diff --git a/README.md b/README.md index 81a8228..129e556 100644 --- a/README.md +++ b/README.md @@ -167,6 +167,14 @@ git clone --depth=1 https://github.com/OpenLMLab/MOSS.git request_llms/moss # # 【可选步骤IV】确保config.py配置文件的AVAIL_LLM_MODELS包含了期望的模型，目前支持的全部模型如下(jittorllms系列目前仅支持docker方案)： AVAIL_LLM_MODELS = ["gpt-3.5-turbo", "api2d-gpt-3.5-turbo", "gpt-4", "api2d-gpt-4", "chatglm", "moss"] # + ["jittorllms_rwkv", "jittorllms_pangualpha", "jittorllms_llama"] + +# 【可选步骤V】支持本地模型INT8,INT4量化（这里所指的模型本身不是量化版本，目前deepseek-coder支持，后面测试后会加入更多模型量化选择） +pip install bitsandbyte +# windows用户安装bitsandbytes需要使用下面bitsandbytes-windows-webui +python -m pip install bitsandbytes --prefer-binary --extra-index-url=https://jllllll.github.io/bitsandbytes-windows-webui +pip install -U git+https://github.com/huggingface/transformers.git +pip install -U git+https://github.com/huggingface/accelerate.git +pip install peft ```

diff --git a/check_proxy.py b/check_proxy.py index e2ba3f1..2df8185 100644 --- a/check_proxy.py +++ b/check_proxy.py @@ -159,7 +159,15 @@ def warm_up_modules(): enc.encode("模块预热", disallowed_special=()) enc = model_info["gpt-4"]['tokenizer'] enc.encode("模块预热", disallowed_special=()) + +def warm_up_vectordb(): + print('正在执行一些模块的预热 ...') + from toolbox import ProxyNetworkActivate + with ProxyNetworkActivate("Warmup_Modules"): + import nltk + with ProxyNetworkActivate("Warmup_Modules"): nltk.download("punkt") + if __name__ == '__main__': import os os.environ['no_proxy'] = '*' # 避免代理网络产生意外污染 diff --git a/config.py b/config.py index a511724..3d80962 100644 --- a/config.py +++ b/config.py @@ -120,7 +120,6 @@ CHATGLM_PTUNING_CHECKPOINT = "" # 例如"/home/hmp/ChatGLM2-6B/ptuning/output/6b LOCAL_MODEL_DEVICE = "cpu" # 可选 "cuda" LOCAL_MODEL_QUANT = "FP16" # 默认 "FP16" "INT4" 启用量化INT4版本 "INT8" 启用量化INT8版本 - # 设置gradio的并行线程数（不需要修改） CONCURRENT_COUNT = 100 @@ -238,6 +237,10 @@ WHEN_TO_USE_PROXY = ["Download_LLM", "Download_Gradio_Theme", "Connect_Grobid", BLOCK_INVALID_APIKEY = False +# 启用插件热加载 +PLUGIN_HOT_RELOAD = False + + # 自定义按钮的最大数量限制 NUM_CUSTOM_BASIC_BTN = 4 diff --git a/crazy_functional.py b/crazy_functional.py index ef78b5a..7c2bf79 100644 --- a/crazy_functional.py +++ b/crazy_functional.py @@ -440,7 +440,7 @@ def get_crazy_functions(): print('Load function plugin failed') try: - from crazy_functions.Langchain知识库 import 知识库问答 + from crazy_functions.知识库问答 import 知识库文件注入 function_plugins.update({ "构建知识库（先上传文件素材,再运行此插件）": { "Group": "对话", @@ -448,7 +448,7 @@ def get_crazy_functions(): "AsButton": False, "AdvancedArgs": True, "ArgsReminder": "此处待注入的知识库名称id, 默认为default。文件进入知识库后可长期保存。可以通过再次调用本插件的方式，向知识库追加更多文档。", - "Function": HotReload(知识库问答) + "Function": HotReload(知识库文件注入) } }) except: @@ -456,9 +456,9 @@ def get_crazy_functions(): print('Load function plugin failed') try: - from crazy_functions.Langchain知识库 import 读取知识库作答 + from crazy_functions.知识库问答 import 读取知识库作答 function_plugins.update({ - "知识库问答（构建知识库后,再运行此插件）": { + "知识库文件注入（构建知识库后,再运行此插件）": { "Group": "对话", "Color": "stop", "AsButton": False, @@ -590,6 +590,20 @@ def get_crazy_functions(): print(trimmed_format_exc()) print('Load function plugin failed') + # try: + # from crazy_functions.互动小游戏 import 随机小游戏 + # function_plugins.update({ + # "随机小游戏": { + # "Group": "智能体", + # "Color": "stop", + # "AsButton": True, + # "Function": HotReload(随机小游戏) + # } + # }) + # except: + # print(trimmed_format_exc()) + # print('Load function plugin failed') + # try: # from crazy_functions.chatglm微调工具 import 微调数据集生成 # function_plugins.update({ diff --git a/crazy_functions/crazy_utils.py b/crazy_functions/crazy_utils.py index afe079f..9778053 100644 --- a/crazy_functions/crazy_utils.py +++ b/crazy_functions/crazy_utils.py @@ -1,4 +1,4 @@ -from toolbox import update_ui, get_conf, trimmed_format_exc, get_max_token +from toolbox import update_ui, get_conf, trimmed_format_exc, get_max_token, Singleton import threading import os import logging @@ -631,89 +631,6 @@ def get_files_from_everything(txt, type): # type='.md' - -def Singleton(cls): - _instance = {} - - def _singleton(*args, **kargs): - if cls not in _instance: - _instance[cls] = cls(*args, **kargs) - return _instance[cls] - - return _singleton - - -@Singleton -class knowledge_archive_interface(): - def __init__(self) -> None: - self.threadLock = threading.Lock() - self.current_id = "" - self.kai_path = None - self.qa_handle = None - self.text2vec_large_chinese = None - - def get_chinese_text2vec(self): - if self.text2vec_large_chinese is None: - # < -------------------预热文本向量化模组--------------- > - from toolbox import ProxyNetworkActivate - print('Checking Text2vec ...') - from langchain.embeddings.huggingface import HuggingFaceEmbeddings - with ProxyNetworkActivate('Download_LLM'): # 临时地激活代理网络 - self.text2vec_large_chinese = HuggingFaceEmbeddings(model_name="GanymedeNil/text2vec-large-chinese") - - return self.text2vec_large_chinese - - - def feed_archive(self, file_manifest, id="default"): - self.threadLock.acquire() - # import uuid - self.current_id = id - from zh_langchain import construct_vector_store - self.qa_handle, self.kai_path = construct_vector_store( - vs_id=self.current_id, - files=file_manifest, - sentence_size=100, - history=[], - one_conent="", - one_content_segmentation="", - text2vec = self.get_chinese_text2vec(), - ) - self.threadLock.release() - - def get_current_archive_id(self): - return self.current_id - - def get_loaded_file(self): - return self.qa_handle.get_loaded_file() - - def answer_with_archive_by_id(self, txt, id): - self.threadLock.acquire() - if not self.current_id == id: - self.current_id = id - from zh_langchain import construct_vector_store - self.qa_handle, self.kai_path = construct_vector_store( - vs_id=self.current_id, - files=[], - sentence_size=100, - history=[], - one_conent="", - one_content_segmentation="", - text2vec = self.get_chinese_text2vec(), - ) - VECTOR_SEARCH_SCORE_THRESHOLD = 0 - VECTOR_SEARCH_TOP_K = 4 - CHUNK_SIZE = 512 - resp, prompt = self.qa_handle.get_knowledge_based_conent_test( - query = txt, - vs_path = self.kai_path, - score_threshold=VECTOR_SEARCH_SCORE_THRESHOLD, - vector_search_top_k=VECTOR_SEARCH_TOP_K, - chunk_conent=True, - chunk_size=CHUNK_SIZE, - text2vec = self.get_chinese_text2vec(), - ) - self.threadLock.release() - return resp, prompt @Singleton class nougat_interface(): diff --git a/crazy_functions/game_fns/game_utils.py b/crazy_functions/game_fns/game_utils.py new file mode 100644 index 0000000..09b6f7a --- /dev/null +++ b/crazy_functions/game_fns/game_utils.py @@ -0,0 +1,35 @@ + +from crazy_functions.json_fns.pydantic_io import GptJsonIO, JsonStringError +from request_llms.bridge_all import predict_no_ui_long_connection +def get_code_block(reply): + import re + pattern = r"```([\s\S]*?)