# from dotenv import load_dotenv
# import os
# import time
#
# print("step 1: loading env")
# load_dotenv()
#
# hf_token = os.getenv("HF_TOKEN")
# print("step 2:", "HF_TOKEN loaded" if hf_token else "No token")
#
# print("step 3: importing sentence-transformers")
# from sentence_transformers import SentenceTransformer
#
# print("step 4: start loading model")
# t0 = time.time()
# model = SentenceTransformer(r"E:\Models\bge-m3", local_files_only=True, device="cuda")
# print(f"step 5: model loaded in {time.time() - t0:.2f}s")
#
# print("step 6: importing sklearn/numpy")
# from sklearn.metrics.pairwise import cosine_similarity
# import numpy as np
#
# titles = [
#     # A组：同品牌同产品，但含义不同
#     "苹果发布新款iPhone，影像系统再次升级",
#     "苹果推出全新iPhone，摄像头性能进一步增强",
#     "苹果回应新款iPhone发热问题：将通过系统更新修复",
#     "苹果下调部分旧款iPhone售价，新机型并未参与促销",
#
#     # B组：看起来都像“苹果新闻”，但主题已变
#     "苹果公司股价上涨，市值再创新高",
#     "苹果供应链承压，部分零部件厂商下调全年预期",
#     "苹果被曝缩减Vision产品产量，市场需求不及预期",
#     "苹果发布新款MacBook，并未更新iPhone产品线",
#
#     # C组：同样是“发布/推出”，但主体不同
#     "华为发布新款手机，影像能力进一步提升",
#     "小米推出全新手机，影像系统迎来升级",
#     "OPPO发布年度旗舰机型，主打夜景拍摄",
# ]
#
# print("step 7: start encoding")
# t1 = time.time()
# embeddings = model.encode(
#     titles,
#     normalize_embeddings=True,
#     show_progress_bar=True,
#     batch_size=16
# )
# print(f"step 8: encode done in {time.time() - t1:.2f}s")
#
# sim = cosine_similarity(embeddings)
# print(np.round(sim, 4))
#
import secrets

print(secrets.token_urlsafe(64))