大语言模型响应结果的可靠性分析实战

pip install --upgrade cleanlab-studio
1.

from cleanlab_studio import Studio
studio = Studio("<CLEANLAB_API_KEY>") # 从上面获取您的API密钥
tlm = studio.TLM(options={"log": ["explanation"], "model": "gpt-4o"}) # GPT, Claude, etc
#设置提示
out = tlm.prompt("How many vowels are there in the word 'Abracadabra'.?")
#TLM响应包含实际输出的“响应”、可信度评分和解释
print(f"Model's response = {out['response']}")
print(f"Trustworthiness score = {out['trustworthiness_score']}")
print(f"Explanation = {out['log']['explanation']}")
1.
2.
3.
4.
5.
6.
7.
8.
9.

Model's response = The word "Abracadabra" contains 6 vowels. The vowels are: A, a, a, a, a, and a.
Trustworthiness score = 0.6842228802750124
Explanation = This response is untrustworthy due to a lack of consistency in possible responses from the model. Here's one inconsistent alternate response that the model considered (which may not be accurate either):
5.
1.
2.
3.
4.

Model's response = Let me count the vowels in 'Abracadabra':
A-b-r-a-c-a-d-a-b-r-a

The vowels are: A, a, a, a, a

There are 5 vowels in the word 'Abracadabra'.
Trustworthiness score = 0.9378276048845285
Explanation = Did not find a reason to doubt trustworthiness.
1.
2.
3.
4.
5.
6.
7.
8.

python
from cleanlab_studio import Studio
import markdown
from IPython.core.display import display, Markdown

# 使用API密钥初始化Cleanlab Studio
studio = Studio("<CLEANLAB_API_KEY>") #替换为您的实际API密钥

# 要评估的模型列表
models = ["gpt-4o", "claude-3.5-sonnet-v2"]

# 定义提示
prompt_text = "Which one of 9.11 and 9.9 is bigger?"

# 遍历每个模型并进行评估
for model in models:
 tlm = studio.TLM(options={"log": ["explanation"], "model": model})
 out = tlm.prompt(prompt_text)

 md_content = f"""
## 模型: {model}

**响应**: {out['response']}

**可信度评分**: {out['trustworthiness_score']}

**解释**: {out['log']['explanation']}

---
"""
 display(Markdown(md_content))
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.

import streamlit as st
from langchain_groq.chat_models import ChatGroq
import os
os.environ["GROQ_API_KEY"]=st.secrets["GROQ_API_KEY"]
#初始化Groq Llama即时模型
groq_llm = ChatGroq(model="deepseek-r1-distill-llama-70b", temperature=0.5)
prompt = "Which one of 9.11 and 9.9 is bigger?"
# Get the response from the model
response = groq_llm.invoke(prompt)
#初始化Cleanlab的studio
studio = Studio("226eeab91e944b23bd817a46dbe3c8ae") 
cleanlab_tlm = studio.TLM(optinotallow={"log": ["explanation"]}) #供解释
#得到包含可信度得分和解释的输出
output = cleanlab_tlm.get_trustworthiness_score(prompt, respnotallow=response.content.strip())
md_content = f"""
## 模型: {model}
**Response:** {response.content.strip()}
**Trustworthiness Score:** {output['trustworthiness_score']}
**Explanation:** {output['log']['explanation']}
---
"""
display(Markdown(md_content))
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.

pip install llama-parse llama-index-core llama-index-embeddings-huggingface 
llama-index-llms-cleanlab requests beautifulsoup4 pdfkit nest-asyncio
1.
2.

from llama_parse import LlamaParse
from llama_index.core import VectorStoreIndex
import requests
from bs4 import BeautifulSoup
import pdfkit
from llama_index.readers.docling import DoclingReader
from llama_index.core import Settings
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
from llama_index.llms.cleanlab import CleanlabTLM
from typing import Dict, List, ClassVar
from llama_index.core.instrumentation.events import BaseEvent
from llama_index.core.instrumentation.event_handlers import BaseEventHandler
from llama_index.core.instrumentation import get_dispatcher
from llama_index.core.instrumentation.events.llm import LLMCompletionEndEvent
import nest_asyncio
import os
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.

options = {
 "model": "gpt-4o",
 "max_tokens": 512,
 "log": ["explanation"]
}
llm = CleanlabTLM(api_key="<CLEANLAB_API_KEY>", optinotallow=options) # 从https://cleanlab.ai/获取您的免费API
Settings.llm = llm
Settings.embed_model = HuggingFaceEmbedding(
 model_name="BAAI/bge-small-en-v1.5"
)
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.

# 可信度评分事件处理程序
class GetTrustworthinessScore(BaseEventHandler):
 events: ClassVar[List[BaseEvent]] = []
 trustworthiness_score: float = 0.0
 @classmethod
 def class_name(cls) -> str:
 return "GetTrustworthinessScore"
 def handle(self, event: BaseEvent) -> Dict:
 if isinstance(event, LLMCompletionEndEvent):
 self.trustworthiness_score = event.response.additional_kwargs.get("trustworthiness_score", 0.0)
 self.events.append(event)
 return {}

# 显示LLM响应的辅助函数
def display_response(response):
 response_str = response.response
 trustworthiness_score = event_handler.trustworthiness_score
 print(f"Response: {response_str}")
 print(f"Trustworthiness score: {round(trustworthiness_score, 2)}")
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.