```" # regex pattern to match code blocks + matches = re.findall(pattern, reply) # find all code blocks in text + if len(matches) == 1: + return "```" + matches[0] + "```" # code block + raise RuntimeError("GPT is not generating proper code.") + +def is_same_thing(a, b, llm_kwargs): + from pydantic import BaseModel, Field + class IsSameThing(BaseModel): + is_same_thing: bool = Field(description="determine whether two objects are same thing.", default=False) + + def run_gpt_fn(inputs, sys_prompt, history=[]): + return predict_no_ui_long_connection( + inputs=inputs, llm_kwargs=llm_kwargs, + history=history, sys_prompt=sys_prompt, observe_window=[] + ) + + gpt_json_io = GptJsonIO(IsSameThing) + inputs_01 = "Identity whether the user input and the target is the same thing: \n target object: {a} \n user input object: {b} \n\n\n".format(a=a, b=b) + inputs_01 += "\n\n\n Note that the user may describe the target object with a different language, e.g. cat and 猫 are the same thing." + analyze_res_cot_01 = run_gpt_fn(inputs_01, "", []) + + inputs_02 = inputs_01 + gpt_json_io.format_instructions + analyze_res = run_gpt_fn(inputs_02, "", [inputs_01, analyze_res_cot_01]) + + try: + res = gpt_json_io.generate_output_auto_repair(analyze_res, run_gpt_fn) + return res.is_same_thing + except JsonStringError as e: + return False \ No newline at end of file diff --git a/crazy_functions/multi_stage/multi_stage_utils.py b/crazy_functions/multi_stage/multi_stage_utils.py index f85d35a..1395e79 100644 --- a/crazy_functions/multi_stage/multi_stage_utils.py +++ b/crazy_functions/multi_stage/multi_stage_utils.py @@ -1,6 +1,7 @@ from pydantic import BaseModel, Field from typing import List from toolbox import update_ui_lastest_msg, disable_auto_promotion +from toolbox import CatchException, update_ui, get_conf, select_api_key, get_log_folder from request_llms.bridge_all import predict_no_ui_long_connection from crazy_functions.json_fns.pydantic_io import GptJsonIO, JsonStringError import time @@ -36,6 +37,57 @@ class GptAcademicState(): state.chatbot = chatbot return state -class GatherMaterials(): - def __init__(self, materials) -> None: - materials = ['image', 'prompt'] \ No newline at end of file + +class GptAcademicGameBaseState(): + """ + 1. first init: __init__ -> + """ + def init_game(self, chatbot, lock_plugin): + self.plugin_name = None + self.callback_fn = None + self.delete_game = False + self.step_cnt = 0 + + def lock_plugin(self, chatbot): + if self.callback_fn is None: + raise ValueError("callback_fn is None") + chatbot._cookies['lock_plugin'] = self.callback_fn + self.dump_state(chatbot) + + def get_plugin_name(self): + if self.plugin_name is None: + raise ValueError("plugin_name is None") + return self.plugin_name + + def dump_state(self, chatbot): + chatbot._cookies[f'plugin_state/{self.get_plugin_name()}'] = pickle.dumps(self) + + def set_state(self, chatbot, key, value): + setattr(self, key, value) + chatbot._cookies[f'plugin_state/{self.get_plugin_name()}'] = pickle.dumps(self) + + @staticmethod + def sync_state(chatbot, llm_kwargs, cls, plugin_name, callback_fn, lock_plugin=True): + state = chatbot._cookies.get(f'plugin_state/{plugin_name}', None) + if state is not None: + state = pickle.loads(state) + else: + state = cls() + state.init_game(chatbot, lock_plugin) + state.plugin_name = plugin_name + state.llm_kwargs = llm_kwargs + state.chatbot = chatbot + state.callback_fn = callback_fn + return state + + def continue_game(self, prompt, chatbot, history): + # 游戏主体 + yield from self.step(prompt, chatbot, history) + self.step_cnt += 1 + # 保存状态，收尾 + self.dump_state(chatbot) + # 如果游戏结束，清理 + if self.delete_game: + chatbot._cookies['lock_plugin'] = None + chatbot._cookies[f'plugin_state/{self.get_plugin_name()}'] = None + yield from update_ui(chatbot=chatbot, history=history) diff --git a/crazy_functions/vector_fns/__init__.py b/crazy_functions/vector_fns/__init__.py new file mode 100644 index 0000000..e69de29 diff --git a/crazy_functions/vector_fns/general_file_loader.py b/crazy_functions/vector_fns/general_file_loader.py new file mode 100644 index 0000000..a512c48 --- /dev/null +++ b/crazy_functions/vector_fns/general_file_loader.py @@ -0,0 +1,70 @@ +# From project chatglm-langchain + + +from langchain.document_loaders import UnstructuredFileLoader +from langchain.text_splitter import CharacterTextSplitter +import re +from typing import List + +class ChineseTextSplitter(CharacterTextSplitter): + def __init__(self, pdf: bool = False, sentence_size: int = None, **kwargs): + super().__init__(**kwargs) + self.pdf = pdf + self.sentence_size = sentence_size + + def split_text1(self, text: str) -> List[str]: + if self.pdf: + text = re.sub(r"\n{3,}", "\n", text) + text = re.sub('\s', ' ', text) + text = text.replace("\n\n", "") + sent_sep_pattern = re.compile('([﹒﹔﹖﹗．。！？]["’”」』]{0,2}|(?=["‘“「『]{1,2}|$))') # del ：； + sent_list = [] + for ele in sent_sep_pattern.split(text): + if sent_sep_pattern.match(ele) and sent_list: + sent_list[-1] += ele + elif ele: + sent_list.append(ele) + return sent_list + + def split_text(self, text: str) -> List[str]: ##此处需要进一步优化逻辑 + if self.pdf: + text = re.sub(r"\n{3,}", r"\n", text) + text = re.sub('\s', " ", text) + text = re.sub("\n\n", "", text) + + text = re.sub(r'([;；.!?。！？\?])([^”’])', r"\1\n\2", text) # 单字符断句符 + text = re.sub(r'(\.{6})([^"’”」』])', r"\1\n\2", text) # 英文省略号 + text = re.sub(r'(\…{2})([^"’”」』])', r"\1\n\2", text) # 中文省略号 + text = re.sub(r'([;；!?。！？\?]["’”」』]{0,2})([^;；!?，。！？\?])', r'\1\n\2', text) + # 如果双引号前有终止符，那么双引号才是句子的终点，把分句符\n放到双引号后，注意前面的几句都小心保留了双引号 + text = text.rstrip() # 段尾如果有多余的\n就去掉它 + # 很多规则中会考虑分号;，但是这里我把它忽略不计，破折号、英文双引号等同样忽略，需要的再做些简单调整即可。 + ls = [i for i in text.split("\n") if i] + for ele in ls: + if len(ele) > self.sentence_size: + ele1 = re.sub(r'([,，.]