Add files using upload-large-folder tool

OpenManus/.github/ISSUE_TEMPLATE/config.yaml

@@ -0,0 +1,4 @@
+blank_issues_enabled: false
+contact_links:
+  - name: "📑 Read online docs"
+    about: Find tutorials, use cases, and guides in the OpenManus documentation.

OpenManus/.github/ISSUE_TEMPLATE/show_me_the_bug.md

@@ -0,0 +1,25 @@
+---
+name: "🪲 Show me the Bug"
+about: Report a bug encountered while using OpenManus and seek assistance.
+title: ''
+labels: kind/bug
+assignees: ''
+---
+
+**Bug description**
+<!-- Clearly describe the bug you encountered -->
+
+**Bug solved method**
+<!-- If resolved, explain the solution. Optionally, include a Pull Request URL. -->
+<!-- If unresolved, provide additional details to aid investigation -->
+
+**Environment information**
+<!-- System: e.g., Ubuntu 22.04, Python: e.g., 3.12, OpenManus version: e.g., 0.1.0 -->
+
+- System version:
+- Python version:
+- OpenManus version or branch:
+- Installation method (e.g., `pip install -r requirements.txt` or `pip install -e .`):
+
+**Screenshots or logs**
+<!-- Attach screenshots or logs to help diagnose the issue -->

OpenManus/.github/workflows/build-package.yaml

@@ -0,0 +1,33 @@
+name: Build and upload Python package
+
+on:
+  workflow_dispatch:
+  release:
+    types: [created, published]
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.12'
+          cache: 'pip'
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -r requirements.txt
+          pip install setuptools wheel twine
+      - name: Set package version
+        run: |
+          export VERSION="${GITHUB_REF#refs/tags/v}"
+          sed -i "s/version=.*/version=\"${VERSION}\",/" setup.py
+      - name: Build and publish
+        env:
+          TWINE_USERNAME: __token__
+          TWINE_PASSWORD: ${{ secrets.PYPI_API_TOKEN }}
+        run: |
+          python setup.py bdist_wheel sdist
+          twine upload dist/*

OpenManus/.github/workflows/pre-commit.yaml

@@ -0,0 +1,26 @@
+name: Pre-commit checks
+
+on:
+  pull_request:
+    branches:
+      - '**'
+  push:
+    branches:
+      - '**'
+
+jobs:
+  pre-commit-check:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout Source Code
+        uses: actions/checkout@v4
+      - name: Set up Python 3.12
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.12'
+      - name: Install pre-commit and tools
+        run: |
+          python -m pip install --upgrade pip
+          pip install pre-commit black==23.1.0 isort==5.12.0 autoflake==2.0.1
+      - name: Run pre-commit hooks
+        run: pre-commit run --all-files

OpenManus/.github/workflows/stale.yaml

@@ -0,0 +1,23 @@
+name: Close inactive issues
+
+on:
+  schedule:
+    - cron: "5 0 * * *"
+
+jobs:
+  close-issues:
+    runs-on: ubuntu-latest
+    permissions:
+      issues: write
+      pull-requests: write
+    steps:
+      - uses: actions/stale@v5
+        with:
+          days-before-issue-stale: 30
+          days-before-issue-close: 14
+          stale-issue-label: "inactive"
+          stale-issue-message: "This issue has been inactive for 30 days. Please comment if you have updates."
+          close-issue-message: "This issue was closed due to 45 days of inactivity. Reopen if still relevant."
+          days-before-pr-stale: -1
+          days-before-pr-close: -1
+          repo-token: ${{ secrets.GITHUB_TOKEN }}

