11月LLM待学习

iterative dpo :

• https://github.com/RLHFlow/Online-RLHF
• https://github.com/YuxiXie/MCTS-DPO (蒙特卡洛树dpo)
  longWriter:https://github.com/THUDM/LongWriter/tree/main
  agent：了解跑通agent的四种功能
  rag：
  文本匹配：TinyBERT、LightMatcher
  '''

Sentence-BERT/BERT相关模型

Word2Vec/Doc2Vec计算语义相似度

TF-IDF + 余弦相似度

BM25算法

使用Sentence-BERT

from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np

加载模型

model = SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2')

工具名称及描述

tools = {
"PPT大纲": "用于生成PPT演示文稿的框架结构",
"申请书": "用于生成各类申请书、报告的模板",
"论文查重": "检测文章重复率"
}

def get_tool_similarity(query, tools):
# 编码query
query_embedding = model.encode(query)

similarities = {}
for tool_name, desc in tools.items():
    # 编码工具描述
    tool_embedding = model.encode(desc)
    # 计算余弦相似度
    similarity = cosine_similarity(
        query_embedding.reshape(1, -1), 
        tool_embedding.reshape(1, -1)
    )[0][0]
    similarities[tool_name] = similarity

return similarities

使用示例

query = "国家励志奖学金申请书"
similarities = get_tool_similarity(query, tools)
best_match = max(similarities.items(), key=lambda x: x[1])
print(f"最匹配的工具是: {best_match[0]}, 相似度: {best_match[1]:.4f}")
'''