["’”」』]{0,2})([^,，.])', r'\1\n\2', ele) + ele1_ls = ele1.split("\n") + for ele_ele1 in ele1_ls: + if len(ele_ele1) > self.sentence_size: + ele_ele2 = re.sub(r'([\n]{1,}| {2,}["’”」』]{0,2})([^\s])', r'\1\n\2', ele_ele1) + ele2_ls = ele_ele2.split("\n") + for ele_ele2 in ele2_ls: + if len(ele_ele2) > self.sentence_size: + ele_ele3 = re.sub('( ["’”」』]{0,2})([^ ])', r'\1\n\2', ele_ele2) + ele2_id = ele2_ls.index(ele_ele2) + ele2_ls = ele2_ls[:ele2_id] + [i for i in ele_ele3.split("\n") if i] + ele2_ls[ + ele2_id + 1:] + ele_id = ele1_ls.index(ele_ele1) + ele1_ls = ele1_ls[:ele_id] + [i for i in ele2_ls if i] + ele1_ls[ele_id + 1:] + + id = ls.index(ele) + ls = ls[:id] + [i for i in ele1_ls if i] + ls[id + 1:] + return ls + +def load_file(filepath, sentence_size): + loader = UnstructuredFileLoader(filepath, mode="elements") + textsplitter = ChineseTextSplitter(pdf=False, sentence_size=sentence_size) + docs = loader.load_and_split(text_splitter=textsplitter) + # write_check_file(filepath, docs) + return docs + diff --git a/crazy_functions/vector_fns/vector_database.py b/crazy_functions/vector_fns/vector_database.py new file mode 100644 index 0000000..cffa22c --- /dev/null +++ b/crazy_functions/vector_fns/vector_database.py @@ -0,0 +1,338 @@ +# From project chatglm-langchain + +import threading +from toolbox import Singleton +import os +import shutil +import os +import uuid +import tqdm +from langchain.vectorstores import FAISS +from langchain.docstore.document import Document +from typing import List, Tuple +import numpy as np +from crazy_functions.vector_fns.general_file_loader import load_file + +embedding_model_dict = { + "ernie-tiny": "nghuyong/ernie-3.0-nano-zh", + "ernie-base": "nghuyong/ernie-3.0-base-zh", + "text2vec-base": "shibing624/text2vec-base-chinese", + "text2vec": "GanymedeNil/text2vec-large-chinese", +} + +# Embedding model name +EMBEDDING_MODEL = "text2vec" + +# Embedding running device +EMBEDDING_DEVICE = "cpu" + +# 基于上下文的prompt模版，请务必保留"{question}"和"{context}" +PROMPT_TEMPLATE = """已知信息： +{context} + +根据上述已知信息，简洁和专业的来回答用户的问题。如果无法从中得到答案，请说 “根据已知信息无法回答该问题” 或 “没有提供足够的相关信息”，不允许在答案中添加编造成分，答案请使用中文。 问题是：{question}""" + +# 文本分句长度 +SENTENCE_SIZE = 100 + +# 匹配后单段上下文长度 +CHUNK_SIZE = 250 + +# LLM input history length +LLM_HISTORY_LEN = 3 + +# return top-k text chunk from vector store +VECTOR_SEARCH_TOP_K = 5 + +# 知识检索内容相关度 Score, 数值范围约为0-1100，如果为0，则不生效，经测试设置为小于500时，匹配结果更精准 +VECTOR_SEARCH_SCORE_THRESHOLD = 0 + +NLTK_DATA_PATH = os.path.join(os.path.dirname(os.path.dirname(__file__)), "nltk_data") + +FLAG_USER_NAME = uuid.uuid4().hex + +# 是否开启跨域，默认为False，如果需要开启，请设置为True +# is open cross domain +OPEN_CROSS_DOMAIN = False + +def similarity_search_with_score_by_vector( + self, embedding: List[float], k: int = 4 +) -> List[Tuple[Document, float]]: + + def seperate_list(ls: List[int]) -> List[List[int]]: + lists = [] + ls1 = [ls[0]] + for i in range(1, len(ls)): + if ls[i - 1] + 1 == ls[i]: + ls1.append(ls[i]) + else: + lists.append(ls1) + ls1 = [ls[i]] + lists.append(ls1) + return lists + + scores, indices = self.index.search(np.array([embedding], dtype=np.float32), k) + docs = [] + id_set = set() + store_len = len(self.index_to_docstore_id) + for j, i in enumerate(indices[0]): + if i == -1 or 0 < self.score_threshold < scores[0][j]: + # This happens when not enough docs are returned. + continue + _id = self.index_to_docstore_id[i] + doc = self.docstore.search(_id) + if not self.chunk_conent: + if not isinstance(doc, Document): + raise ValueError(f"Could not find document for id {_id}, got {doc}") + doc.metadata["score"] = int(scores[0][j]) + docs.append(doc) + continue + id_set.add(i) + docs_len = len(doc.page_content) + for k in range(1, max(i, store_len - i)): + break_flag = False + for l in [i + k, i - k]: + if 0 <= l < len(self.index_to_docstore_id): + _id0 = self.index_to_docstore_id[l] + doc0 = self.docstore.search(_id0) + if docs_len + len(doc0.page_content) > self.chunk_size: + break_flag = True + break + elif doc0.metadata["source"] == doc.metadata["source"]: + docs_len += len(doc0.page_content) + id_set.add(l) + if break_flag: + break + if not self.chunk_conent: + return docs + if len(id_set) == 0 and self.score_threshold > 0: + return [] + id_list = sorted(list(id_set)) + id_lists = seperate_list(id_list) + for id_seq in id_lists: + for id in id_seq: + if id == id_seq[0]: + _id = self.index_to_docstore_id[id] + doc = self.docstore.search(_id) + else: + _id0 = self.index_to_docstore_id[id] + doc0 = self.docstore.search(_id0) + doc.page_content += " " + doc0.page_content + if not isinstance(doc, Document): + raise ValueError(f"Could not find document for id {_id}, got {doc}") + doc_score = min([scores[0][id] for id in [indices[0].tolist().index(i) for i in id_seq if i in indices[0]]]) + doc.metadata["score"] = int(doc_score) + docs.append(doc) + return docs + + +class LocalDocQA: + llm: object = None + embeddings: object = None + top_k: int = VECTOR_SEARCH_TOP_K + chunk_size: int = CHUNK_SIZE + chunk_conent: bool = True + score_threshold: int = VECTOR_SEARCH_SCORE_THRESHOLD + + def init_cfg(self, + top_k=VECTOR_SEARCH_TOP_K, + ): + + self.llm = None + self.top_k = top_k + + def init_knowledge_vector_store(self, + filepath, + vs_path: str or os.PathLike = None, + sentence_size=SENTENCE_SIZE, + text2vec=None): + loaded_files = [] + failed_files = [] + if isinstance(filepath, str): + if not os.path.exists(filepath): + print("路径不存在") + return None + elif os.path.isfile(filepath): + file = os.path.split(filepath)[-1] + try: + docs = load_file(filepath, SENTENCE_SIZE) + print(f"{file} 已成功加载") + loaded_files.append(filepath) + except Exception as e: + print(e) + print(f"{file} 未能成功加载") + return None + elif os.path.isdir(filepath): + docs = [] + for file in tqdm(os.listdir(filepath), desc="加载文件"): + fullfilepath = os.path.join(filepath, file) + try: + docs += load_file(fullfilepath, SENTENCE_SIZE) + loaded_files.append(fullfilepath) + except Exception as e: + print(e) + failed_files.append(file) + + if len(failed_files) > 0: + print("以下文件未能成功加载：") + for file in failed_files: + print(f"{file}\n") + + else: + docs = [] + for file in filepath: + docs += load_file(file, SENTENCE_SIZE) + print(f"{file} 已成功加载") + loaded_files.append(file) + + if len(docs) > 0: + print("文件加载完毕，正在生成向量库") + if vs_path and os.path.isdir(vs_path): + try: + self.vector_store = FAISS.load_local(vs_path, text2vec) + self.vector_store.add_documents(docs) + except: + self.vector_store = FAISS.from_documents(docs, text2vec) + else: + self.vector_store = FAISS.from_documents(docs, text2vec) # docs 为Document列表 + + self.vector_store.save_local(vs_path) + return vs_path, loaded_files + else: + raise RuntimeError("文件加载失败，请检查文件格式是否正确") + + def get_loaded_file(self, vs_path): + ds = self.vector_store.docstore + return set([ds._dict[k].metadata['source'].split(vs_path)[-1] for k in ds._dict]) + + + # query 查询内容 + # vs_path 知识库路径 + # chunk_conent 是否启用上下文关联 + # score_threshold 搜索匹配score阈值 + # vector_search_top_k 搜索知识库内容条数，默认搜索5条结果 + # chunk_sizes 匹配单段内容的连接上下文长度 + def get_knowledge_based_conent_test(self, query, vs_path, chunk_conent, + score_threshold=VECTOR_SEARCH_SCORE_THRESHOLD, + vector_search_top_k=VECTOR_SEARCH_TOP_K, chunk_size=CHUNK_SIZE, + text2vec=None): + self.vector_store = FAISS.load_local(vs_path, text2vec) + self.vector_store.chunk_conent = chunk_conent + self.vector_store.score_threshold = score_threshold + self.vector_store.chunk_size = chunk_size + + embedding = self.vector_store.embedding_function.embed_query(query) + related_docs_with_score = similarity_search_with_score_by_vector(self.vector_store, embedding, k=vector_search_top_k) + + if not related_docs_with_score: + response = {"query": query, + "source_documents": []} + return response, "" + # prompt = f"{query}. You should answer this question using information from following documents: \n\n" + prompt = f"{query}. 