deep_search/data_syn/bing_search.py

@@ -0,0 +1,421 @@
+import os
+import json
+import requests
+from requests.exceptions import Timeout
+from bs4 import BeautifulSoup
+from tqdm import tqdm
+import time
+import concurrent
+from concurrent.futures import ThreadPoolExecutor
+import pdfplumber
+from io import BytesIO
+import re
+import string
+from typing import Optional, Tuple
+from nltk.tokenize import sent_tokenize
+
+# os.environ['http_proxy'] = 'http://127.0.0.1:7890'
+# os.environ['https_proxy'] = 'http://127.0.0.1:7890'
+
+
+# ----------------------- Custom Headers -----------------------
+headers = {
+    'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) '
+                  'AppleWebKit/537.36 (KHTML, like Gecko) '
+                  'Chrome/58.0.3029.110 Safari/537.36',
+    'Referer': 'https://www.google.com/',
+    'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8',
+    'Accept-Language': 'en-US,en;q=0.5',
+    'Connection': 'keep-alive',
+    'Upgrade-Insecure-Requests': '1'
+}
+
+# Initialize session
+session = requests.Session()
+session.headers.update(headers)
+
+
+
+def remove_punctuation(text: str) -> str:
+    """Remove punctuation from the text."""
+    return text.translate(str.maketrans("", "", string.punctuation))
+
+def f1_score(true_set: set, pred_set: set) -> float:
+    """Calculate the F1 score between two sets of words."""
+    intersection = len(true_set.intersection(pred_set))
+    if not intersection:
+        return 0.0
+    precision = intersection / float(len(pred_set))
+    recall = intersection / float(len(true_set))
+    return 2 * (precision * recall) / (precision + recall)
+
+def extract_snippet_with_context(full_text: str, snippet: str, context_chars: int = 2500) -> Tuple[bool, str]:
+    """
+    Extract the sentence that best matches the snippet and its context from the full text.
+
+    Args:
+        full_text (str): The full text extracted from the webpage.
+        snippet (str): The snippet to match.
+        context_chars (int): Number of characters to include before and after the snippet.
+
+    Returns:
+        Tuple[bool, str]: The first element indicates whether extraction was successful, the second element is the extracted context.
+    # 这个 extract_snippet_with_context 函数的作用是 从一段长文本中找到最符合给定片段（snippet）的句子，并返回包含该句子的一定上下文范围的文本。它的核心逻辑包括 文本预处理、句子匹配、F1 评分计算、上下文截取 等几个步骤。
+    """
+    try:
+        full_text = full_text[:50000]
+
+        snippet = snippet.lower()
+        snippet = remove_punctuation(snippet)
+        snippet_words = set(snippet.split())
+
+        best_sentence = None
+        best_f1 = 0.2
+
+        # sentences = re.split(r'(?<=[.!?]) +', full_text)  # Split sentences using regex, supporting ., !, ? endings
+        sentences = sent_tokenize(full_text)  # Split sentences using nltk's sent_tokenize
+
+        for sentence in sentences:
+            key_sentence = sentence.lower()
+            key_sentence = remove_punctuation(key_sentence)
+            sentence_words = set(key_sentence.split())
+            f1 = f1_score(snippet_words, sentence_words)
+            if f1 > best_f1:
+                best_f1 = f1
+                best_sentence = sentence
+
+        if best_sentence:
+            para_start = full_text.find(best_sentence)
+            para_end = para_start + len(best_sentence)
+            start_index = max(0, para_start - context_chars)
+            end_index = min(len(full_text), para_end + context_chars)
+            context = full_text[start_index:end_index]
+            return True, context
+        else:
+            # If no matching sentence is found, return the first context_chars*2 characters of the full text
+            return False, full_text[:context_chars * 2]
+    except Exception as e:
+        return False, f"Failed to extract snippet context due to {str(e)}"
+
+def extract_text_from_url(url, use_jina=False, jina_api_key=None, snippet: Optional[str] = None):
+    """
+    Extract text from a URL. If a snippet is provided, extract the context related to it.
+
+    Args:
+        url (str): URL of a webpage or PDF.
+        use_jina (bool): Whether to use Jina for extraction.
+        snippet (Optional[str]): The snippet to search for.
+
+    Returns:
+        str: Extracted text or context.
+    """
+    try:
+        print(f"extract_text_from_url use_jina: {use_jina}")
+        if use_jina:
+            jina_headers = {
+                'Authorization': f'Bearer {jina_api_key}',
+                'X-Return-Format': 'markdown',
+                # 'X-With-Links-Summary': 'true'
+            }
+            response = requests.get(f'https://r.jina.ai/{url}', headers=jina_headers).text
+            # Remove URLs
+            pattern = r"\(https?:.*?\)|\[https?:.*?\]"
+            text = re.sub(pattern, "", response).replace('---','-').replace('===','=').replace('   ',' ').replace('   ',' ')
+            print("use jina to extract text successfully")
+        else:
+            print(f"don't use jina to extract text")
+            response = session.get(url, timeout=20)  # Set timeout to 20 seconds
+            response.raise_for_status()  # Raise HTTPError if the request failed
+            # Determine the content type
+            content_type = response.headers.get('Content-Type', '')
+            if 'pdf' in content_type:
+                # If it's a PDF file, extract PDF text
+                print("Extracting text from PDF...")
+                return extract_pdf_text(url)
+            # Try using lxml parser, fallback to html.parser if unavailable
+            try:
+                soup = BeautifulSoup(response.text, 'lxml')
+            except Exception:
+                print("lxml parser not found or failed, falling back to html.parser")
+                soup = BeautifulSoup(response.text, 'html.parser')
+            text = soup.get_text(separator=' ', strip=True)
+
+        if snippet:
+            success, context = extract_snippet_with_context(text, snippet)
+            if success:
+                return context
+            else:
+                return text
+        else:
+            # If no snippet is provided, return directly
+            return text[:8000]
+    except requests.exceptions.HTTPError as http_err:
+        return f"HTTP error occurred: {http_err}"
+    except requests.exceptions.ConnectionError:
+        return "Error: Connection error occurred"
+    except requests.exceptions.Timeout:
+        return "Error: Request timed out after 20 seconds"
+    except Exception as e:
+        return f"Unexpected error: {str(e)}"
+
+def fetch_page_content(urls, max_workers=4, use_jina=False, jina_api_key=None, snippets: Optional[dict] = None):
+    """
+    Concurrently fetch content from multiple URLs.
+
+    Args:
+        urls (list): List of URLs to scrape.
+        max_workers (int): Maximum number of concurrent threads.
+        use_jina (bool): Whether to use Jina for extraction.
+        snippets (Optional[dict]): A dictionary mapping URLs to their respective snippets.
+
+    Returns:
+        dict: A dictionary mapping URLs to the extracted content or context.
+    这段代码定义了一个名为 fetch_page_content 的函数，用于并发地从多个 URL 中提取内容。该函数使用 ThreadPoolExecutor 来并发执行提取操作
+    """
+    results = {}
+    with ThreadPoolExecutor(max_workers=64) as executor:
+        # Use tqdm to display a progress bar
+        futures = {
+            executor.submit(extract_text_from_url, url, use_jina, jina_api_key,snippets.get(url) if snippets else None): url
+            for url in urls
+        }
+        for future in tqdm(concurrent.futures.as_completed(futures), desc="Fetching URLs", total=len(urls)):
+            url = futures[future]
+            try:
+                data = future.result()
+                results[url] = data
+            except Exception as exc:
+                results[url] = f"Error fetching {url}: {exc}"
+            time.sleep(0.2)  # Simple rate limiting
+    return results
+
+
+# def bing_web_search(query, subscription_key, endpoint, market='en-US', language='en', timeout=20):
+#     """
+#     Perform a search using the Bing Web Search API with a set timeout.
+
+#     Args:
+#         query (str): Search query.
+#         subscription_key (str): Subscription key for the Bing Search API.
+#         endpoint (str): Endpoint for the Bing Search API.
+#         market (str): Market, e.g., "en-US" or "zh-CN".
+#         language (str): Language of the results, e.g., "en".
+#         timeout (int or float or tuple): Request timeout in seconds.
+#                                          Can be a float representing the total timeout,
+#                                          or a tuple (connect timeout, read timeout).
+
+#     Returns:
+#         dict: JSON response of the search results. Returns None or raises an exception if the request times out.
+#     函数的目标是使用 Bing Web Search API 执行搜索，并返回 JSON 格式的结果。
+#     如果请求超时或出现其他问题，返回空字典（{}）或抛出异常
+#     """
+#     headers = {
+#         "Ocp-Apim-Subscription-Key": subscription_key
+#     }
+#     params = {
+#         "q": query,
+#         "mkt": market,
+#         "setLang": language,
+#         "textDecorations": True,
+#         "textFormat": "HTML"
+#     }
+
+#     try:
+#         response = requests.get(endpoint, headers=headers, params=params, timeout=timeout)
+#         response.raise_for_status()  # Raise exception if the request failed 检查响应的状态码。如果返回的状态码是 4xx 或 5xx（表示客户端或服务器错误），它将引发 requests.exceptions.HTTPError 异常
+#         search_results = response.json() # response.json() 会返回一个 Python 字典（或类似字典的对象），其中包含了 Bing Web Search 返回的搜索结果
+#         return search_results
+#     except Timeout:
+#         print(f"Bing Web Search request timed out ({timeout} seconds) for query: {query}")
+#         return {}  # Or you can choose to raise an exception
+#     except requests.exceptions.RequestException as e:
+#         print(f"Error occurred during Bing Web Search request: {e}")
+#         return {}
+
+proxies = {
+    "http": "http://127.0.0.1:7880",
+    "https": "http://127.0.0.1:7880"
+}
+
+
+def bing_web_search(query, subscription_key, endpoint, market='en-US', language='en', timeout=2000):
+    """
+    Perform a search using the Bing Web Search API with a set timeout.
+
+    Args:
+        query (str): Search query.
+        subscription_key (str): Subscription key for the Bing Search API.
+        endpoint (str): Endpoint for the Bing Search API.
+        market (str): Market, e.g., "en-US" or "zh-CN".
+        language (str): Language of the results, e.g., "en".
+        timeout (int or float or tuple): Request timeout in seconds.
+                                         Can be a float representing the total timeout,
+                                         or a tuple (connect timeout, read timeout).
+
+    Returns:
+        dict: JSON response of the search results. Returns None or raises an exception if the request times out.
+    函数的目标是使用 Bing Web Search API 执行搜索，并返回 JSON 格式的结果。
+    如果请求超时或出现其他问题，返回空字典（{}）或抛出异常
+    """
+    payload = json.dumps({
+        "q": query,  # 设置查询内容
+        "mkt": market,  # 设置市场
+        "setLang": language,  # 设置语言
+        "textDecorations": True,  # 启用文本装饰
+        "textFormat": "HTML"  # 设置文本格式
+    })
+
+    headers = {
+        'X-API-KEY': subscription_key,
+        'Content-Type': 'application/json'
+    }
+    error_cnt = 0
+    while True:
+        if error_cnt == 20:
+            print(f"qery: {query} has tried {error_cnt} times without success, just skip it.")
+            break
+        try:
+            # 发送POST请求
+            response = requests.request("POST", endpoint, headers=headers, data=payload, proxies=proxies, timeout=timeout)
+            response.raise_for_status()  # Raise exception if the request failed 检查响应的状态码。如果返回的状态码是 4xx 或 5xx（表示客户端或服务器错误），它将引发 requests.exceptions.HTTPError 异常
+            search_results = response.json() #
+            return search_results
+        except Timeout:
+            error_cnt += 1
+            print(f"error_cnt: {error_cnt}, Bing Web Search request timed out ({timeout} seconds) for query: {query}")
+            time.sleep(5)
+            # return {}  # Or you can choose to raise an exception
+        except requests.exceptions.RequestException as e:
+            error_cnt += 1
+            print(f"error_cnt: {error_cnt}, Error occurred during Bing Web Search request: {e}, payload: {payload}")
+            time.sleep(5)
+            # return {}
+
+
+def extract_pdf_text(url):
+    """
+    Extract text from a PDF.
+
+    Args:
+        url (str): URL of the PDF file.
+
+    Returns:
+        str: Extracted text content or error message.
+    """
+    try:
+        response = session.get(url, timeout=20)  # Set timeout to 20 seconds
+        if response.status_code != 200:
+            return f"Error: Unable to retrieve the PDF (status code {response.status_code})"
+        
+        # Open the PDF file using pdfplumber
+        with pdfplumber.open(BytesIO(response.content)) as pdf:
+            full_text = ""
+            for page in pdf.pages:
+                text = page.extract_text()
+                if text:
+                    full_text += text
+        
+        # Limit the text length
+        cleaned_text = ' '.join(full_text.split()[:600])
+        return cleaned_text
+    except requests.exceptions.Timeout:
+        return "Error: Request timed out after 20 seconds"
+    except Exception as e:
+        return f"Error: {str(e)}"
+
+# def extract_relevant_info(search_results):
+#     """
+#     Extract relevant information from Bing search results.
+
+#     Args:
+#         search_results (dict): JSON response from the Bing Web Search API.
+
+#     Returns:
+#         list: A list of dictionaries containing the extracted information.
+#     """
+#     useful_info = []
+    
+#     if 'webPages' in search_results and 'value' in search_results['webPages']: # value 通常是一个列表，包含了搜索结果的每个页面信息
+#         for id, result in enumerate(search_results['webPages']['value']):
+#             info = {
+#                 'id': id + 1,  # Increment id for easier subsequent operations 为每个结果分配一个 id，id + 1 是为了让 ID 从 1 开始，而不是从 0 开始。这对后续操作更直观
+#                 'title': result.get('name', ''), # 每个搜索结果中提取标题
+#                 'url': result.get('url', ''), # 每个搜索结果中提取 URL
+#                 'site_name': result.get('siteName', ''), # 每个搜索结果中提取站点名称
+#                 'date': result.get('datePublished', '').split('T')[0], # 提取���索结果的发布时间
+#                 'snippet': result.get('snippet', ''),  # Remove HTML tags : 提取搜索结果的简短描述（即摘要或片段），result.get('snippet', '')。这里的 snippet 可能包含 HTML 标签，因此需要在后续的处理中可能会清除这些标签
+#                 # Add context content to the information
+#                 'context': ''  # Reserved field to be filled later
+#             }
+#             useful_info.append(info)
+    
+#     return useful_info
+
+def extract_relevant_info(search_results):
+    """
+    Extract relevant information from Bing search results.
+
+    Args:
+        search_results (dict): JSON response from the Bing Web Search API.
+
+    Returns:
+        list: A list of dictionaries containing the extracted information.
+    """
+    useful_info = []
+    
+    if 'organic' in search_results : # value 通常是一个列表，包含了搜索结果的每个页面信息
+        for id, result in enumerate(search_results['organic']):
+            info = {
+                'id': id + 1,  # Increment id for easier subsequent operations 为每个结果分配一个 id，id + 1 是为了让 ID 从 1 开始，而不是从 0 开始。这对后续操作更直观
+                'title': result.get('title', ''), # 每个搜索结果中提取标题
+                'url': result.get('link', ''), # 每个搜索结果中提取 URL
+                'site_name': result.get('siteName', ''), # 每个搜索结果中提取站点名称
+                'date': result.get('datePublished', '').split('T')[0], # 提取搜索结果的发布时间
+                'snippet': result.get('snippet', ''),  # Remove HTML tags : 提取搜索结果的简短描述（即摘要或片段），result.get('snippet', '')。这里的 snippet 可能包含 HTML 标签，因此需要在后续的处理中可能会清除这些标签
+                # Add context content to the information
+                'context': ''  # Reserved field to be filled later
+            }
+            useful_info.append(info)
+    
+    return useful_info
+
+
+# ------------------------------------------------------------
+
+if __name__ == "__main__":
+    # Example usage
+    # Define the query to search
+    query = "Structure of dimethyl fumarate"
+    
+    # Subscription key and endpoint for Bing Search API
+    BING_SUBSCRIPTION_KEY = "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d"
+    if not BING_SUBSCRIPTION_KEY:
+        raise ValueError("Please set the BING_SEARCH_V7_SUBSCRIPTION_KEY environment variable.")
+    
+    bing_endpoint = "https://google.serper.dev/search"
+    
+    # Perform the search
+    print("Performing Bing Web Search...")
+    search_results = bing_web_search(query, BING_SUBSCRIPTION_KEY, bing_endpoint)
+    
+    print("Extracting relevant information from search results...")
+    extracted_info = extract_relevant_info(search_results)
+
+    print("Fetching and extracting context for each snippet...")
+    for info in tqdm(extracted_info, desc="Processing Snippets"):
+        full_text = extract_text_from_url(info['url'], use_jina=False, jina_api_key="jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ", snippet=info["snippet"])  # Get full webpage text
+        if full_text and not full_text.startswith("Error"):
+            success, context = extract_snippet_with_context(full_text, info['snippet'])
+            if success:
+                info['context'] = context
+                print("-------------------")
+                print(f"Snippet: {info['snippet']}\nContext: {context}")
+                
+            else:
+                info['context'] = f"Could not extract context. Returning first 8000 chars: {full_text[:8000]}"
+        else:
+            info['context'] = f"Failed to fetch full text: {full_text}"
+
+    # print("Your Search Query:", query)
+    # print("Final extracted information with context:")
+    # print(json.dumps(extracted_info, indent=2, ensure_ascii=False))

deep_search/data_syn/data_keypoints_stats.py

@@ -0,0 +1,50 @@
+import json
+from collections import Counter, OrderedDict
+import matplotlib.pyplot as plt
+
+# 读取数据
+file_path = '/share/project/sunshuang/deep_search/data_for_rl/musique_tagged/final_selected_dataset.json'
+
+with open(file_path, 'r') as f:
+    data = json.load(f)
+
+# 提取 key_points 并统计每个关键词的出现次数
+key_points_counter = Counter()
+
+for entry in data:
+    domain_keypoints = entry.get('domain_keypoints', {})
+    key_points = domain_keypoints.get('key_points', [])
+    key_points_counter.update(key_points)
+
+
+
+key_points_counter = OrderedDict(sorted(key_points_counter.items(), key=lambda item: item[1], reverse=True))
+# 打印统计结果
+# print("Key Points Count:")
+# for key, count in key_points_counter.items():
+#     print(f"{key}: {count}")
+
+# 将统计结果保存为 JSON 文件
+key_points_count_file = 'key_points_count.json'
+with open(key_points_count_file, 'w') as f:
+    json.dump(dict(key_points_counter), f, indent=4)
+
+print(f"Key points count saved to '{key_points_count_file}'")
+
+# 绘制饼图
+labels = [f'{key} ({count})' for key, count in key_points_counter.items()]
+sizes = list(key_points_counter.values())
+
+plt.figure(figsize=(12, 7))
+plt.pie(
+    sizes,
+    labels=labels,
+    autopct='%1.1f%%',
+    startangle=140
+)
+plt.title('Key Points Distribution')
+plt.tight_layout()
+plt.savefig('key_points_distribution_pie.png', bbox_inches='tight')
+plt.close()
+
+print("Pie chart saved as 'key_points_distribution_pie.png'")