##########################################
# 从多个URL生成PDF
##########################################
# 配置wkhtmltopdf路径
wkhtml_path = r'C:\Program Files\wkhtmltopdf\bin\wkhtmltopdf.exe'
config = pdfkit.configuration(wkhtmltopdf=wkhtml_path)
# 定义URL和分配文档名称
urls = {
 "LLMs": "https://en.wikipedia.org/wiki/Large_language_model"
}
# 保存PDF的目录
pdf_directory = "PDFs"
os.makedirs(pdf_directory, exist_ok=True)
pdf_paths = {}
for doc_name, url in urls.items():
 try:
 print(f"Processing {doc_name} from {url} ...")
 response = requests.get(url)
 soup = BeautifulSoup(response.text, "html.parser")
 main_content = soup.find("div", {"id": "mw-content-text"})
 if main_content is None:
 raise ValueError("Main content not found")
 # 将协议相对URL替换为绝对URL
 html_string = str(main_content).replace('src="//', 'src="https://').replace('href="//', 'href="https://')
 pdf_file_path = os.path.join(pdf_directory, f"{doc_name}.pdf")
 pdfkit.from_string(
 html_string,
 pdf_file_path,
 optinotallow={'encoding': 'UTF-8', 'quiet': ''},
 cnotallow=config
 )
 pdf_paths[doc_name] = pdf_file_path
 print(f"Saved PDF for {doc_name} at {pdf_file_path}")
 except Exception as e:
 print(f"Error processing {doc_name}: {e}")
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.

##########################################
# 使用LlamaParse解析PDF并注入元数据
##########################################

# 定义解析指令（如果您的解析器支持）
parsing_instructions = """提取文档的markdown格式内容。
按页将文档拆分为节点（例如）。
确保每个节点具有文档名称和页码的元数据。"""

# 创建LlamaParse实例
parser = LlamaParse(
 api_key="<LLAMACLOUD_API_KEY>", # 替换为您的实际密钥
 parsing_instructinotallow=parsing_instructions,
 result_type="markdown",
 premium_mode=True,
 max_timeout=600
)
# 保存合并的Markdown文件的目录（每个PDF一个）
output_md_dir = os.path.join(pdf_directory, "markdown_docs")
os.makedirs(output_md_dir, exist_ok=True)
# 列表，用于保存所有更新后的节点以供索引
all_nodes = []
for doc_name, pdf_path in pdf_paths.items():
 try:
 print(f"Parsing PDF for {doc_name} from {pdf_path} ...")
 nodes = parser.load_data(pdf_path) # 返回节点列表
 updated_nodes = []
 # 处理每个节点：更新元数据并在文本中注入引用标题。
 for i, node in enumerate(nodes, start=1):
 # 复制现有元数据（如果有），并添加我们自己的键。
 new_metadata = dict(node.metadata) if node.metadata else {}
 new_metadata["document_name"] = doc_name
 if "page_number" not in new_metadata:
 new_metadata["page_number"] = str(i)
 # 构建引用标题。
 citation_header = f"[{new_metadata['document_name']}, page {new_metadata['page_number']}]\n\n"
 # 在节点的文本前添加引用标题。
 updated_text = citation_header + node.text
 new_node = node.__class__(text=updated_text, metadata=new_metadata)
 updated_nodes.append(new_node)
 # 使用更新后的节点文本为文档保存一个合并的Markdown文件。
 combined_texts = [node.text for node in updated_nodes]
 combined_md = "\n\n---\n\n".join(combined_texts)
 md_filename = f"{doc_name}.md"
 md_filepath = os.path.join(output_md_dir, md_filename)
 with open(md_filepath, "w", encoding="utf-8") as f:
 f.write(combined_md)
 print(f"Saved combined markdown for {doc_name} to {md_filepath}")
 # 将更新后的节点添加到全局列表以供索引。
 all_nodes.extend(updated_nodes)
 print(f"Parsed {len(updated_nodes)} nodes from {doc_name}.")
 except Exception as e:
 print(f"Error parsing {doc_name}: {e}")
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.

##########################################

# 创建索引和查询引擎

##########################################

# 从所有节点创建索引。

index = VectorStoreIndex.from_documents(documents=all_nodes)

# 定义一个自定义提示模板，强制包含引用。

prompt_template = """
你是一个具有主题专业知识的AI助手。
仅使用提供的上下文回答问题。
在必要时，以格式良好的Markdown格式回答，包含项目符号和章节。
如果提供的上下文不支持答案，请回复“我不知道。”
上下文：
{context_str}
问题：
{query_str}
答案：
"""
# 使用自定义提示创建查询引擎。
query_engine = index.as_query_engine(similarity_top_k=3, llm=llm, prompt_template=prompt_template)
print("Combined index and query engine created successfully!")
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.

query = "When is mixture of experts approach used?"
response = query_engine.query(query)
display_response(response)
1.
2.
3.

query = "How do you compare Deepseek model with OpenAI's models?"
response = query_engine.query(query)
display_response(response)
1.
2.
3.