你必须利用以下文档中包含的信息回答这个问题: \n\n---\n\n" + prompt += "\n\n".join([f"({k}): " + doc.page_content for k, doc in enumerate(related_docs_with_score)]) + prompt += "\n\n---\n\n" + prompt = prompt.encode('utf-8', 'ignore').decode() # avoid reading non-utf8 chars + # print(prompt) + response = {"query": query, "source_documents": related_docs_with_score} + return response, prompt + + + + +def construct_vector_store(vs_id, vs_path, files, sentence_size, history, one_conent, one_content_segmentation, text2vec): + for file in files: + assert os.path.exists(file), "输入文件不存在：" + file + import nltk + if NLTK_DATA_PATH not in nltk.data.path: nltk.data.path = [NLTK_DATA_PATH] + nltk.data.path + local_doc_qa = LocalDocQA() + local_doc_qa.init_cfg() + filelist = [] + if not os.path.exists(os.path.join(vs_path, vs_id)): + os.makedirs(os.path.join(vs_path, vs_id)) + for file in files: + file_name = file.name if not isinstance(file, str) else file + filename = os.path.split(file_name)[-1] + shutil.copyfile(file_name, os.path.join(vs_path, vs_id, filename)) + filelist.append(os.path.join(vs_path, vs_id, filename)) + vs_path, loaded_files = local_doc_qa.init_knowledge_vector_store(filelist, os.path.join(vs_path, vs_id), sentence_size, text2vec) + + if len(loaded_files): + file_status = f"已添加 {'、'.join([os.path.split(i)[-1] for i in loaded_files if i])} 内容至知识库，并已加载知识库，请开始提问" + else: + pass + # file_status = "文件未成功加载，请重新上传文件" + # print(file_status) + return local_doc_qa, vs_path + +@Singleton +class knowledge_archive_interface(): + def __init__(self) -> None: + self.threadLock = threading.Lock() + self.current_id = "" + self.kai_path = None + self.qa_handle = None + self.text2vec_large_chinese = None + + def get_chinese_text2vec(self): + if self.text2vec_large_chinese is None: + # < -------------------预热文本向量化模组--------------- > + from toolbox import ProxyNetworkActivate + print('Checking Text2vec ...') + from langchain.embeddings.huggingface import HuggingFaceEmbeddings + with ProxyNetworkActivate('Download_LLM'): # 临时地激活代理网络 + self.text2vec_large_chinese = HuggingFaceEmbeddings(model_name="GanymedeNil/text2vec-large-chinese") + + return self.text2vec_large_chinese + + + def feed_archive(self, file_manifest, vs_path, id="default"): + self.threadLock.acquire() + # import uuid + self.current_id = id + self.qa_handle, self.kai_path = construct_vector_store( + vs_id=self.current_id, + vs_path=vs_path, + files=file_manifest, + sentence_size=100, + history=[], + one_conent="", + one_content_segmentation="", + text2vec = self.get_chinese_text2vec(), + ) + self.threadLock.release() + + def get_current_archive_id(self): + return self.current_id + + def get_loaded_file(self, vs_path): + return self.qa_handle.get_loaded_file(vs_path) + + def answer_with_archive_by_id(self, txt, id, vs_path): + self.threadLock.acquire() + if not self.current_id == id: + self.current_id = id + self.qa_handle, self.kai_path = construct_vector_store( + vs_id=self.current_id, + vs_path=vs_path, + files=[], + sentence_size=100, + history=[], + one_conent="", + one_content_segmentation="", + text2vec = self.get_chinese_text2vec(), + ) + VECTOR_SEARCH_SCORE_THRESHOLD = 0 + VECTOR_SEARCH_TOP_K = 4 + CHUNK_SIZE = 512 + resp, prompt = self.qa_handle.get_knowledge_based_conent_test( + query = txt, + vs_path = self.kai_path, + score_threshold=VECTOR_SEARCH_SCORE_THRESHOLD, + vector_search_top_k=VECTOR_SEARCH_TOP_K, + chunk_conent=True, + chunk_size=CHUNK_SIZE, + text2vec = self.get_chinese_text2vec(), + ) + self.threadLock.release() + return resp, prompt \ No newline at end of file diff --git a/crazy_functions/互动小游戏.py b/crazy_functions/互动小游戏.py index e00ef32..a6871b3 100644 --- a/crazy_functions/互动小游戏.py +++ b/crazy_functions/互动小游戏.py @@ -1,159 +1,59 @@ -from toolbox import CatchException, update_ui, get_conf, select_api_key, get_log_folder -from crazy_functions.multi_stage.multi_stage_utils import GptAcademicState +from toolbox import CatchException, update_ui, update_ui_lastest_msg +from crazy_functions.multi_stage.multi_stage_utils import GptAcademicGameBaseState from crazy_functions.crazy_utils import request_gpt_model_in_new_thread_with_ui_alive +from request_llms.bridge_all import predict_no_ui_long_connection +from crazy_functions.game_fns.game_utils import get_code_block, is_same_thing import random -class 小游戏(GptAcademicState): - def __init__(self): - self.need_game_reset = True - self.llm_kwargs = None - super().__init__() + +class MiniGame_ASCII_Art(GptAcademicGameBaseState): - def lock_plugin(self, chatbot): - chatbot._cookies['lock_plugin'] = 'crazy_functions.互动小游戏->谁是卧底' - self.dump_state(chatbot) - - def unlock_plugin(self, chatbot): - self.reset() - chatbot._cookies['lock_plugin'] = None - self.dump_state(chatbot) - - def set_state(self, chatbot, key, value): - return super().set_state(chatbot, key, value) - - def init_game(self, chatbot): - chatbot.get_cookies()['lock_plugin'] = '' - - def clean_up_game(self, chatbot): - chatbot.get_cookies()['lock_plugin'] = None - - def init_player(self): - pass - - def step(self, prompt, chatbot): - pass - - def continue_game(self, prompt, chatbot): - if self.need_game_reset: - self.need_game_reset = False - yield from self.init_game(chatbot) - yield from self.step(prompt, chatbot) - self.dump_state(chatbot) - yield update_ui(chatbot=chatbot, history=[]) - -class 小游戏_谁是卧底_玩家(): - def __init__(self, game_handle, card, llm_model, name) -> None: - self.game_handle = game_handle - self.card = card - self.name = name - self.is_out = False - self.llm_model = llm_model - self.is_human = llm_model == 'human' - self.what_player_has_spoken = [] - - def speek(self, content=None): - if content is