deep_search/data_syn/data_stats.py

@@ -0,0 +1,250 @@
+# # import json
+# # import matplotlib.pyplot as plt
+# # from collections import Counter
+
+# # # 读取数据
+# # file_path = '/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/tagged_domain_keypoints/merged_tagged_domain_keypoints_keywords_count_hop.json'
+
+# # with open(file_path, 'r') as f:
+# #     data = json.load(f)
+
+# # # 提取所需字段
+# # domains = []
+# # totals = []
+# # special_totals = []
+# # hops = []
+
+# # for entry in data:
+# #     domain_info = entry.get('domain_keypoints', {})
+# #     # print(domain_info)
+# #     # print(f"idx: {entry['idx']}")
+# #     keywords_info = entry.get('keywords_count', {})
+    
+# #     domains.append(domain_info.get('domain', 'Unknown'))
+# #     totals.append(keywords_info.get('total', 0))
+# #     special_totals.append(keywords_info.get('special_total', 0))
+# #     hops.append(entry.get('hop', 0))
+
+# # # 统计domain分布并绘制饼图
+# # domain_counts = Counter(domains)
+# # domain_labels = [f'{dom} ({cnt})' for dom, cnt in domain_counts.items()]
+# # domain_sizes = list(domain_counts.values())
+
+# # plt.figure(figsize=(12, 7))
+# # plt.pie(domain_sizes, labels=domain_labels, autopct='%1.1f%%', startangle=140)
+# # plt.title('Domain Distribution (Count in Parentheses)')
+# # plt.savefig('domain_distribution_pie.png')
+# # plt.close()
+
+# # # 辅助函数：绘制带数值标签的直方图
+# # def plot_histogram(data, title, xlabel, output_filename):
+# #     counts = Counter(data)
+# #     sorted_items = sorted(counts.items())
+# #     labels, values = zip(*sorted_items)
+    
+# #     plt.figure(figsize=(10, 6))
+# #     bars = plt.bar(labels, values)
+# #     plt.title(title)
+# #     plt.xlabel(xlabel)
+# #     plt.ylabel('Count')
+    
+# #     for bar in bars:
+# #         height = bar.get_height()
+# #         plt.text(bar.get_x() + bar.get_width()/2., height, str(height),
+# #                  ha='center', va='bottom')
+    
+# #     plt.savefig(output_filename)
+# #     plt.close()
+
+# # # 绘制各个直方图
+# # plot_histogram(totals, 'Total Distribution', 'Total Value', 'total_histogram.png')
+# # plot_histogram(special_totals, 'Special Total Distribution', 'Special Total Value', 'special_total_histogram.png')
+# # plot_histogram(hops, 'Hop Distribution', 'Hop Value', 'hop_histogram.png')
+
+# import json
+# import matplotlib.pyplot as plt
+# from collections import Counter, OrderedDict
+
+# # 读取数据
+# file_path = '/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/tagged_domain_keypoints/merged_tagged_domain_keypoints_keywords_count_hop.json'
+
+# with open(file_path, 'r') as f:
+#     data = json.load(f)
+
+# # 提取所需字段
+# domains = []
+# totals = []
+# special_totals = []
+# hops = []
+
+# for entry in data:
+#     domain_info = entry.get('domain_keypoints', {})
+#     keywords_info = entry.get('keywords_count', {})
+    
+#     domains.append(domain_info.get('domain', 'Unknown'))
+#     totals.append(keywords_info.get('total', 0))
+#     special_totals.append(keywords_info.get('special_total', 0))
+#     hops.append(entry.get('hop', 0))
+
+# # 统计domain分布
+# domain_counts = Counter(domains)
+
+# # 按照 value 降序排序
+# domain_counts = OrderedDict(sorted(domain_counts.items(), key=lambda item: item[1], reverse=True))
+
+# # 保存所有domain及其计数到JSON文件
+# domain_count_file = 'domain_counts.json'
+# with open(domain_count_file, 'w') as f:
+#     json.dump(domain_counts, f, indent=4)
+
+# # 合并数目少于10的domain为Other
+# threshold = 100
+# filtered_domains = {}
+# other_count = 0
+
+# for domain, count in domain_counts.items():
+#     if count < threshold:
+#         other_count += count
+#     else:
+#         filtered_domains[domain] = count
+
+# if other_count > 0:
+#     filtered_domains['Other'] = other_count
+
+# # 绘制饼图
+# domain_labels = [f'{dom} ({cnt})' for dom, cnt in filtered_domains.items()]
+# domain_sizes = list(filtered_domains.values())
+
+# plt.figure(figsize=(12, 7))
+# plt.pie(domain_sizes, labels=domain_labels, autopct='%1.1f%%', startangle=140)
+# plt.title('Domain Distribution (Count in Parentheses)')
+# plt.savefig('domain_distribution_pie.png')
+# plt.close()
+
+# # 辅助函数：绘制带数值标签的直方图
+# def plot_histogram(data, title, xlabel, output_filename):
+#     counts = Counter(data)
+#     sorted_items = sorted(counts.items())
+#     labels, values = zip(*sorted_items)
+    
+#     plt.figure(figsize=(10, 6))
+#     bars = plt.bar(labels, values)
+#     plt.title(title)
+#     plt.xlabel(xlabel)
+#     plt.ylabel('Count')
+    
+#     for bar in bars:
+#         height = bar.get_height()
+#         plt.text(bar.get_x() + bar.get_width()/2., height, str(height),
+#                  ha='center', va='bottom')
+    
+#     plt.savefig(output_filename)
+#     plt.close()
+
+# # 绘制各个直方图
+# plot_histogram(totals, 'Total Distribution', 'Total Value', 'total_histogram.png')
+# plot_histogram(special_totals, 'Special Total Distribution', 'Special Total Value', 'special_total_histogram.png')
+# plot_histogram(hops, 'Hop Distribution', 'Hop Value', 'hop_histogram.png')
+
+
+
+import json
+import matplotlib
+import matplotlib.pyplot as plt
+from collections import Counter, OrderedDict
+matplotlib.rcParams['font.family'] = ['Songti SC']  # macOS 用户用宋体
+# 读取数据
+file_path = '/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/merged_4k/4k.json'
+
+with open(file_path, 'r') as f:
+    data = json.load(f)
+
+# 提取所需字段
+domains = []
+totals = []
+special_totals = []
+hops = []
+
+for entry in data:
+    domain_info = entry.get('domain_keypoints', {})
+    keywords_info = entry.get('keywords_count', {})
+    
+    domains.append(domain_info.get('domain', 'Unknown'))
+    totals.append(keywords_info.get('total', 0))
+    special_totals.append(keywords_info.get('special_total', 0))
+    hops.append(entry.get('hop', 0))
+
+# 统计domain分布
+domain_counts = Counter(domains)
+
+# 按照 value 降序排序
+domain_counts = OrderedDict(sorted(domain_counts.items(), key=lambda item: item[1], reverse=True))
+
+# 保存所有domain及其计数到JSON文件
+domain_count_file = 'domain_counts.json'
+with open(domain_count_file, 'w') as f:
+    json.dump(domain_counts, f, indent=4)
+
+# 合并数目少于100的domain为Other
+threshold = 1
+filtered_domains = {}
+other_count = 0
+
+for domain, count in domain_counts.items():
+    if count < threshold:
+        other_count += count
+    else:
+        filtered_domains[domain] = count
+
+if other_count > 0:
+    filtered_domains['Other'] = other_count
+
+# 绘制饼图
+domain_labels = [f'{dom}' for dom, cnt in filtered_domains.items()]  # 只显示domain名称
+domain_sizes = list(filtered_domains.values())
+
+plt.figure(figsize=(12, 7))
+plt.pie(
+    domain_sizes,
+    labels=None,  # 不直接在饼图上显示标签
+    autopct='%1.1f%%',
+    startangle=140
+)
+
+# 添加图例
+plt.legend(
+    labels=[f'{dom} ({cnt})' for dom, cnt in filtered_domains.items()],
+    loc='upper right',
+    bbox_to_anchor=(1.2, 1),  # 图例位置调整到右侧
+    fontsize=10
+)
+
+plt.title('Domain Distribution (Count in Parentheses)')
+plt.tight_layout()  # 优化布局
+plt.savefig('domain_distribution_pie.png', bbox_inches='tight', dpi=1000)  # 避免裁剪
+plt.close()
+
+# 辅助函数：绘制带数值标签的直方图
+def plot_histogram(data, title, xlabel, output_filename):
+    counts = Counter(data)
+    sorted_items = sorted(counts.items())
+    labels, values = zip(*sorted_items)
+    
+    plt.figure(figsize=(10, 6))
+    bars = plt.bar(labels, values)
+    plt.title(title)
+    plt.xlabel(xlabel)
+    plt.ylabel('数量')
+    
+    for bar in bars:
+        height = bar.get_height()
+        plt.text(bar.get_x() + bar.get_width()/2., height, str(height),
+                 ha='center', va='bottom')
+    
+    plt.savefig(output_filename)
+    plt.close()
+
+# 绘制各个直方图
+# plot_histogram(totals, 'Total Distribution', 'Total Value', 'total_histogram.png')
+plot_histogram(special_totals, 'Special Question Words Distribution', '问题中疑问词数目', 'special_total_histogram.png')
+# plot_histogram(hops, 'Hop Distribution', 'Hop Value', 'hop_histogram.png')

