# coding=utf-8 # Copyright 2023 The AIWaves Inc. team. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """helper functions for an LLM autonoumous agent""" import csv import random import json import pandas import numpy as np import requests import torch from tqdm import tqdm import re import datetime import string import random import os import openai from text2vec import semantic_search import re import datetime from langchain.document_loaders import UnstructuredFileLoader from langchain.text_splitter import CharacterTextSplitter from sentence_transformers import SentenceTransformer embed_model_name = os.environ["Embed_Model"] if "Embed_Model" in os.environ else "text-embedding-ada-002" if embed_model_name in ["text-embedding-ada-002"]: pass else: embedding_model = SentenceTransformer( embed_model_name, device=torch.device("cpu") ) def get_embedding(sentence): if embed_model_name in ["text-embedding-ada-002"]: openai.api_key = os.environ["API_KEY"] if "PROXY" in os.environ: assert "http:" in os.environ["PROXY"] or "socks" in os.environ["PROXY"],"PROXY error,PROXY must be http or socks" openai.proxy = os.environ["PROXY"] if "API_BASE" in os.environ: openai.api_base = os.environ["API_BASE"] embedding_model = openai.Embedding embed = embedding_model.create( model=embed_model_name, input=sentence ) embed = embed["data"][0]["embedding"] embed = torch.tensor(embed,dtype=torch.float32) else: embed = embedding_model.encode(sentence,convert_to_tensor=True) if len(embed.shape)==1: embed = embed.unsqueeze(0) return embed def get_code(): return "".join(random.sample(string.ascii_letters + string.digits, 8)) def get_content_between_a_b(start_tag, end_tag, text): """ Args: start_tag (str): start_tag end_tag (str): end_tag text (str): complete sentence Returns: str: the content between start_tag and end_tag """ extracted_text = "" start_index = text.find(start_tag) while start_index != -1: end_index = text.find(end_tag, start_index + len(start_tag)) if end_index != -1: extracted_text += text[start_index + len(start_tag):end_index] + " " start_index = text.find(start_tag, end_index + len(end_tag)) else: break return extracted_text.strip() def extract(text, type): """extract the content between Args: text (str): complete sentence type (str): tag Returns: str: content between """ target_str = get_content_between_a_b(f"<{type}>", f"", text) return target_str def count_files_in_directory(directory): # 获取指定目录下的文件数目 file_count = len([f for f in os.listdir(directory) if os.path.isfile(os.path.join(directory, f))]) return file_count def delete_oldest_files(directory, num_to_keep): # 获取目录下文件列表,并按修改时间排序 files = [(f, os.path.getmtime(os.path.join(directory, f))) for f in os.listdir(directory) if os.path.isfile(os.path.join(directory, f))] # 删除最开始的 num_to_keep 个文件 for i in range(min(num_to_keep, len(files))): file_to_delete = os.path.join(directory, files[i][0]) os.remove(file_to_delete) def delete_files_if_exceed_threshold(directory, threshold, num_to_keep): # 获取文件数目并进行处理 file_count = count_files_in_directory(directory) if file_count > threshold: delete_count = file_count - num_to_keep delete_oldest_files(directory, delete_count) def save_logs(log_path, messages, response): if not os.path.exists(log_path): os.mkdir(log_path) delete_files_if_exceed_threshold(log_path, 20, 10) log_path = log_path if log_path else "logs" log = {} log["input"] = messages log["output"] = response os.makedirs(log_path, exist_ok=True) log_file = os.path.join( log_path, datetime.datetime.now().strftime("%Y-%m-%d-%H-%M-%S") + ".json") with open(log_file, "w", encoding="utf-8") as f: json.dump(log, f, ensure_ascii=False, indent=2) def semantic_search_word2vec(query_embedding, kb_embeddings, top_k): return semantic_search(query_embedding, kb_embeddings, top_k=top_k) def cut_sent(para): para = re.sub("([。!?\?])([^”’])", r"\1\n\2", para) para = re.sub("(\.{6})([^”’])", r"\1\n\2", para) para = re.sub("(\…{2})([^”’])", r"\1\n\2", para) para = re.sub("([。!?\?][”’])([^,。!?