None: - assert not self.is_human - speak_what = yield from + def step(self, prompt, chatbot, history): + if self.step_cnt == 0: + chatbot.append(["我画你猜（动物）", "请稍等..."]) else: - self.what_player_has_spoken.append(content) + if prompt.strip() == 'exit': + self.delete_game = True + yield from update_ui_lastest_msg(lastmsg=f"谜底是{self.obj}，游戏结束。", chatbot=chatbot, history=history, delay=0.) + return + chatbot.append([prompt, ""]) + yield from update_ui(chatbot=chatbot, history=history) - def agi_speek(self): - inputs = f'please say something about {self.card}' - res = yield from request_gpt_model_in_new_thread_with_ui_alive( - inputs = inputs, - inputs_show_user=inputs, - llm_kwargs=self.game_handle.llm_kwargs, - chatbot=chatbot, - history=history, - sys_prompt=sys_prompt - ) - pass + if self.step_cnt == 0: + self.lock_plugin(chatbot) + self.cur_task = 'draw' - def vote(self, content=None): - if content is None: - assert not self.is_human - self.vote_who = yield from - else: - try: - self.vote_who = int(content) - except: - self.vote_who = None + if self.cur_task == 'draw': + avail_obj = ["狗","猫","鸟","鱼","老鼠","蛇"] + self.obj = random.choice(avail_obj) + inputs = "I want to play a game called Guess the ASCII art. You can draw the ASCII art and I will try to guess it. " + f"This time you draw a {self.obj}. Note that you must not indicate what you have draw in the text, and you should only produce the ASCII art wrapped by ```. " + raw_res = predict_no_ui_long_connection(inputs=inputs, llm_kwargs=self.llm_kwargs, history=[], sys_prompt="") + self.cur_task = 'identify user guess' + res = get_code_block(raw_res) + history += ['', f'the answer is {self.obj}', inputs, res] + yield from update_ui_lastest_msg(lastmsg=res, chatbot=chatbot, history=history, delay=0.) - def agi_vote(self): - pass - -class 小游戏_谁是卧底(小游戏): - def __init__(self): - self.game_phase = '发言' # 投票 - super().__init__() - - def init_game(self, chatbot): - self.n_players = 3 - self.n_ai_players = self.n_players - 1 - card = "橙子" - undercover_card = "橘子" - llm_model = self.llm_kwargs['llm_model'] - self.players = [ - 小游戏_谁是卧底(self, card, llm_model, str(i)) for i in range(self.n_players) - ] - - undercover = random.randint(0, self.n_players-1) - human = 0 - - self.players[undercover].card = undercover_card - self.players[human].llm_model = 'human' - super().init_game(chatbot) - - def who_is_out(self): - votes = {} - for player in self.players: - if player.is_out: continue - if player.vote is None: continue - if player.vote not in votes: votes[player.vote] = 0 - votes[player.vote] += 1 - max_votes = max(votes.values()) - players_with_max_votes = [player for player, vote_count in votes.items() if vote_count == max_votes] - for player in players_with_max_votes: - print('淘汰了', player.name) - player.is_out = True - return players_with_max_votes - - def step(self, prompt, chatbot): - - if self.game_phase == '发言': - for player in self.players: - if player.is_out: continue - if player.is_human: - player.speek(prompt) - else: - player.speek() - self.game_phase = '投票' - - elif self.game_phase == '投票': - for player in self.players: - if player.is_out: continue - if player.is_human: - player.vote(prompt) - else: - player.vote() - self.who_is_out() - if len([player for player in self.players if not player.is_out]) <= 2: - if sum([player for player in self.players if player.is_undercover]) == 1: - print('卧底获胜') - else: - print('平民获胜') - self.need_game_reset = True - self.game_phase = '发言' - - else: - raise RuntimeError - + elif self.cur_task == 'identify user guess': + if is_same_thing(self.obj, prompt, self.llm_kwargs): + self.delete_game = True + yield from update_ui_lastest_msg(lastmsg="你猜对了！", chatbot=chatbot, history=history, delay=0.) + else: + self.cur_task = 'identify user guess' + yield from update_ui_lastest_msg(lastmsg="猜错了，再试试，输入“exit”获取答案。", chatbot=chatbot, history=history, delay=0.) + @CatchException -def 谁是卧底(prompt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port): - # 尚未完成 - history = [] # 清空历史 - state = 小游戏_谁是卧底.get_state(chatbot, 小游戏_谁是卧底) - state.llm_kwargs = llm_kwargs - yield from state.continue_game(prompt, chatbot) +def 随机小游戏(prompt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port): + # 清空历史 + history = [] + # 选择游戏 + cls = MiniGame_ASCII_Art + # 如果之前已经初始化了游戏实例，则继续该实例；否则重新初始化 + state = cls.sync_state(chatbot, + llm_kwargs, + cls, + plugin_name='MiniGame_ASCII_Art', + callback_fn='crazy_functions.互动小游戏->随机小游戏', + lock_plugin=True + ) + yield from state.continue_game(prompt, chatbot, history) diff --git a/crazy_functions/Langchain知识库.py b/crazy_functions/知识库问答.py similarity index 69% rename from crazy_functions/Langchain知识库.py rename to crazy_functions/知识库问答.py index 8433895..e1cd00c 100644 --- a/crazy_functions/Langchain知识库.py +++ b/crazy_functions/知识库问答.py @@ -1,10 +1,19 @@ -from toolbox import CatchException, update_ui, ProxyNetworkActivate, update_ui_lastest_msg +from toolbox import CatchException, update_ui, ProxyNetworkActivate, update_ui_lastest_msg, get_log_folder, get_user from .crazy_utils import request_gpt_model_in_new_thread_with_ui_alive, get_files_from_everything +install_msg =""" +1. python -m pip install torch --index-url https://download.pytorch.org/whl/cpu + +2. python -m pip install transformers protobuf langchain sentence-transformers faiss-cpu nltk beautifulsoup4 bitsandbytes tabulate icetk --upgrade + +3. python -m pip install unstructured[all-docs] --upgrade + +4. python -c 'import nltk; nltk.download("punkt")' +""" @CatchException -def 知识库问答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port): +def 知识库文件注入(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port): """ txt 输入栏用户输入的文本，例如需要翻译的一段话，再例如一个包含了待处理文件的路径 llm_kwargs gpt模型参数, 如温度和top_p等, 一般原样传递下去就行 @@ -25,15 +34,15 @@ def 知识库问答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_pro # resolve deps try: - from zh_langchain import construct_vector_store - from langchain.embeddings.huggingface import HuggingFaceEmbeddings - from .crazy_utils import knowledge_archive_interface + # from zh_langchain import construct_vector_store + # from langchain.embeddings.huggingface import HuggingFaceEmbeddings + from crazy_functions.vector_fns.vector_database import knowledge_archive_interface except Exception as e: - chatbot.append(["依赖不足", "导入依赖失败。正在尝试自动安装，请查看终端的输出或耐心等待..."]) + chatbot.append(["依赖不足", f"{str(e)}\n\n导入依赖失败。