deep_search/data_syn/data_tag.py

@@ -0,0 +1,186 @@
+import os
+import argparse
+import json
+from vllm import LLM, SamplingParams
+from datasets import Dataset
+from transformers import AutoTokenizer
+import random
+from tqdm import tqdm
+
+# os.environ["CUDA_VISIBLE_DEVICES"] = "0,1"
+
+
+user_prompt = f"""You are an expert in complex problem reasoning. Below are definitions and examples of various reasoning graph shapes. Following these, you will receive a candidate question. Your task is to:
+
+Analyze the reasoning graph shape of the candidate question, using the provided examples as a guide.
+Determine the subject area of the question (choose from options such as Music, Sports, Geography, Art, Politics, Science and Technology, Video Games, History, TV Shows, or Other).
+Identify the key factual elements needed to answer the question. 
+
+```
+[Difinitions and examples of various reasoning graph shapes]:
+
++ 1. 2-hop Reasoning (2-step chain)
+Basic Pattern: A simple linear chain with two consecutive steps, no branching or cycles.
+Example Question:
+"Who succeeded the first President of Namibia?"
+Decomposition:
+Step 1: Identify the first President of Namibia.
+Step 2: Determine who succeeded that president.
+
++ 2. 3-hop Reasoning (3-step chain)
+Basic Pattern: A linear chain with three consecutive steps, still without any branching.
+Example Question:
+"What currency is used where Billy Giles died?"
+Decomposition:
+Step 1: Find out where Billy Giles died.
+Step 2: Determine which country or region that place belongs to.
+Step 3: Identify the currency used in that country or region.
+
++ 3. 3-hop Reasoning with Branching (3-step chain + branching)
+Basic Pattern: Three steps where the first two steps run in parallel (branching), and the final step combines the results.
+Example Question:
+"When was the first establishment that McDonaldization is named after, open in the country Horndean is located?"
+Decomposition:
+Branch 1: Determine the country where Horndean is located.
+Branch 2: Identify the establishment that McDonaldization is named after (and find out details such as its first branch).
+Final Step: Merge the information from both branches to determine when that establishment opened in the identified country.
+
++ 4. 4-hop Reasoning (4-step chain)
+Basic Pattern: A linear chain requiring four consecutive steps, with no branching.
+Example Question:
+"When did Napoleon occupy the city where the mother of the woman who brought Louis XVI style to the court died?"
+Decomposition:
+Step 1: Identify the woman who brought Louis XVI style to the court.
+Step 2: Determine who her mother was.
+Step 3: Find out in which city her mother died.
+Step 4: Ascertain when Napoleon occupied that city.
+
++ 5. 4-hop Reasoning with Branching (4-step chain + branching)
+Basic Pattern: A four-step reasoning process that involves parallel queries (branching) before combining the results.
+Example Question:
+"How many Germans live in the colonial holding in Aruba’s continent that was governed by Prazeres’s country?"
+Decomposition:
+Branch 1: Identify which continent Aruba is located in.
+Branch 2: Determine which country Prazeres is from.
+Step 3: Identify the colonial holding on that continent governed by the country from Branch 2.
+Step 4: Find out how many Germans live in that colonial holding.
+
++ 6. 3-hop Reasoning with Branching and Cycle (3-step chain + branching + cycle)
+Basic Pattern: A three-step process with parallel branches that later require recombination, with a cyclic element where earlier results are referenced again in the final step.
+Example Question:
+"When did the people who captured Malakoff come to the region where Philipsburg is located?"
+Decomposition:
+Branch 1: Determine the region where Philipsburg is located.
+Branch 2: Identify who captured Malakoff.
+Final Step: Combine the results from Branch 1 and Branch 2—this involves a cyclic reference to the earlier information—to determine when those people arrived in the region.
+
++ Overall Summary:
+
+n-hop: Refers to the number of steps or connected information points needed to reach the answer.
+Branching: Occurs when multiple independent queries (sub-questions) need to be solved in parallel before being combined.
+Cycle: Involves a loop or a back-reference where later steps depend on revisiting or combining earlier results.
+
+[Examples of key points]
+Example 1:
+Question: “Are director of film Move (1970 Film) and director of film Mediterranee (1963 Film) from the same country?”
+Key points: director, country.
+
+Example 2:
+Question: "Which film whose director is younger, Charge It To Me or Danger: Diabolik?"
+Key points: director, age.
+
+Example 4:
+Question: "Does Mukasa Mbidde have the same nationality as Erich Maas?"
+Key points: nationality.
+
+Example 5:
+Question: "Which film was released more recently, Die schöne Lurette or Sabhash?"
+Key points: film, release date.
+
+**Notably, name is not the key point.**
+```
+
+[Candidate Question]
+{{question}}
+
+**Please do not generate any analysis and exactly format your answer  as follows:
+
+Reasoning Graph: [the name of predicted reasoning graph here, do not generate other analysis]
+Area: [the subject area]
+Key Points: [the key factual elements]**"""
+
+
+def main(input_file_path, output_path):
+    
+    data = []
+    with open(input_file_path, "r", encoding="utf-8") as file:
+        # for line in file:
+        #     data.append(json.loads(line))
+        data = json.load(file)
+
+    
+    prompts = []
+    new_data = []
+
+    for item in tqdm(data, desc="gen prompts"):
+        prompt = user_prompt.format(question=item["Question"])
+        # item = f"请解释下面的内容：{item}"
+        # message = {"role": "user", "content": item}
+        prompt = {"role": "user", "content": prompt}
+        
+        message = tokenizer.apply_chat_template(
+            [prompt], add_generation_prompt=True, tokenize=False
+        )
+        prompts.append(message)
+        item["tag_prompt"] = message
+        # new_data.append(prompt)
+    print(prompts[0])
+    outputs = llm.generate(prompts, sampling_params)
+
+    generated_texts = []
+    for i, output in tqdm(enumerate(outputs), desc="process outputs"):
+        # prompt = output.prompt
+        generated_text = output.outputs[0].text.strip()
+        data[i]["tag_qwq"] = generated_text
+
+        generated_texts.append(data[i])
+
+    
+    os.makedirs(os.path.dirname(output_path), exist_ok=True) 
+    with open(output_path, "w", encoding="utf-8") as json_file:
+        json.dump(generated_texts, json_file, ensure_ascii=False, indent=4)
+    # with open(output_path, 'w', encoding='utf-8') as f:
+    #     for item in generated_texts:
+    #         f.write(json.dumps(item, ensure_ascii=False) + '\n')
+
+    print(f"save to {output_path}")
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Process input file and CUDA devices.")
+    parser.add_argument("--input_file_path", type=str, required=True, help="Path to the input JSON file.")
+    parser.add_argument("--cuda_visible_devices", type=str, required=True, help="CUDA devices to use (e.g., '0,1').")
+    args = parser.parse_args()
+
+    os.environ["CUDA_VISIBLE_DEVICES"] = args.cuda_visible_devices
+    input_file_path = args.input_file_path
+
+
+    model_name = "qwq"
+    model_path = "/share/project/zhipengchen/model/QwQ-32B"
+    
+    
+    # input_file_path = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/split_1.json"
+    file_name = input_file_path.split(".")[0]
+    output_path = f"{file_name}_tagged.json"
+
+    tokenizer = AutoTokenizer.from_pretrained(model_path)
+    temperature = 0.3
+    sampling_params = SamplingParams(n=1, temperature=temperature, max_tokens=8000, top_p=0.99, top_k=50)
+    llm = LLM(model=model_path, tensor_parallel_size=2, gpu_memory_utilization=0.95,trust_remote_code=True)
+    
+    print(f"CUDA_VISIBLE_DEVICES: {os.environ.get('CUDA_VISIBLE_DEVICES')}")
+    print(f"model path: {model_path}")
+    print(f"temperature: {temperature}")
+    print(f"input_file_path: {input_file_path}")
+
+    main(input_file_path, output_path)
+

deep_search/data_syn/data_tag.sh

@@ -0,0 +1,113 @@
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag.py \
+    --input_file_path  "data/mixed_data/splits/split_1.json" \
+    --cuda_visible_devices "0,1" > data/mixed_data/logs/split_1.log 2>&1 &
+
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag.py \
+    --input_file_path  "data/mixed_data/splits/split_2.json" \
+    --cuda_visible_devices "2,3" > data/mixed_data/logs/split_2.log 2>&1 &
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag.py \
+    --input_file_path  "data/mixed_data/splits/split_3.json" \
+    --cuda_visible_devices "4,5" > data/mixed_data/logs/split_3.log 2>&1 &
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag.py \
+    --input_file_path  "data/mixed_data/splits/split_4.json" \
+    --cuda_visible_devices "6,7" > data/mixed_data/logs/split_4.log 2>&1 &
+
+
+# worker 3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag.py \
+    --input_file_path  "data/mixed_data/splits/split_1.json" \
+    --cuda_visible_devices "0,1" > data/mixed_data/logs/split_1.log 2>&1 &
+
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag.py \
+    --input_file_path  "data/mixed_data/splits/split_2.json" \
+    --cuda_visible_devices "2,3" > data/mixed_data/logs/split_2.log 2>&1 &
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag.py \
+    --input_file_path  "data/mixed_data/splits/split_3.json" \
+    --cuda_visible_devices "4,5" > data/mixed_data/logs/split_3.log 2>&1 &
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag.py \
+    --input_file_path  "data/mixed_data/splits/split_4.json" \
+    --cuda_visible_devices "6,7" > data/mixed_data/logs/split_4.log 2>&1 &
+
+###############  tag domain and key points  ###############
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag_domain_keypoints.py \
+    --input_file_path  "data/mixed_data/splits/split_4.json" \
+    --cuda_visible_devices "6,7" > data/mixed_data/logs/tag_domain_keypoints_split_4.log 2>&1 &
+
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag_domain_keypoints.py \
+    --input_file_path  "data/mixed_data/splits/split_1.json" \
+    --cuda_visible_devices "0,1" > data/mixed_data/logs/tag_domain_keypoints_split_1.log 2>&1 &
+
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag_domain_keypoints.py \
+    --input_file_path  "data/mixed_data/splits/split_2.json" \
+    --cuda_visible_devices "2,3" > data/mixed_data/logs/tag_domain_keypoints_split_2.log 2>&1 &
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag_domain_keypoints.py \
+    --input_file_path  "data/mixed_data/splits/split_3.json" \
+    --cuda_visible_devices "4,5" > data/mixed_data/logs/tag_domain_keypoints_split_3.log 2>&1 &
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag_domain_keypoints.py \
+    --input_file_path  "data/mixed_data/splits/split_4.json" \
+    --cuda_visible_devices "6,7" > data/mixed_data/logs/tag_domain_keypoints_split_4.log 2>&1 &
+
+
+# worker 3
+
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag_domain_keypoints.py \
+    --input_file_path  "data/mixed_data/splits/split_5.json" \
+    --cuda_visible_devices "2,3" > data/mixed_data/logs/tag_domain_keypoints_split_5.log 2>&1 &
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag_domain_keypoints.py \
+    --input_file_path  "data/mixed_data/splits/split_6.json" \
+    --cuda_visible_devices "4,5" > data/mixed_data/logs/tag_domain_keypoints_split_6.log 2>&1 &
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag_domain_keypoints.py \
+    --input_file_path  "data/mixed_data/splits/split_7.json" \
+    --cuda_visible_devices "6,7" > data/mixed_data/logs/tag_domain_keypoints_split_7.log 2>&1 &
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u data_tag_domain_keypoints.py \
+    --input_file_path  "data/mixed_data/splits/split_8.json" \
+    --cuda_visible_devices "0,1" > data/mixed_data/logs/tag_domain_keypoints_split_8.log 2>&1 &
+
+
+####################
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u /share/project/sunshuang/deep_search/data_syn/data_tag_domain_keypoints.py \
+    --input_file_path  "/share/project/sunshuang/deep_search/data_for_rl/musique_ans_v1.0_train_only_qa.json" \
+    --cuda_visible_devices "0,1,2,3,4,5,6,7" > /share/project/sunshuang/deep_search/data_for_rl/tag_musique.log 2>&1 &
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u /share/project/sunshuang/deep_search/data_syn/data_tag_domain_keypoints.py \
+    --input_file_path  "/share/project/sunshuang/deep_search/data_for_rl/MultiHopRAG_only_qa.json" \
+    --cuda_visible_devices "0,1,2,3,4,5,6,7" > /share/project/sunshuang/deep_search/data_for_rl/MultiHopRAG_only_qa.log 2>&1 &
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u /share/project/sunshuang/deep_search/data_syn/data_tag_domain_keypoints.py \
+    --input_file_path  "/share/project/sunshuang/deep_search/data_for_rl/2wiki_train_only_compositional_qa.json" \
+    --cuda_visible_devices "0,1,2,3,4,5,6,7" > /share/project/sunshuang/deep_search/data_for_rl/2wiki_train_only_compositional_qa.log 2>&1 &
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u /share/project/sunshuang/deep_search/data_syn/data_tag_domain_keypoints.py \
+    --input_file_path  "/share/project/sunshuang/deep_search/data_for_rl/merged_data.json" \
+    --cuda_visible_devices "0,1,2,3,4,5,6,7" > /share/project/sunshuang/deep_search/data_for_rl/merged_data.log 2>&1 &
+
+