\?])", r"\1\n\2", para) para = para.rstrip() pieces = [i for i in para.split("\n") if i] batch_size = 3 chucks = [ " ".join(pieces[i:i + batch_size]) for i in range(0, len(pieces), batch_size) ] return chucks def process_document(file_path): """ Save QA_csv to json. Args: model: LLM to generate embeddings qa_dict: A dict contains Q&A save_path: where to save the json file. Json format: Dict[num,Dict[q:str,a:str,chunk:str,emb:List[float]] """ final_dict = {} count = 0 if file_path.endswith(".csv"): dataset = pandas.read_csv(file_path) questions = dataset["question"] answers = dataset["answer"] # embedding q+chunk for q, a in zip(questions, answers): for text in cut_sent(a): temp_dict = {} temp_dict["q"] = q temp_dict["a"] = a temp_dict["chunk"] = text temp_dict["emb"] = get_embedding(q + text).tolist() final_dict[count] = temp_dict count += 1 # embedding chunk for q, a in zip(questions, answers): for text in cut_sent(a): temp_dict = {} temp_dict["q"] = q temp_dict["a"] = a temp_dict["chunk"] = text temp_dict["emb"] = get_embedding(text).tolist() final_dict[count] = temp_dict count += 1 # embedding q for q, a in zip(questions, answers): temp_dict = {} temp_dict["q"] = q temp_dict["a"] = a temp_dict["chunk"] = a temp_dict["emb"] = get_embedding(q).tolist() final_dict[count] = temp_dict count += 1 # embedding q+a for q, a in zip(questions, answers): temp_dict = {} temp_dict["q"] = q temp_dict["a"] = a temp_dict["chunk"] = a temp_dict["emb"] = get_embedding(q + a).tolist() final_dict[count] = temp_dict count += 1 # embedding a for q, a in zip(questions, answers): temp_dict = {} temp_dict["q"] = q temp_dict["a"] = a temp_dict["chunk"] = a temp_dict["emb"] = get_embedding(a).tolist() final_dict[count] = temp_dict count += 1 print(f"finish updating {len(final_dict)} data!") os.makedirs("temp_database", exist_ok=True) save_path = os.path.join( "temp_database/", file_path.split("/")[-1].replace("." + file_path.split(".")[1], ".json"), ) print(save_path) with open(save_path, "w") as f: json.dump(final_dict, f, ensure_ascii=False, indent=2) return {"knowledge_base": save_path, "type": "QA"} else: loader = UnstructuredFileLoader(file_path) docs = loader.load() text_spiltter = CharacterTextSplitter(chunk_size=200, chunk_overlap=100) docs = text_spiltter.split_text(docs[0].page_content) os.makedirs("temp_database", exist_ok=True) save_path = os.path.join( "temp_database/", file_path.replace("." + file_path.split(".")[1], ".json")) final_dict = {} count = 0 for c in tqdm(docs): temp_dict = {} temp_dict["chunk"] = c temp_dict["emb"] = get_embedding(c).tolist() final_dict[count] = temp_dict count += 1 print(f"finish updating {len(final_dict)} data!") with open(save_path, "w") as f: json.dump(final_dict, f, ensure_ascii=False, indent=2) return {"knowledge_base": save_path, "type": "UnstructuredFile"} def load_knowledge_base_qa(path): """ Load json format knowledge base. """ print("path", path) with open(path, "r") as f: data = json.load(f) embeddings = [] questions = [] answers = [] chunks = [] for idx in range(len(data.keys())): embeddings.append(data[str(idx)]["emb"]) questions.append(data[str(idx)]["q"]) answers.append(data[str(idx)]["a"]) chunks.append(data[str(idx)]["chunk"]) embeddings = np.array(embeddings, dtype=np.float32) embeddings = torch.from_numpy(embeddings).squeeze() return embeddings, questions, answers, chunks def load_knowledge_base_UnstructuredFile(path): """ Load json format knowledge base. """ with open(path, "r") as f: data = json.load(f) embeddings = [] chunks = [] for idx in range(len(data.keys())): embeddings.append(data[str(idx)]["emb"]) chunks.append(data[str(idx)]["chunk"]) embeddings = np.array(embeddings, dtype=np.float32) embeddings = torch.from_numpy(embeddings).squeeze() return embeddings, chunks def cos_sim(a: torch.Tensor, b: torch.Tensor): """ Computes the cosine similarity cos_sim(a[i], b[j]) for all i and j. :return: Matrix with res[i][j] = cos_sim(a[i], b[j]) """ if not isinstance(a, torch.Tensor): a = torch.tensor(a) if not isinstance(b, torch.Tensor): b = torch.tensor(b) if len(a.shape) == 1: a = a.unsqueeze(0) if len(b.shape) == 1: b = b.unsqueeze(0) a_norm = torch.nn.functional.normalize(a, p=2, dim=1) b_norm = torch.nn.functional.normalize(b, p=2, dim=1) return torch.mm(a_norm, b_norm.transpose(0, 1)) def matching_a_b(a, b, requirements=None): a_embedder = get_embedding(a) # 获取embedder b_embeder = get_embedding(b) sim_scores = cos_sim(a_embedder, b_embeder)[0] return sim_scores def matching_category(inputtext, forest_name, requirements=None, cat_embedder=None, top_k=3): """ Args: inputtext: the category name to be matched forest: search tree top_k: the default three highest scoring results Return: topk matching_result. List[List] [[top1_name,top2_name,top3_name],[top1_score,top2_score,top3_score]] """ sim_scores = torch.zeros([100]) if inputtext: input_embeder = get_embedding(inputtext) sim_scores = cos_sim(input_embeder, cat_embedder)[0] if requirements: requirements = requirements.split(" ") requirements_embedder = get_embedding(requirements) req_scores = cos_sim(requirements_embedder, cat_embedder) req_scores = torch.mean(req_scores, dim=0) total_scores = req_scores else: total_scores = sim_scores top_k_cat = torch.topk(total_scores, k=top_k) top_k_score, top_k_idx = top_k_cat[0], top_k_cat[1] top_k_name = [forest_name[top_k_idx[i]] for i in range(0, top_k)] return [top_k_name, top_k_score.tolist(), top_k_idx] def sample_with_order_preserved(lst, num): """Randomly sample from the list while maintaining the original order.""" indices = list(range(len(lst))) sampled_indices = random.sample(indices, num) sampled_indices.sort() # 保持原顺序 return [lst[i] for i in sampled_indices] def limit_values(data, max_values): """Reduce each key-value list in the dictionary to the specified size, keeping the order of the original list unchanged.""" for key, values in data.items(): if len(values) > max_values: data[key] = sample_with_order_preserved(values, max_values) return data def limit_keys(data, max_keys): """Reduce the dictionary to the specified number of keys.""" keys = list(data.keys()) if len(keys) > max_keys: keys = sample_with_order_preserved(keys, max_keys) data = {key: data[key] for key in keys} return data def flatten_dict(nested_dict): """ flatten the dictionary """ flattened_dict = {} for key, value in nested_dict.items(): if isinstance(value, dict): flattened_subdict = flatten_dict(value) flattened_dict.update(flattened_subdict) else: flattened_dict[key] = value return flattened_dict def merge_list(list1, list2): for l in list2: if l not in list1: list1.append(l) return list1 def Search_Engines(req): FETSIZE = eval(os.environ["FETSIZE"]) if "FETSIZE" in os.environ else 5 new_dict = {"keyword": req, "catLeafName": "", "fetchSize": FETSIZE} url = os.environ["SHOPPING_SEARCH"] res = requests.post( url= url, json=new_dict, ) user_dict = json.loads(res.text) if "data" in user_dict.keys(): request_items = user_dict["data"]["items"] # 查询到的商品信息JSON top_category = user_dict["data"]["topCategories"] return request_items, top_category else: return [] def search_with_api(requirements, categery): FETSIZE = eval(os.environ["FETSIZE"]) if "FETSIZE" in os.environ else 5 request_items = [] all_req_list = requirements.split(" ") count = 0 while len(request_items) < FETSIZE and len(all_req_list) > 0: if count: all_req_list.pop(0) all_req = (" ").join(all_req_list) if categery not in all_req_list: all_req = all_req + " " + categery now_request_items, top_category = Search_Engines(all_req) request_items = merge_list(request_items, now_request_items) count += 1 new_top = [] for category in top_category: if "其它" in category or "其它" in category: continue else: new_top.append(category) if len(request_items) > FETSIZE: request_items = request_items[:FETSIZE] return request_items, new_top def get_relevant_history(query,history,embeddings): """ Retrieve a list of key history entries based on a query using semantic search. Args: query (str): The input query for which key history is to be retrieved. history (list): A list of historical key entries. embeddings (numpy.ndarray): An array of embedding vectors for historical entries. Returns: list: A list of key history entries most similar to the query. """ TOP_K = eval(os.environ["TOP_K"]) if "TOP_K" in os.environ else 2 relevant_history = [] query_embedding = get_embedding(query) hits = semantic_search(query_embedding, embeddings, top_k=min(TOP_K,embeddings.shape[0])) hits = hits[0] for hit in hits: matching_idx = hit["corpus_id"] try: relevant_history.append(history[matching_idx]) except: return [] return relevant_history