请用以下命令安装" + install_msg]) yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 - from .crazy_utils import try_install_deps - try_install_deps(['zh_langchain==0.2.1', 'pypinyin'], reload_m=['pypinyin', 'zh_langchain']) - yield from update_ui_lastest_msg("安装完成，您可以再次重试。", chatbot, history) + # from .crazy_utils import try_install_deps + # try_install_deps(['zh_langchain==0.2.1', 'pypinyin'], reload_m=['pypinyin', 'zh_langchain']) + # yield from update_ui_lastest_msg("安装完成，您可以再次重试。", chatbot, history) return # < --------------------读取文件--------------- > @@ -42,7 +51,7 @@ def 知识库问答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_pro for sp in spl: _, file_manifest_tmp, _ = get_files_from_everything(txt, type=f'.{sp}') file_manifest += file_manifest_tmp - + if len(file_manifest) == 0: chatbot.append(["没有找到任何可读取文件", "当前支持的格式包括: txt, md, docx, pptx, pdf, json等"]) yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 @@ -62,13 +71,14 @@ def 知识库问答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_pro print('Establishing knowledge archive ...') with ProxyNetworkActivate('Download_LLM'): # 临时地激活代理网络 kai = knowledge_archive_interface() - kai.feed_archive(file_manifest=file_manifest, id=kai_id) - kai_files = kai.get_loaded_file() + vs_path = get_log_folder(user=get_user(chatbot), plugin_name='vec_store') + kai.feed_archive(file_manifest=file_manifest, vs_path=vs_path, id=kai_id) + kai_files = kai.get_loaded_file(vs_path=vs_path) kai_files = '
'.join(kai_files) # chatbot.append(['知识库构建成功', "正在将知识库存储至cookie中"]) # yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 # chatbot._cookies['langchain_plugin_embedding'] = kai.get_current_archive_id() - # chatbot._cookies['lock_plugin'] = 'crazy_functions.Langchain知识库->读取知识库作答' + # chatbot._cookies['lock_plugin'] = 'crazy_functions.知识库文件注入->读取知识库作答' # chatbot.append(['完成', "“根据知识库作答”函数插件已经接管问答系统, 提问吧! 但注意, 您接下来不能再使用其他插件了，刷新页面即可以退出知识库问答模式。"]) chatbot.append(['构建完成', f"当前知识库内的有效文件：\n\n---\n\n{kai_files}\n\n---\n\n请切换至“知识库问答”插件进行知识库访问, 或者使用此插件继续上传更多文件。"]) yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 # 由于请求gpt需要一段时间，我们先及时地做一次界面更新 @@ -77,15 +87,15 @@ def 知识库问答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_pro def 读取知识库作答(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, web_port=-1): # resolve deps try: - from zh_langchain import construct_vector_store - from langchain.embeddings.huggingface import HuggingFaceEmbeddings - from .crazy_utils import knowledge_archive_interface + # from zh_langchain import construct_vector_store + # from langchain.embeddings.huggingface import HuggingFaceEmbeddings + from crazy_functions.vector_fns.vector_database import knowledge_archive_interface except Exception as e: - chatbot.append(["依赖不足", "导入依赖失败。正在尝试自动安装，请查看终端的输出或耐心等待..."]) + chatbot.append(["依赖不足", f"{str(e)}\n\n导入依赖失败。请用以下命令安装" + install_msg]) yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 - from .crazy_utils import try_install_deps - try_install_deps(['zh_langchain==0.2.1', 'pypinyin'], reload_m=['pypinyin', 'zh_langchain']) - yield from update_ui_lastest_msg("安装完成，您可以再次重试。", chatbot, history) + # from .crazy_utils import try_install_deps + # try_install_deps(['zh_langchain==0.2.1', 'pypinyin'], reload_m=['pypinyin', 'zh_langchain']) + # yield from update_ui_lastest_msg("安装完成，您可以再次重试。", chatbot, history) return # < ------------------- --------------- > @@ -93,7 +103,8 @@ def 读取知识库作答(txt, llm_kwargs, plugin_kwargs, chatbot, history, syst if ("advanced_arg" in plugin_kwargs) and (plugin_kwargs["advanced_arg"] == ""): plugin_kwargs.pop("advanced_arg") kai_id = plugin_kwargs.get("advanced_arg", 'default') - resp, prompt = kai.answer_with_archive_by_id(txt, kai_id) + vs_path = get_log_folder(user=get_user(chatbot), plugin_name='vec_store') + resp, prompt = kai.answer_with_archive_by_id(txt, kai_id, vs_path) chatbot.append((txt, f'[知识库 {kai_id}] ' + prompt)) yield from update_ui(chatbot=chatbot, history=history) # 刷新界面 # 由于请求gpt需要一段时间，我们先及时地做一次界面更新 diff --git a/docs/GithubAction+NoLocal+Vectordb b/docs/GithubAction+NoLocal+Vectordb new file mode 100644 index 0000000..98595e3 --- /dev/null +++ b/docs/GithubAction+NoLocal+Vectordb @@ -0,0 +1,26 @@ +# 此Dockerfile适用于“无本地模型”的环境构建，如果需要使用chatglm等本地模型，请参考 docs/Dockerfile+ChatGLM +# 如何构建: 先修改 `config.py`， 然后 docker build -t gpt-academic-nolocal-vs -f docs/GithubAction+NoLocal+Vectordb . +# 如何运行: docker run --rm -it --net=host gpt-academic-nolocal-vs +FROM python:3.11 + +# 指定路径 +WORKDIR /gpt + +# 装载项目文件 +COPY . . + +# 安装依赖 +RUN pip3 install -r requirements.txt + +# 安装知识库插件的额外依赖 +RUN apt-get update && apt-get install libgl1 -y +RUN pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cpu +RUN pip3 install transformers protobuf langchain sentence-transformers faiss-cpu nltk beautifulsoup4 bitsandbytes tabulate icetk --upgrade +RUN pip3 install unstructured[all-docs] --upgrade + +# 可选步骤，用于预热模块 +RUN python3 -c 'from check_proxy import warm_up_modules; warm_up_modules()' +RUN python3 -c 'from check_proxy import warm_up_vectordb; warm_up_vectordb()' + +# 启动 +CMD ["python3", "-u", "main.py"] diff --git a/docs/translate_english.json b/docs/translate_english.json index cbfdafd..02d3b64 100644 --- a/docs/translate_english.json +++ b/docs/translate_english.json @@ -2902,5 +2902,107 @@ "高优先级": "High priority", "请配置ZHIPUAI_API_KEY": "Please configure ZHIPUAI_API_KEY", "单个azure模型": "Single Azure model", - "预留参数 context 未实现": "Reserved parameter 'context' not implemented" + "预留参数 context 未实现": "Reserved parameter 'context' not implemented", + "在输入区输入临时API_KEY后提交": "Submit after entering temporary API_KEY in the input area", + "鸟": "Bird", + "图片中需要修改的位置用橡皮擦擦除为纯白色": "Erase the areas in the image that need to be modified with an eraser to pure white", + "└── PDF文档精准解析": "└── Accurate parsing of PDF documents", + "└── ALLOW_RESET_CONFIG 是否允许通过自然语言描述修改本页的配置": "└── ALLOW_RESET_CONFIG Whether to allow modifying the configuration of this page through natural language description", + "等待指令": "Waiting for instructions", + "不存在": "Does not exist", + "选择游戏": "Select game", + "本地大模型示意图": "Local large model diagram", + "无视此消息即可": "You can ignore this message", + "即RGB=255": "That is, RGB=255", + "如需追问": "If you have further questions", + "也可以是具体的模型路径": "It can also be a specific model path", + "才会起作用": "Will take effect", + "下载失败": "Download failed", + "网页刷新后失效": "Invalid after webpage refresh", + "crazy_functions.互动小游戏-": "crazy_functions.Interactive mini game-", + "右对齐": "Right alignment", + "您可以调用下拉菜单中的“LoadConversationHistoryArchive”还原当下的对话": "You can use the 'LoadConversationHistoryArchive' in the drop-down menu to restore the current conversation", + "左对齐": "Left alignment", + "使用默认的 FP16": "Use default FP16", + "一小时": "One hour", + "从而方便内存的释放": "Thus facilitating memory release", + "如何临时更换API_KEY": "How to temporarily change API_KEY", + "请输入 1024x1024-HD": "Please enter 1024x1024-HD", + "使用 INT8 量化": "Use INT8 quantization", + "3. 