deep_search/data_syn/data_tag_domain_keypoints.py

@@ -0,0 +1,118 @@
+import os
+import argparse
+import json
+from vllm import LLM, SamplingParams
+from datasets import Dataset
+from transformers import AutoTokenizer
+import random
+from tqdm import tqdm
+import torch
+# os.environ["CUDA_VISIBLE_DEVICES"] = "0,1"
+
+user_prompt = f"""You are an advanced semantic analyzer. For the given question, perform the following tasks step-by-step:
+
+1. **Domain Identification**:
+- Determine the broad subject category (domain) this question belongs to.
+- Examples: film, history, biology, geography, politics, technology, etc (or any other suitable domain)
+
+2. **Key Point Extraction**:
+- Identify 2-4 core semantic components that are crucial for answering
+- Include:
+• Key entities (e.g., films, people, locations)
+• Critical attributes (e.g., age, duration, population)
+• Core relationships (e.g., comparison, causality)
+• Measurement dimensions (e.g., time, quantity)
+- Exclude filler words and non-essential descriptors
+
+**Output Requirements**:
+- Use JSON format: {{"domain": "...", "key_points": [...]}}
+- Keep key_points concise (1-2 words each)
+- Use lowercase for all outputs
+- Separate multiple key_points with commas
+
+**Examples**:
+Question: "Which film whose director is younger, Charge It To Me or Danger: Diabolik?"
+Output: {{"domain": "film", "key_points": ["director", "age"]}}
+
+**Now process this question:**
+{{Question}}"""
+
+def main(input_file_path, output_path):
+    
+    data = []
+    with open(input_file_path, "r", encoding="utf-8") as file:
+        # for line in file:
+        #     data.append(json.loads(line))
+        data = json.load(file)
+
+    
+    prompts = []
+    new_data = []
+
+    for item in tqdm(data, desc="gen prompts"):
+        prompt = user_prompt.replace('{Question}', item["question"])
+        # item = f"请解释下面的内容：{item}"
+        # message = {"role": "user", "content": item}
+        messages = [{"role": "user", "content": prompt}]
+        
+        text = tokenizer.apply_chat_template(
+            messages, add_generation_prompt=True, tokenize=False
+        )
+        prompts.append(text)
+        item["tag_prompt"] = text
+        # new_data.append(prompt)
+    print(prompts[0])
+    outputs = llm.generate(prompts, sampling_params)
+
+    generated_texts = []
+    for i, output in tqdm(enumerate(outputs), desc="process outputs"):
+        # prompt = output.prompt
+        generated_text = output.outputs[0].text.strip()
+        data[i]["tag_qwq"] = generated_text
+
+        generated_texts.append(data[i])
+
+    
+    os.makedirs(os.path.dirname(output_path), exist_ok=True) 
+    with open(output_path, "w", encoding="utf-8") as json_file:
+        json.dump(generated_texts, json_file, ensure_ascii=False, indent=4)
+    # with open(output_path, 'w', encoding='utf-8') as f:
+    #     for item in generated_texts:
+    #         f.write(json.dumps(item, ensure_ascii=False) + '\n')
+
+    print(f"save to {output_path}")
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Process input file and CUDA devices.")
+    parser.add_argument("--input_file_path", type=str, required=True, help="Path to the input JSON file.")
+    parser.add_argument("--cuda_visible_devices", type=str, required=True, help="CUDA devices to use (e.g., '0,1').")
+    args = parser.parse_args()
+
+    os.environ["CUDA_VISIBLE_DEVICES"] = args.cuda_visible_devices
+    print(f"CUDA_VISIBLE_DEVICES: {os.environ.get('CUDA_VISIBLE_DEVICES')}")
+    input_file_path = args.input_file_path
+
+
+    model_name = "qwq"
+    model_path = "/share/project/zhipengchen/model/QwQ-32B"
+    
+    
+    # input_file_path = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/split_1.json"
+    # file_name = input_file_path.split(".")[0]
+    file_name = os.path.basename(input_file_path).split(".")[0]
+    base_path = os.path.dirname(input_file_path)
+    output_dir = os.path.join(base_path, "tagged_domain_keypoints")
+    os.makedirs(output_dir, exist_ok=True)
+    output_path = os.path.join(output_dir, f"{file_name}_tagged.json")
+
+    tokenizer = AutoTokenizer.from_pretrained(model_path)
+    
+    sampling_params = SamplingParams(n=1, temperature=0.6, max_tokens=10000, top_p=0.95, top_k=40)
+    llm = LLM(model=model_path, tensor_parallel_size=torch.cuda.device_count(), gpu_memory_utilization=0.95,trust_remote_code=True)
+    
+    print(f"CUDA_VISIBLE_DEVICES: {os.environ.get('CUDA_VISIBLE_DEVICES')}")
+    print(f"model path: {model_path}")
+    print(f"input_file_path: {input_file_path}")
+    print(f"output_path: {output_path}")
+
+    main(input_file_path, output_path)
+

deep_search/data_syn/extract_domain_keypoints.py

@@ -0,0 +1,87 @@
+import json
+import os
+import torch
+from tqdm import tqdm
+import ast
+
+def save_to_json(data, filename):
+    with open(filename, 'w', encoding='utf-8') as f:
+        json.dump(data, f, ensure_ascii=False, indent=4)
+    print(f"save to {filename}, data len: {len(data)}")
+def load_json(file_path):
+    with open(file_path, "r", encoding="utf-8") as f:
+        data = json.load(f)
+    print(f"load from {file_path}, data len: {len(data)}")
+    return data
+
+# def extract_braces(text):
+#     # 匹配所有 {} 及其内容（非贪婪模式）
+#     matches = re.findall(r'\{.*?\}', text)
+#     return matches
+
+def extract_brace_content(s):
+    matches = re.findall(r'\{[^}]*\}', s)
+    return matches[-1] if matches else ''
+
+def extract_domain_keypoints(input_file_path, output_file_path):
+    data = load_json(input_file_path)
+
+    error_count = 0
+    error_two_count = 0
+    # match_zero = 0
+    # match_one = 0
+    # match_more = 0
+
+    new_data = []
+    for item in tqdm(data):
+        domain_keypoints = item["tag_qwq"].split("\n</think>\n\n")[-1]
+        
+        try: # 能够直接解析的
+            domain_keypoints_formatted = json.loads(domain_keypoints)
+            item["domain_keypoints"] = domain_keypoints_formatted
+        except:
+            error_count += 1
+            # print(f"error: {domain_keypoints_formatted}")
+            try: # 二次解析的
+                domain_keypoints = domain_keypoints.split("\n\n")[0]
+                # print("-----------")
+                # print(domain_keypoints)
+                item["domain_keypoints"] = json.loads(domain_keypoints)
+            except:
+                error_two_count += 1
+                print(f"error_two_count: {[domain_keypoints]}")
+                item["domain_keypoints"] = domain_keypoints
+            # try:
+            #     matches = extract_brace_content(domain_keypoints)
+            #     print(matches)
+            #     if len(matches) == 0:
+            #         match_zero += 1
+            #         item["domain_keypoints"] = domain_keypoints_formatted
+            #     elif len(matches) == 1:
+            #         match_one += 1
+            #         item["domain_keypoints"] = json.loads(matches[-1])
+            #     else:
+            #         match_more += 1
+            #         # print(f"more: {matches}")
+            #         item["domain_keypoints"] = json.loads(matches[-1])
+            # except:
+            #     print(f"match_error: {domain_keypoints_formatted}")
+            #     # print(f"match_error": )
+            #     item["domain_keypoints"] = domain_keypoints_formatted
+            # # item["domain_keypoints"] = domain_keypoints
+        if isinstance(item["domain_keypoints"], dict):
+            new_data.append(item)
+    print(f"error_count: {error_count}")
+    print(f"error_two_count: {error_two_count}")
+    # print(f"match_zero: {match_zero}")
+    # print(f"match_one: {match_one}")
+    # print(f"match_more: {match_more}")
+    print(f"len data: {len(data)}")
+    print(f"len new_data: {len(new_data)}")
+    save_to_json(new_data, output_file_path)
+
+
+if __name__ == "__main__":
+    input_file_path = "/share/project/sunshuang/deep_search/data_for_rl/tagged_domain_keypoints/merged_data_tagged.json"
+    output_file_path = "/share/project/sunshuang/deep_search/data_for_rl/tagged_domain_keypoints/merged_data_tagged_domain_keypoints.json"
+    extract_domain_keypoints(input_file_path, output_file_path)

deep_search/data_syn/graph_selection.py

@@ -0,0 +1,91 @@
+import os
+import json
+
+def save_to_json(data, filename):
+    with open(filename, 'w', encoding='utf-8') as f:
+        json.dump(data, f, ensure_ascii=False, indent=4)
+    print(f"save to {filename}, data len: {len(data)}")
+def load_json(file_path):
+    with open(file_path, "r", encoding="utf-8") as f:
+        data = json.load(f)
+    print(f"load from {file_path}, data len: {len(data)}")
+    return data
+
+
+# 输入和输出路径
+input_folder = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits"
+# output_folder = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/code/graph"
+
+# # 创建输出文件夹
+# os.makedirs(output_folder, exist_ok=True)
+
+# 初始化分类字典
+classified_data = {
+    "1-hop": [],
+    "2-hop": [],
+    "3-hop": [],
+    "4-hop": [],
+    "5-hop": [],
+    "other": []
+}
+
+new_data = []
+# 遍历目标文件夹中的所有文件
+for filename in sorted(os.listdir(input_folder)):
+    
+    if filename == "tagged_domain_keypoints":
+        continue
+    if "tagged" in filename and filename.endswith(".json"):
+        file_path = os.path.join(input_folder, filename)
+        print(filename)
+        # 读取 JSON 文件
+        with open(file_path, "r", encoding="utf-8") as f:
+            data = json.load(f)
+
+        # 遍历数据中的每个条目
+        for item in data:
+            # 提取 Reasoning Graph 部分
+            tag_qwq = item.get("tag_qwq", "")
+            reasoning_graph = tag_qwq.split("</think>")[-1].split("Graph:")[-1].split("Area:")[0].strip()
+
+            # 判断 hop 数
+            hop = None
+            if "1-hop" in reasoning_graph:
+                hop = "1-hop"
+                item["hop"] = 1
+            elif "2-hop" in reasoning_graph:
+                hop = "2-hop"
+                item["hop"] = 2
+            elif "3-hop" in reasoning_graph:
+                hop = "3-hop"
+                item["hop"] = 3
+            elif "4-hop" in reasoning_graph:
+                hop = "4-hop"
+                item["hop"] = 4
+            elif "5-hop" in reasoning_graph:
+                hop = "5-hop"
+                item["hop"] = 5
+            else:
+                hop = "other"
+                item["hop"] = -1
+            new_data.append(item)
+            # 将条目添加到对应的分类
+            # classified_data[hop].append(item)
+
+# 保存分类后的数据到不同文件
+# for hop, items in classified_data.items():
+#     output_file = os.path.join(output_folder, f"{hop}.json")
+#     with open(output_file, "w", encoding="utf-8") as f:
+#         json.dump(items, f, ensure_ascii=False, indent=4)
+#     print(f"Saved {len(items)} items to {output_file}")
+
+data_1 = load_json("/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/tagged_domain_keypoints/merged_tagged_domain_keypoints_keywords_count.json")
+
+assert len(new_data) == len(data_1)
+for item, item_1 in zip(new_data, data_1):
+    assert item["idx"] == item_1["idx"], f"{item['idx']} {item_1['idx']}"
+    assert item["Question"] == item_1["Question"]
+
+    item_1["hop"] = item["hop"]
+
+save_to_json(data_1, "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/tagged_domain_keypoints/merged_tagged_domain_keypoints_keywords_count_hop.json")

