +from modelscope_agent.tools.base import BaseTool +from modelscope_agent.tools import register_tool + +class Modelscope(Blackbox): + + def __call__(self, *args, **kwargs): + return self.processing(*args, **kwargs) + + def valid(self, *args, **kwargs) -> bool: + data = args[0] + return isinstance(data, list) + + # model_name有 Qwen1.5-14B-Chat , internlm2-chat-20b + def processing(self, model_name, prompt, template, context: list) -> str: + if context == None: + context = [] + + @register_tool('ChromaQuery') + class AliyunRenewInstanceTool(BaseTool): + description = '查询chroma数据库中的数据' + name = 'ChromaQuery' + parameters: list = [{ + 'name': 'id', + 'description': '用户的chroma id', + 'required': True, + 'type': 'string' + }, { + 'name': 'query', + 'description': '用户需要在chroma中查询的问题', + 'required': True, + 'type': 'string' + }] + + def call(self, params: str, **kwargs): + params = self._verify_args(params) + id = params['id'] + query = params['query'] + query_data = { + "chroma_query_data": { + "id": id, + "question": query + } + } + url = "http://10.6.80.75:7003" + response = requests.post(f"{url}/api/chroma_query", json=query_data) + result = response.json()['response'] + return str({'result': f'Chroma ID为{id}的用户，查询结果为{response}。'}) + + + @register_tool('WebSearch') + class WebSearchTool(BaseTool): + description = '查询网络中的内容' + name = 'WebSearch' + parameters: list = [ { + 'name': 'search_term', + 'description': '用户需要在Web中查询的问题', + 'required': True, + 'type': 'string' + }] + + def call(self, params: str, **kwargs): + params = self._verify_args(params) + search_term = params['search_term'] + + api_key='9e51be0aaecb5a56fe2faead6e2c702fde92e62a' + headers = { + 'X-API-KEY': api_key, + 'Content-Type': 'application/json', + } + params = { + 'q': search_term + } + try: + response = requests.post( + f'https://google.serper.dev/search', + headers=headers, + params=params, + timeout=5) + except Exception as e: + return -1, str(e) + + result = response.json()['answerBox']['snippet'] + + return str({'result': f'WebSearch查询结果为{search_term}{result}。'}) + + + # define LLM + api_base_url = "http://120.196.116.194:48892/v1" + api_key= "EMPTY" + LLM_MODEL = model_name + + llm_config = { + 'model': LLM_MODEL, + 'model_server': 'openai', + 'api_base':api_base_url, + 'api_key': api_key + } + + function_list = ['WebSearch', 'ChromaQuery'] + + bot = RolePlay(function_list=function_list,llm=llm_config, instruction=template) + + response = bot.run(prompt, history=context, lang='zh') + + text = '' + for chunk in response: + text += chunk + + return text + + + async def fast_api_handler(self, request: Request) -> Response: + try: + data = await request.json() + except: + return JSONResponse(content={"error": "json parse error"}, status_code=status.HTTP_400_BAD_REQUEST) + + user_model_name = data.get("model_name") + user_context = data.get("context") + user_prompt = data.get("prompt") + user_template = data.get("template") + + if user_prompt is None: + return JSONResponse(content={"error": "question is required"}, status_code=status.HTTP_400_BAD_REQUEST) + + if user_model_name is None or user_model_name.isspace(): + user_model_name = "Qwen1.5-14B-Chat" + + if user_template is None or user_template.isspace(): + # user_template 是定义LLM的语气，例如template = "使用小丑的语气说话。"，user_template可以为空字串，或者是用户自定义的语气，或者是使用我们提供的语气 + user_template = "" + else: + user_template = f"使用{user_template}的语气说话。" + + return JSONResponse(content={"response": self.processing(user_model_name, user_prompt, user_template, user_context)}, status_code=status.HTTP_200_OK) \ No newline at end of file diff --git a/swagger.yml b/swagger.yml index 75adb7c..ba0247d 100644 --- a/swagger.yml +++ b/swagger.yml @@ -80,13 +80,10 @@ components: type: string description: "Blackbox name" enum: - - text_to_audio - - audio_to_text - asr - tts - sentiment_engine - emotion - - tesou - fastchat - audio_chat - g2e diff --git a/test.pdf b/test.pdf new file mode 100644 index 0000000..e69de29