输入修改需求": "3. Enter modification requirements", + "刷新界面 由于请求gpt需要一段时间": "Refreshing the interface takes some time due to the request for gpt", + "随机小游戏": "Random mini game", + "那么请在下面的QWEN_MODEL_SELECTION中指定具体的模型": "So please specify the specific model in QWEN_MODEL_SELECTION below", + "表值": "Table value", + "我画你猜": "I draw, you guess", + "狗": "Dog", + "2. 输入分辨率": "2. Enter resolution", + "鱼": "Fish", + "尚未完成": "Not yet completed", + "表头": "Table header", + "填localhost或者127.0.0.1": "Fill in localhost or 127.0.0.1", + "请上传jpg格式的图片": "Please upload images in jpg format", + "API_URL_REDIRECT填写格式是错误的": "The format of API_URL_REDIRECT is incorrect", + "├── RWKV的支持见Wiki": "Support for RWKV is available in the Wiki", + "如果中文Prompt效果不理想": "If the Chinese prompt is not effective", + "/SEAFILE_LOCAL/50503047/我的资料库/学位/paperlatex/aaai/Fu_8368_with_appendix": "/SEAFILE_LOCAL/50503047/My Library/Degree/paperlatex/aaai/Fu_8368_with_appendix", + "只有当AVAIL_LLM_MODELS包含了对应本地模型时": "Only when AVAIL_LLM_MODELS contains the corresponding local model", + "选择本地模型变体": "Choose the local model variant", + "如果您确信自己没填错": "If you are sure you haven't made a mistake", + "PyPDF2这个库有严重的内存泄露问题": "PyPDF2 library has serious memory leak issues", + "整理文件集合 输出消息": "Organize file collection and output message", + "没有检测到任何近期上传的图像文件": "No recently uploaded image files detected", + "游戏结束": "Game over", + "调用结束": "Call ended", + "猫": "Cat", + "请及时切换模型": "Please switch models in time", + "次中": "In the meantime", + "如需生成高清图像": "If you need to generate high-definition images", + "CPU 模式": "CPU mode", + "项目目录": "Project directory", + "动物": "Animal", + "居中对齐": "Center alignment", + "请注意拓展名需要小写": "Please note that the extension name needs to be lowercase", + "重试第": "Retry", + "实验性功能": "Experimental feature", + "猜错了": "Wrong guess", + "打开你的代理软件查看代理协议": "Open your proxy software to view the proxy agreement", + "您不需要再重复强调该文件的路径了": "You don't need to emphasize the file path again", + "请阅读": "Please read", + "请直接输入您的问题": "Please enter your question directly", + "API_URL_REDIRECT填错了": "API_URL_REDIRECT is filled incorrectly", + "谜底是": "The answer is", + "第一个模型": "The first model", + "你猜对了！": "You guessed it right!", + "已经接收到您上传的文件": "The file you uploaded has been received", + "您正在调用“图像生成”插件": "You are calling the 'Image Generation' plugin", + "刷新界面 界面更新": "Refresh the interface, interface update", + "如果之前已经初始化了游戏实例": "If the game instance has been initialized before", + "文件": "File", + "老鼠": "Mouse", + "列2": "Column 2", + "等待图片": "Waiting for image", + "使用 INT4 量化": "Use INT4 quantization", + "from crazy_functions.互动小游戏 import 随机小游戏": "TranslatedText", + "游戏主体": "TranslatedText", + "该模型不具备上下文对话能力": "TranslatedText", + "列3": "TranslatedText", + "清理": "TranslatedText", + "检查量化配置": "TranslatedText", + "如果游戏结束": "TranslatedText", + "蛇": "TranslatedText", + "则继续该实例；否则重新初始化": "TranslatedText", + "e.g. cat and 猫 are the same thing": "TranslatedText", + "第三个模型": "TranslatedText", + "如果你选择Qwen系列的模型": "TranslatedText", + "列4": "TranslatedText", + "输入“exit”获取答案": "TranslatedText", + "把它放到子进程中运行": "TranslatedText", + "列1": "TranslatedText", + "使用该模型需要额外依赖": "TranslatedText", + "再试试": "TranslatedText", + "1. 上传图片": "TranslatedText", + "保存状态": "TranslatedText", + "GPT-Academic对话存档": "TranslatedText", + "Arxiv论文精细翻译": "TranslatedText" } diff --git a/multi_language.py b/multi_language.py index a20fb5a..a807dbd 100644 --- a/multi_language.py +++ b/multi_language.py @@ -182,12 +182,12 @@ cached_translation = read_map_from_json(language=LANG) def trans(word_to_translate, language, special=False): if len(word_to_translate) == 0: return {} from crazy_functions.crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency - from toolbox import get_conf, ChatBotWithCookies - proxies, WEB_PORT, LLM_MODEL, CONCURRENT_COUNT, AUTHENTICATION, CHATBOT_HEIGHT, LAYOUT, API_KEY = \ - get_conf('proxies', 'WEB_PORT', 'LLM_MODEL', 'CONCURRENT_COUNT', 'AUTHENTICATION', 'CHATBOT_HEIGHT', 'LAYOUT', 'API_KEY') + from toolbox import get_conf, ChatBotWithCookies, load_chat_cookies + + cookies = load_chat_cookies() llm_kwargs = { - 'api_key': API_KEY, - 'llm_model': LLM_MODEL, + 'api_key': cookies['api_key'], + 'llm_model': cookies['llm_model'], 'top_p':1.0, 'max_length': None, 'temperature':0.4, @@ -245,15 +245,15 @@ def trans(word_to_translate, language, special=False): def trans_json(word_to_translate, language, special=False): if len(word_to_translate) == 0: return {} from crazy_functions.crazy_utils import request_gpt_model_multi_threads_with_very_awesome_ui_and_high_efficiency - from toolbox import get_conf, ChatBotWithCookies - proxies, WEB_PORT, LLM_MODEL, CONCURRENT_COUNT, AUTHENTICATION, CHATBOT_HEIGHT, LAYOUT, API_KEY = \ - get_conf('proxies', 'WEB_PORT', 'LLM_MODEL', 'CONCURRENT_COUNT', 'AUTHENTICATION', 'CHATBOT_HEIGHT', 'LAYOUT', 'API_KEY') + from toolbox import get_conf, ChatBotWithCookies, load_chat_cookies + + cookies = load_chat_cookies() llm_kwargs = { - 'api_key': API_KEY, - 'llm_model': LLM_MODEL, + 'api_key': cookies['api_key'], + 'llm_model': cookies['llm_model'], 'top_p':1.0, 'max_length': None, - 'temperature':0.1, + 'temperature':0.4, } import random N_EACH_REQ = random.randint(16, 32) diff --git a/request_llms/bridge_all.py b/request_llms/bridge_all.py index 8dece54..dcfeba9 100644 --- a/request_llms/bridge_all.py +++ b/request_llms/bridge_all.py @@ -552,7 +552,7 @@ if "deepseekcoder" in AVAIL_LLM_MODELS: # deepseekcoder "fn_with_ui": deepseekcoder_ui, "fn_without_ui": deepseekcoder_noui, "endpoint": None, - "max_token": 4096, + "max_token": 2048, "tokenizer": tokenizer_gpt35, "token_cnt": get_token_num_gpt35, } diff --git a/request_llms/bridge_deepseekcoder.py b/request_llms/bridge_deepseekcoder.py index 2242eec..89964ab 100644 --- a/request_llms/bridge_deepseekcoder.py +++ b/request_llms/bridge_deepseekcoder.py @@ -6,6 +6,7 @@ from toolbox import ProxyNetworkActivate from toolbox import get_conf from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns from threading import Thread +import torch def download_huggingface_model(model_name, max_retry, local_dir): from huggingface_hub import snapshot_download @@ -36,9 +37,46 @@ class GetCoderLMHandle(LocalLLMHandle): # tokenizer = download_huggingface_model(model_name, max_retry=128, local_dir=local_dir) tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True) self._streamer = TextIteratorStreamer(tokenizer) - model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True) + device_map = { + "transformer.word_embeddings": 0, + "transformer.word_embeddings_layernorm": 0, + "lm_head": 0, + "transformer.h": 0, + "transformer.ln_f": 0, + "model.embed_tokens": 0, + "model.layers": 0, + "model.norm": 0, + } + + # 检查量化配置 + quantization_type = get_conf('LOCAL_MODEL_QUANT') + if get_conf('LOCAL_MODEL_DEVICE') != 'cpu': - model = model.cuda() + if quantization_type == "INT8": + from transformers import BitsAndBytesConfig + # 使用 INT8 量化 + model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, load_in_8bit=True, + device_map=device_map) + elif quantization_type == "INT4": + from transformers import BitsAndBytesConfig + # 使用 INT4 量化 + bnb_config = BitsAndBytesConfig( + load_in_4bit=True, + bnb_4bit_use_double_quant=True, + bnb_4bit_quant_type="nf4", + bnb_4bit_compute_dtype=torch.bfloat16 + ) + model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, + quantization_config=bnb_config, device_map=device_map) + else: + # 使用默认的 FP16 + model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, + torch_dtype=torch.bfloat16, device_map=device_map) + else: + # CPU 模式 + model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, + torch_dtype=torch.bfloat16) + return model, tokenizer def llm_stream_generator(self, **kwargs): @@ -54,7 +92,10 @@ class GetCoderLMHandle(LocalLLMHandle): query, max_length, top_p, temperature, history = adaptor(kwargs) history.append({ 'role': 'user', 'content': query}) messages = history - inputs = self._tokenizer.apply_chat_template(messages, return_tensors="pt").to(self._model.device) + inputs = self._tokenizer.apply_chat_template(messages, return_tensors="pt") + if inputs.shape[1] > max_length: + inputs = inputs[:, -max_length:] + inputs = inputs.to(self._model.device) generation_kwargs = dict( inputs=inputs, max_new_tokens=max_length, diff --git a/tests/test_plugins.py b/tests/test_plugins.py index 8470895..13ec259 100644 --- a/tests/test_plugins.py +++ b/tests/test_plugins.py @@ -48,11 +48,11 @@ if __name__ == "__main__": # for lang in ["English", "French", "Japanese", "Korean", "Russian", "Italian", "German", "Portuguese", "Arabic"]: # plugin_test(plugin='crazy_functions.批量Markdown翻译->Markdown翻译指定语言', main_input="README.md", advanced_arg={"advanced_arg": lang}) - # plugin_test(plugin='crazy_functions.Langchain知识库->知识库问答', main_input="./") + # plugin_test(plugin='crazy_functions.知识库文件注入->知识库文件注入', main_input="./") - # plugin_test(plugin='crazy_functions.Langchain知识库->读取知识库作答', main_input="What is the installation method？") + # plugin_test(plugin='crazy_functions.知识库文件注入->读取知识库作答', main_input="What is the installation method？") - # plugin_test(plugin='crazy_functions.Langchain知识库->读取知识库作答', main_input="远程云服务器部署？") + # plugin_test(plugin='crazy_functions.知识库文件注入->读取知识库作答', main_input="远程云服务器部署？") # plugin_test(plugin='crazy_functions.Latex输出PDF结果->Latex翻译中文并重新编译PDF', main_input="2210.03629") diff --git a/tests/test_utils.py b/tests/test_utils.py index 1fdca1e..c87908f 100644 --- a/tests/test_utils.py +++ b/tests/test_utils.py @@ -56,11 +56,11 @@ vt.get_plugin_handle = silence_stdout_fn(get_plugin_handle) vt.get_plugin_default_kwargs = silence_stdout_fn(get_plugin_default_kwargs) vt.get_chat_handle = silence_stdout_fn(get_chat_handle) vt.get_chat_default_kwargs = silence_stdout_fn(get_chat_default_kwargs) -vt.chat_to_markdown_str = chat_to_markdown_str +vt.chat_to_markdown_str = (chat_to_markdown_str) proxies, WEB_PORT, LLM_MODEL, CONCURRENT_COUNT, AUTHENTICATION, CHATBOT_HEIGHT, LAYOUT, API_KEY = \ vt.get_conf('proxies', 'WEB_PORT', 'LLM_MODEL', 'CONCURRENT_COUNT', 'AUTHENTICATION', 'CHATBOT_HEIGHT', 'LAYOUT', 'API_KEY') -def plugin_test(main_input, plugin, advanced_arg=None): +def plugin_test(main_input, plugin, advanced_arg=None, debug=True): from rich.live import Live from rich.markdown import Markdown @@ -72,7 +72,10 @@ def plugin_test(main_input, plugin, advanced_arg=None): plugin_kwargs['main_input'] = main_input if advanced_arg is not None: plugin_kwargs['plugin_kwargs'] = advanced_arg - my_working_plugin = silence_stdout(plugin)(**plugin_kwargs) + if debug: + my_working_plugin = (plugin)(**plugin_kwargs) + else: + my_working_plugin = silence_stdout(plugin)(**plugin_kwargs) with Live(Markdown(""), auto_refresh=False, vertical_overflow="visible") as live: for cookies, chat, hist, msg in my_working_plugin: diff --git a/tests/test_vector_plugins.py b/tests/test_vector_plugins.py new file mode 100644 index 0000000..9b75463 --- /dev/null +++ b/tests/test_vector_plugins.py @@ -0,0 +1,17 @@ +""" +对项目中的各个插件进行测试。运行方法：直接运行 python tests/test_plugins.py +""" + + +import os, sys +def validate_path(): dir_name = os.path.dirname(__file__); root_dir_assume = os.path.abspath(dir_name + '/..'); os.chdir(root_dir_assume); sys.path.append(root_dir_assume) +validate_path() # 返回项目根路径 + +if __name__ == "__main__": + from tests.test_utils import plugin_test + + plugin_test(plugin='crazy_functions.知识库问答->知识库文件注入', main_input="./README.md") + + plugin_test(plugin='crazy_functions.知识库问答->读取知识库作答', main_input="What is the installation method？") + + plugin_test(plugin='crazy_functions.知识库问答->读取知识库作答', main_input="远程云服务器部署？") \ No newline at end of file diff --git a/toolbox.py b/toolbox.py index f830a3d..8d91035 100644 --- a/toolbox.py +++ b/toolbox.py @@ -180,12 +180,15 @@ def HotReload(f): 最后，使用yield from语句返回重新加载过的函数，并在被装饰的函数上执行。 最终，装饰器函数返回内部函数。这个内部函数可以将函数的原始定义更新为最新版本，并执行函数的新版本。 """ - @wraps(f) - def decorated(*args, **kwargs): - fn_name = f.__name__ - f_hot_reload = getattr(importlib.reload(inspect.getmodule(f)), fn_name) - yield from f_hot_reload(*args, **kwargs) - return decorated + if get_conf('PLUGIN_HOT_RELOAD'): + @wraps(f) + def decorated(*args, **kwargs): + fn_name = f.__name__ + f_hot_reload = getattr(importlib.reload(inspect.getmodule(f)), fn_name) + yield from f_hot_reload(*args, **kwargs) + return decorated + else: + return f """ @@ -916,7 +919,14 @@ def read_single_conf_with_lru_cache(arg): @lru_cache(maxsize=128) def get_conf(*args): - # 建议您复制一个config_private.py放自己的秘密, 如API和代理网址, 避免不小心传github被别人看到 + """ + 本项目的所有配置都集中在config.py中。 修改配置有三种方法，您只需要选择其中一种即可： + - 直接修改config.py + - 创建并修改config_private.py + - 修改环境变量（修改docker-compose.yml等价于修改容器内部的环境变量） + + 注意：如果您使用docker-compose部署，请修改docker-compose（等价于修改容器内部的环境变量） + """ res = [] for arg in args: r = read_single_conf_with_lru_cache(arg)