deep_search/data_syn/json_to_text.py

@@ -0,0 +1,48 @@
+import json
+import os
+import random
+from tqdm import tqdm
+
+# 设置随机种子以保证结果可复现
+random.seed(0)
+
+def save_to_json(data, filename):
+    """保存数据到 JSON 文件"""
+    with open(filename, 'w', encoding='utf-8') as f:
+        json.dump(data, f, ensure_ascii=False, indent=4)
+    print(f"Saved to {filename}, data length: {len(data)}")
+
+def load_json(file_path):
+    """从 JSON 文件加载数据"""
+    with open(file_path, "r", encoding="utf-8") as f:
+        data = json.load(f)
+    print(f"Loaded from {file_path}, data length: {len(data)}")
+    return data
+
+def save_questions_to_txt(data, output_txt_file):
+    """将数据中的 Question 字段保存到 TXT 文件"""
+    with open(output_txt_file, 'w', encoding='utf-8') as f:
+        for item in data:
+            question = item.get("Question", "")  # 获取 Question 字段，默认为空字符串
+            if question:  # 如果 Question 存在，则写入文件
+                f.write(f"idx: {item['idx']}\n")
+                f.write(f"question: {item['Question']}\n\n")  # 每个问题占一行
+    print(f"Questions saved to {output_txt_file}")
+
+# 输入和输出文件路径
+input_file = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/tagged_domain_keypoints/final_selected_dataset.json"  # 替换为你的输入文件路径
+output_file = "shuffled_data.json"  # 替换为你的输出文件路径
+output_txt_file = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/tagged_domain_keypoints/final_selected_dataset.txt"  # 输出的 TXT 文件路径
+
+# 加载数据
+data = load_json(input_file)
+
+# 打乱数据
+random.shuffle(data)
+print("Data shuffled successfully.")
+
+# # 保存打乱后的数据
+# save_to_json(data, output_file)
+
+# 将数据中的 Question 字段保存到 TXT 文件
+save_questions_to_txt(data, output_txt_file)

deep_search/data_syn/keywords_count.py

@@ -0,0 +1,60 @@
+import json
+import os
+import torch
+from tqdm import tqdm
+import ast
+import re
+
+keywords = ["who", "whom", "whose","where", "what", "when", "how", "why", "which", "whether", "is", "are", "do", "does", "was", "were"]
+
+def save_to_json(data, filename):
+    with open(filename, 'w', encoding='utf-8') as f:
+        json.dump(data, f, ensure_ascii=False, indent=4)
+    print(f"save to {filename}, data len: {len(data)}")
+def load_json(file_path):
+    with open(file_path, "r", encoding="utf-8") as f:
+        data = json.load(f)
+    print(f"load from {file_path}, data len: {len(data)}")
+    return data
+
+
+def count_keywords(text, keywords):
+    """
+    统计字符串中指定关键词的出现次数。
+    
+    参数:
+        text (str): 输入的字符串。
+        keywords (list): 需要统计的关键词列表。
+    
+    返回:
+        dict: 每个关键词及其对应的出现次数。
+    """
+    # 将文本转换为小写以忽略大小写
+    text = text.lower()
+    
+    # 创建一个字典存储每个关键词的计数结果
+    keyword_counts = {}
+    total = 0
+    special_total = 0
+    for keyword in keywords:
+        # 使用正则表达式匹配独立的单词
+        pattern = r'\b' + re.escape(keyword.lower()) + r'\b'
+        matches = re.findall(pattern, text)
+        keyword_counts[keyword] = len(matches)
+        if keyword not in ["is", "are", "do", "does", "was", "were"]:
+            special_total += len(matches)
+        total += len(matches)
+    keyword_counts["total"] = total
+    keyword_counts["special_total"] = special_total
+    return keyword_counts
+
+
+
+if __name__ == "__main__":
+    input_file_path = "/share/project/sunshuang/deep_search/data_for_rl/tagged_domain_keypoints/merged_data_tagged_domain_keypoints.json"
+    output_file_path = "/share/project/sunshuang/deep_search/data_for_rl/tagged_domain_keypoints/merged_data_tagged_domain_keypoints_keywords_count.json"
+    data = load_json(input_file_path)
+    for i in tqdm(range(len(data))):
+        text = data[i]["question"]
+        data[i]["keywords_count"] = count_keywords(text, keywords)
+    save_to_json(data, output_file_path)

deep_search/data_syn/merge.py

@@ -0,0 +1,28 @@
+import os
+import json
+
+# 指定路径
+base_path = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/tagged_domain_keypoints"
+
+# 用于存储合并后的数据
+merged_data = []
+
+# 遍历目录下的所有文件
+for file in sorted(os.listdir(base_path)):
+    print(f"Processing file: {file}")
+    if file.endswith("tagged.json"): 
+        if file.endswith("tagged.json"):  # 筛选以 "tagged.json" 结尾的文件
+            file_path = os.path.join(base_path, file)  # 获取完整路径
+            print(f"Processing file: {file_path}")
+            
+            # 读取 JSON 文件内容并合并
+            with open(file_path, "r", encoding="utf-8") as f:
+                data = json.load(f)
+                merged_data.extend(data)  # 将数据添加到合并列表中
+
+# 输出合并后的数据到新文件（可选）
+output_file = os.path.join(base_path, "merged_tagged.json")
+with open(output_file, "w", encoding="utf-8") as out_f:
+    json.dump(merged_data, out_f, ensure_ascii=False, indent=4)
+
+print(f"合并完成，共处理 {len(merged_data)} 条数据，结果已保存至 {output_file}")

deep_search/data_syn/merge_two_data.py

@@ -0,0 +1,33 @@
+import json
+from collections import defaultdict,OrderedDict, Counter
+from tqdm import tqdm
+import random
+import matplotlib.pyplot as plt
+
+
+
+def save_to_json(data, filename):
+    """保存数据到 JSON 文件"""
+    with open(filename, 'w', encoding='utf-8') as f:
+        json.dump(data, f, ensure_ascii=False, indent=4)
+    print(f"Saved to {filename}, data length: {len(data)}")
+
+def load_json(file_path):
+    """从 JSON 文件加载数据"""
+    with open(file_path, "r", encoding="utf-8") as f:
+        data = json.load(f)
+    print(f"Loaded from {file_path}, data length: {len(data)}")
+    return data
+
+
+file1 = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/final_dataset_new/final_selected_dataset.json"
+file2 = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data/final_selected_dataset.json"
+
+data1 = load_json(file1)
+data2 = load_json(file2)
+
+merged_data = data1 + data2
+
+output_file = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/merged_4k/4k.json"
+
+save_to_json(merged_data, output_file)

deep_search/data_syn/search_o1_data_syn.py

@@ -0,0 +1,937 @@
+# run_search_o1.py
+import os
+import json
+import time
+import re
+from tqdm import tqdm
+import numpy as np
+import torch
+import string
+from typing import Optional, Tuple, List, Dict
+import argparse
+
+from transformers import AutoTokenizer
+from vllm import LLM, SamplingParams
+
+from bing_search import (
+    bing_web_search, 
+    extract_relevant_info, 
+    fetch_page_content, 
+    extract_snippet_with_context
+)
+from evaluate import (
+    run_evaluation, 
+    extract_answer
+)
+from prompts import (
+    get_gpqa_search_o1_instruction, 
+    get_math_search_o1_instruction, 
+    get_code_search_o1_instruction, 
+    get_singleqa_search_o1_instruction, 
+    get_multiqa_search_o1_instruction, 
+    get_webpage_to_reasonchain_instruction,
+    get_task_instruction_openqa, 
+    get_task_instruction_math, 
+    get_task_instruction_multi_choice, 
+    get_task_instruction_code, 
+    get_singleqa_search_o1_instruction_1,
+    get_multiqa_search_o1_instruction_1,
+    get_webpage_to_reasonchain_instruction_1,
+    get_math_search_o1_instruction_1,
+)
+
+
+from openai import OpenAI
+
+from add_eval import add_eval
+# Modify OpenAI's API key and API base to use vLLM's API server.
+# 使用 vLLM 的 API 服务器需要修改 OpenAI 的 API 密钥和 API 库。
+
+# openai_api_key = "EMPTY"
+# openai_api_base = "http://localhost:8000/v1"
+# client = OpenAI(
+#     api_key=openai_api_key,
+#     base_url=openai_api_base,
+# )
+
+# Define special tokens
+BEGIN_SEARCH_QUERY = "<|begin_search_query|>"
+END_SEARCH_QUERY = "<|end_search_query|>"
+BEGIN_SEARCH_RESULT = "<|begin_search_result|>"
+END_SEARCH_RESULT = "<|end_search_result|>"
+
+
+# os.environ['http_proxy'] = 'http://127.0.0.1:7880'
+# os.environ['https_proxy'] = 'http://127.0.0.1:7880'
+
+# 增加了cache共享和has answer评测,truncate doc
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Run Search O1 for various datasets and models.")
+
+    # Dataset and split configuration
+    parser.add_argument(
+        '--dataset_name',
+        type=str,
+        required=True,
+        choices=['simpleqa', 'gpqa', 'math500', 'aime', 'amc', 'livecode', 'nq', 'triviaqa', 'hotpotqa', '2wiki', 'musique', 'bamboogle'],
+        help="Name of the dataset to use."
+    )
+
+    parser.add_argument(
+        '--split',
+        type=str,
+        required=True,
+        choices=['test', 'diamond', 'main', 'extended'],
+        help="Dataset split to use."
+    )
+
+    parser.add_argument(
+        '--subset_num',
+        type=int,
+        default=-1,
+        help="Number of examples to process. Defaults to all if not specified."
+    )
+
+    # Search and document retrieval configuration
+    parser.add_argument(
+        '--max_search_limit',
+        type=int,
+        default=10,
+        help="Maximum number of searches per question."
+    )
+
+    parser.add_argument(
+        '--max_turn',
+        type=int,
+        default=15,
+        help="Maximum number of turns."
+    )
+
+    parser.add_argument( # 使用搜索引擎时，返回的最大文档数
+        '--top_k',
+        type=int,
+        default=10,
+        help="Maximum number of search documents to return."
+    )
+
+    parser.add_argument(
+        '--max_doc_len',
+        type=int,
+        default=3000,
+        help="Maximum length of each searched document."
+    )
+
+    # parser.add_argument(
+    #     '--use_jina',
+    #     type=bool,
+    #     default=False,
+    #     help="Whether to use Jina API for document fetching."
+    # )
+    parser.add_argument(
+        '--use_jina',
+        action='store_true',
+        help="Whether to use Jina API for document fetching."
+    )
+
+    parser.add_argument(
+        '--jina_api_key',
+        type=str,
+        default='None',
+        help="Your Jina API Key to Fetch URL Content."
+    )
+
+    # Model configuration
+    parser.add_argument(
+        '--model_path',
+        type=str,
+        required=True,
+        help="Path to the pre-trained model."
+    )
+
+    # Sampling parameters
+    parser.add_argument(
+        '--temperature',
+        type=float,
+        default=0.7,
+        help="Sampling temperature."
+    )
+
+    parser.add_argument(
+        '--top_p',
+        type=float,
+        default=0.8,
+        help="Top-p sampling parameter."
+    )
+
+    parser.add_argument(
+        '--top_k_sampling',
+        type=int,
+        default=20,
+        help="Top-k sampling parameter."
+    )
+
+    parser.add_argument(
+        '--repetition_penalty',
+        type=float,
+        default=None,
+        help="Repetition penalty. If not set, defaults based on the model."
+    )
+
+    parser.add_argument(
+        '--max_tokens',
+        type=int,
+        default=32768,
+        help="Maximum number of tokens to generate. If not set, defaults based on the model and dataset."
+    )
+
+    # Bing API Configuration
+    parser.add_argument(
+        '--bing_subscription_key',
+        type=str,
+        required=True,
+        help="Bing Search API subscription key."
+    )
+
+    parser.add_argument(
+        '--bing_endpoint',
+        type=str,
+        default="https://api.bing.microsoft.com/v7.0/search",
+        help="Bing Search API endpoint."
+    )
+
+    parser.add_argument(
+        '--cache_dir_base',
+        type=str,
+        required=True,
+        help="cache path."
+    )
+
+    parser.add_argument(
+        '--output_dir_base',
+        type=str,
+        required=True,
+        help="output_dir"
+    )
+
+    # parser.add_argument(
+    #     '--model_doc_reason_path',
+    #     type=str,
+    #     required=True,
+    #     help="Path to the document reasoning model."
+    # )
+
+    # openai_api_base
+    # parser.add_argument(
+    #     '--openai_api_base',
+    #     type=str,
+    #     required=True,
+    #     help="openai_api_base"
+    # )
+    # parser.add_argument(
+    #     '--data_path',
+    #     type=str,
+    #     required=True,
+    #     help="Path to the document reasoning model."
+    # )
+    return parser.parse_args()
+
+def main():
+    args = parse_args()
+    print(f"args.use_jina: {args.use_jina}")
+    # Extract arguments
+    dataset_name = args.dataset_name
+    split = args.split
+    subset_num = args.subset_num
+    MAX_SEARCH_LIMIT = args.max_search_limit
+    MAX_TURN = args.max_turn
+    top_k = args.top_k
+    max_doc_len = args.max_doc_len
+    model_path = args.model_path
+    # model_doc_reason_path = args.model_doc_reason_path
+    temperature = args.temperature
+    top_p = args.top_p
+    top_k_sampling = args.top_k_sampling
+    repetition_penalty = args.repetition_penalty
+    max_tokens = args.max_tokens
+    bing_subscription_key = args.bing_subscription_key
+    bing_endpoint = args.bing_endpoint
+    use_jina = args.use_jina
+    jina_api_key = args.jina_api_key
+    cache_dir_base = args.cache_dir_base
+    output_dir_base = args.output_dir_base
+    # openai_api_base = args.openai_api_base
+    use_jina = False
+    print(f"use_jina: {use_jina}")
+
+    print(f"CUDA_VISIBLE_DEVICES is set to: {os.environ['CUDA_VISIBLE_DEVICES']}")
+
+    # openai_api_key = "EMPTY"
+    # openai_api_base = openai_api_base
+    # client = OpenAI(
+    #     api_key=openai_api_key,
+    #     base_url=openai_api_base,
+    # )
+
+    # Adjust parameters based on dataset
+    if dataset_name in ['nq', 'triviaqa', 'hotpotqa', 'musique', 'bamboogle', '2wiki', 'medmcqa', 'pubhealth']:
+        MAX_SEARCH_LIMIT = 5
+        if dataset_name in ['simpleqa' ,'hotpotqa', 'musique', 'bamboogle', '2wiki']:
+            MAX_SEARCH_LIMIT = 10
+            MAX_TURN = 15
+        top_k = 10
+        max_doc_len = 3000
+    
+    if args.jina_api_key == 'None':
+        jina_api_key = None
+
+    # Set default repetition_penalty if not provided
+    if repetition_penalty is None:
+        repetition_penalty = 1.05 if 'qwq' in model_path.lower() else 1.0
+
+    # Data paths based on dataset
+    if split == "test": # 测试用的数据集地址
+        data_path = f"./data/test/{dataset_name}.json"
+    else: # 训练用的数据集地址
+        if dataset_name == 'livecode':
+            data_path = f'./data/LiveCodeBench/{split}.json'
+        elif dataset_name in ['math500', 'gpqa', 'aime', 'amc']:
+            data_path = f'./data/{dataset_name.upper()}/{split}.json'
+        else:
+            data_path = f'./data/QA_Datasets/{dataset_name}.json'
+
+    print('-----------------------')
+    print(f'Using {dataset_name} {split} set.')
+    print('-----------------------')
+
+    # ---------------------- Caching Mechanism ----------------------
+    # Define cache directories and file paths
+    # cache_dir = './cache'
+    model_name = model_path.split('/')[-1].replace('-instruct', '')
+    # cache_dir = f'./{cache_dir_base}_{dataset_name}_{model_name}'
+    cache_dir = cache_dir_base
+    search_cache_path = os.path.join(cache_dir, 'search_cache.json')
+    url_cache_path = os.path.join(cache_dir, 'url_cache.json')
+
+    # Ensure cache directory exists
+    os.makedirs(cache_dir, exist_ok=True)
+
+    # Load existing caches or initialize empty dictionaries
+    if os.path.exists(search_cache_path):
+        with open(search_cache_path, 'r', encoding='utf-8') as f:
+            search_cache = json.load(f)
+    else:
+        search_cache = {}
+
+    if os.path.exists(url_cache_path):
+        with open(url_cache_path, 'r', encoding='utf-8') as f:
+            url_cache = json.load(f)
+    else:
+        url_cache = {}
+
+    # Function to save caches
+    def save_caches():
+        with open(search_cache_path, 'w', encoding='utf-8') as f:
+            json.dump(search_cache, f, ensure_ascii=False, indent=2)
+        with open(url_cache_path, 'w', encoding='utf-8') as f:
+            json.dump(url_cache, f, ensure_ascii=False, indent=2)
+
+    # ---------------------- Model Loading ----------------------
+    print(f"Loading tokenizer from {model_path}...")
+    tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
+    if tokenizer.pad_token is None:
+        tokenizer.pad_token = tokenizer.eos_token
+    tokenizer.padding_side = 'left' # 主要是左填充
+    print("Tokenizer loaded successfully.")
+
+    # Define output directory based on model and dataset
+  
+    
+    # if 'qwq' in model_path.lower():
+    #     if dataset_name in ['math500', 'gpqa', 'aime', 'amc', 'livecode']:
+    #         output_dir = f'./{output_dir_base}/{dataset_name}.qwq.search_o1'
+    #         if dataset_name == 'gpqa' and (MAX_SEARCH_LIMIT != 5 or top_k != 10):
+    #             output_dir = f'./{output_dir_base}/runs.analysis/{dataset_name}.qwq.search_o1.{MAX_SEARCH_LIMIT}.{top_k}'
+    #     else:
+    #         output_dir = f'./{output_dir_base}/runs.qa/{dataset_name}.qwq.search_o1'
+    # else:
+    #     model_short_name = model_path.split('/')[-1].lower().replace('-instruct', '')
+    #     output_dir = f'./{output_dir_base}/runs.baselines/{dataset_name}.{model_short_name}.search_o1'
+    output_dir = output_dir_base
+    os.makedirs(output_dir, exist_ok=True)
+
+    print(f"Loading model from {model_path}...")
+    print(f"device_count: {torch.cuda.device_count()}")
+    
+    # Initialize the LLM
+    llm = LLM(
+        model=model_path,
+        tensor_parallel_size=torch.cuda.device_count(),
+        gpu_memory_utilization=0.95,
+
+    )
+    print("Model loaded successfully.")
+
+    # # ----------------------Loading model to reason in document ----------------------
+
+    # print(f"Loading tokenizer_doc_reason from {model_doc_reason_path}...")
+    # tokenizer_doc_reason = AutoTokenizer.from_pretrained(model_doc_reason_path, trust_remote_code=True)
+    # if tokenizer_doc_reason.pad_token is None:
+    #     tokenizer_doc_reason.pad_token = tokenizer_doc_reason.eos_token
+    # tokenizer_doc_reason.padding_side = 'left' # 主要是左填充
+    # print("tokenizer_doc_reason loaded successfully.")
+
+    # print(f"Loading tokenizer_doc_reason from {model_doc_reason_path}...")
+    
+    # # Initialize the LLM
+    # # torch.cuda.set_device(6,7)
+    
+    # llm_doc_reason = LLM(
+    #     model=model_doc_reason_path,
+    #     tensor_parallel_size=2,
+    #     gpu_memory_utilization=0.95,
+
+    # )
+    # print("Model_doc_reason loaded successfully.")
+
+    # ---------------------- Data Loading ----------------------
+    print(f"Loading data from {data_path}...")
+    with open(data_path, 'r', encoding='utf-8') as json_file:
+        filtered_data = json.load(json_file)
+    print(f"Data loaded successfully. Total examples: {len(filtered_data)}")
+
+    # ---------------------- Batch Generation Function ----------------------
+    def generate_webpage_to_reasonchain_batch( # 模型基于网页内容生成推理，然后从模型的回复中提取答案
+            original_questions: List[str],
+            prev_reasonings: List[str],
+            search_queries: List[str],
+            documents: List[str],
+            dataset_name: str,
+            batch_output_records: List[Dict],  # New parameter to collect outputs
+            max_tokens: int = 32768,
+            coherent: bool = False,
+        ) -> List[str]:
+
+            if "Qwen2.5" in model_path:
+                max_tokens = 8192
+                
+                encode_docs = tokenizer(documents, truncation=True, max_length=20000, add_special_tokens=False)["input_ids"]
+                documents = tokenizer.batch_decode(encode_docs)
+
+                # 统计每个文档的长度
+                doc_lengths = [len(doc) for doc in encode_docs]
+
+                # # 打印每个文档的长度
+                # for i, length in enumerate(doc_lengths):
+                #     print(f"Document {i + 1}: {length} tokens")
+
+                # 如果需要返回长度列表，可以直接使用 doc_lengths
+                print(f"for {model_path}, set max_tokens={max_tokens} for doc gen, truncate documnets. ")
+                print("All document lengths:", doc_lengths)
+
+            assert len(documents) / len(search_queries) == top_k, "Number of documents should be equal to top_k * number of search queries."
+
+            for i, doc in enumerate(documents):
+                query_idx = i // top_k  # 每 query 有 top_k 条文档
+                question = original_questions[query_idx]
+                reasoning = prev_reasonings[query_idx]
+                query = search_queries[query_idx]
+
+                # 拼装用户提示
+                up = get_webpage_to_reasonchain_instruction(
+                    reasoning,     # 之前的推理
+                    query,         # 该文档对应的搜索query
+                    doc            # 当前文档内容
+                )
+                user_prompts.append(up)
+
+
+            user_prompts = [ # 根据之前的推理，搜索query和搜索到的doc生成用户提示
+                get_webpage_to_reasonchain_instruction(r, sq, doc)
+                for r, sq, doc in zip(prev_reasonings, search_queries, documents)
+            ]
+
+            prompts = [{"role": "user", "content": up} for up in user_prompts]
+            prompts = [tokenizer.apply_chat_template([p], tokenize=False, add_generation_prompt=True) for p in prompts]
+
+            output = llm.generate( # 生成模型回复
+                prompts,
+                sampling_params=SamplingParams(
+                    max_tokens=max_tokens,
+                    temperature=0.7,
+                    top_p=0.8,
+                    top_k=20,
+                    repetition_penalty=1.05,
+                )
+            )
+
+            raw_outputs = [out.outputs[0].text for out in output]
+            extracted_infos = [extract_answer(raw, mode='infogen') for raw in raw_outputs] # 提取模型基于网页生成的推理
+
+            for i, (p, r, e) in enumerate(zip(prompts, raw_outputs, extracted_infos)):
+                batch_output_records.append({
+                    'prompt': p,
+                    'raw_output': r,
+                    'extracted_info': e
+                })
+
+            return extracted_infos
+
+    # ---------------------- Preparation of Input Prompts ----------------------
+    input_list = []
+    for item in filtered_data: # 生成prompts
+        question = item['Question']
+
+        if dataset_name in ['nq', 'triviaqa', 'hotpotqa', 'musique', 'bamboogle', '2wiki']:
+            if dataset_name in ['nq', 'triviaqa']:
+                instruction = get_singleqa_search_o1_instruction_1(MAX_SEARCH_LIMIT)
+            elif dataset_name in ['simpleqa', 'hotpotqa', 'musique', 'bamboogle', '2wiki']:
+                instruction = get_multiqa_search_o1_instruction_1(MAX_SEARCH_LIMIT)
+            if 'qwq' in model_path.lower():
+                user_prompt = get_task_instruction_openqa(question, model_name='qwq')
+            else:
+                user_prompt = get_task_instruction_openqa(question)
+
+        elif dataset_name in ['math500', 'aime', 'amc']:
+            instruction = get_math_search_o1_instruction_1(MAX_SEARCH_LIMIT)
+            if 'qwq' in model_path.lower():
+                user_prompt = get_task_instruction_math(question, model_name='qwq')
+            else:
+                user_prompt = get_task_instruction_math(question)
+
+        elif dataset_name == 'gpqa':
+            instruction = get_gpqa_search_o1_instruction(MAX_SEARCH_LIMIT)
+            if 'qwq' in model_path.lower():
+                user_prompt = get_task_instruction_multi_choice(question, model_name='qwq')
+            elif 'llama' in model_path.lower():
+                user_prompt = get_task_instruction_multi_choice(question, model_name='llama')
+            else:
+                user_prompt = get_task_instruction_multi_choice(question)
+
+        elif dataset_name == 'livecode':
+            instruction = get_code_search_o1_instruction(MAX_SEARCH_LIMIT)
+            question_title = item.get('question_title', '')
+            if 'qwq' in model_path.lower():
+                user_prompt = get_task_instruction_code(question, question_title=question_title, model_name='qwq')
+            else:
+                user_prompt = get_task_instruction_code(question)
+        else:
+            user_prompt = ""  # Default to empty if dataset not matched
+
+        prompt = [{"role": "user", "content": instruction + user_prompt}] # instruction是告诉模型怎么进行搜索，user_prompt是用户具体问题
+        prompt = tokenizer.apply_chat_template(prompt, tokenize=False, add_generation_prompt=True)
+        input_list.append(prompt)
+
+    if subset_num != -1:
+        input_list = input_list[:subset_num]
+        filtered_data = filtered_data[:subset_num]
+
+    # Initialize active sequences
+    active_sequences = [{ # 记录每个问题的搜索历史
+        'item': item,
+        'prompt': prompt,
+        'output': '',
+        'finished': False, # 一开始均为未完成
+        'history': [],
+        'search_count': 0,
+        'executed_search_queries': set(),
+        'all_info': [],
+    } for item, prompt in zip(filtered_data, input_list)]
+
+    # ---------------------- Set Max Tokens ----------------------
+    # if 'qwq' in model_path.lower():
+    #     if dataset_name in ['aime', 'amc', 'livecode']:
+    #         max_tokens = 32768
+    #     else:
+    #         max_tokens = 20480
+    # else:
+    #     max_tokens = 8192
+    # max_tokens = 16384
+    if dataset_name in ['aime', 'amc', 'livecode']:
+        max_tokens = 32768
+    else:
+        max_tokens = 20480
+    # ---------------------- Generation Function ----------------------
+    def run_generation(sequences: List[Dict], max_tokens: int) -> List:
+        prompts = [s['prompt'] for s in sequences] # 提取prompt
+        sampling_params = SamplingParams(
+            max_tokens=max_tokens,
+            temperature=temperature,
+            top_p=top_p,
+            top_k=top_k_sampling,
+            repetition_penalty=repetition_penalty,
+            stop=[END_SEARCH_QUERY, tokenizer.eos_token],
+            include_stop_str_in_output=True,
+        )
+        output_list = llm.generate(prompts, sampling_params=sampling_params) # 模型根据prompt生成回答
+        print(f"run_generation completed {len(output_list)}")
+        return output_list
+
+    # Function to extract text between two tags 提取位于 start_tag 和 end_tag 之间的内容
+    def extract_between(text: str, start_tag: str, end_tag: str) -> Optional[str]:
+        pattern = re.escape(start_tag) + r"(.*?)" + re.escape(end_tag)
+        matches = re.findall(pattern, text, flags=re.DOTALL)
+        if matches:
+            return matches[-1].strip()
+        return None
+
+    def replace_recent_steps(origin_str, replace_str): # 使用replace_str更新origin_str
+        """
+        Replaces specific steps in the original reasoning steps with new steps.
+        If a replacement step contains "DELETE THIS STEP", that step is removed.
+
+        Parameters:
+        - origin_str (str): The original reasoning steps.
+        - replace_str (str): The steps to replace or delete.
+
+        Returns:
+        - str: The updated reasoning steps after applying replacements.
+        这个函数的主要功能是替换给定的推理步骤（reasoning steps），
+        并根据传入的 replace_str 进行更新。
+        如果新的步骤包含 "DELETE THIS STEP"，则删除该步骤
+        """
+
+        def parse_steps(text):
+            """
+            Parses the reasoning steps from a given text.
+
+            Parameters:
+            - text (str): The text containing reasoning steps.
+
+            Returns:
+            - dict: A dictionary mapping step numbers to their content.
+            """
+            step_pattern = re.compile(r"Step\s+(\d+):\s*") # 这个模式会匹配 "Step" 后面跟一个或多个空格，然后是一个数字（步骤编号），最后是冒号
+            steps = {}
+            current_step_num = None
+            current_content = []
+
+            for line in text.splitlines(): # 将输入的文本按行分割，并逐行遍历。每一行会被检查是否包含一个步骤
+                step_match = step_pattern.match(line)
+                if step_match: # 匹配到一个新的步骤
+                    # If there's an ongoing step, save its content，如果当前的步骤不为空，将其为上一个步骤，先将上一个步骤的内容（存在current_content中）保存，然后再更新current_step_num和current_content
+                    if current_step_num is not None:
+                        steps[current_step_num] = "\n".join(current_content).strip()
+                    current_step_num = int(step_match.group(1))
+                    content = line[step_match.end():].strip()
+                    current_content = [content] if content else []
+                else:
+                    if current_step_num is not None:
+                        current_content.append(line)
+            
+            # Save the last step if any
+            if current_step_num is not None: # 保存最后一个步骤
+                steps[current_step_num] = "\n".join(current_content).strip()
+            
+            return steps
+
+        # Parse the original and replacement steps
+        origin_steps = parse_steps(origin_str) # 解析原始的推理步骤
+        replace_steps = parse_steps(replace_str) # 解析要替换的推理步骤
+
+        # Apply replacements
+        for step_num, content in replace_steps.items(): # 遍历要替换的步骤
+            if "DELETE THIS STEP" in content: 
+                # Remove the step if it exists
+                if step_num in origin_steps: # 如果要删除的步骤在原始的推理步骤中存在，则删除该步骤
+                    del origin_steps[step_num]
+            else: # 如果要替换的步骤不是要删除的步骤，则替换该步骤
+                # Replace or add the step 
+                origin_steps[step_num] = content
+
+        # Sort the steps by step number
+        sorted_steps = sorted(origin_steps.items()) # 按照步骤编号对步骤进行排序
+
+        # Reconstruct the reasoning steps as a single string
+        new_reasoning_steps = "\n\n".join([f"{content}" for num, content in sorted_steps]) # 根据排序后的步骤构建新的推理步骤字符串，步骤之间以两个换行符分隔
+
+        return new_reasoning_steps
+
+    # ---------------------- Initialize Collection Structure ----------------------
+    # Initialize a list to collect batch outputs
+    batch_output_records = []
+
+    start_time = time.time()
+    turn = 0
+
+    # 流程
+    # 首先根据prompt让模型生成回复
+    # 从模型的回复中提取搜索查询
+    # 如果有（回复要以END_SEARCH_QUERY结尾）
+    #     根据搜索查询，从互联网上提取相关信息
+    #     处理查询的信息
+    #     让模型基于之前的步骤，检索query和查询到的信息生成新的推理，得到search result
+    #     然后回到第一步（这里模型就会根据前面的search result，再次生成新的回复
+    # 如果没有查询则该条问题结束
+
+    # Main loop until all sequences are finished or maximum turns reached
+    while True:
+        # Identify sequences that need generation
+        sequences_needing_generation = [seq for seq in active_sequences if not seq['finished']] # 筛选出需要生成的新内容的序列，active_sequences 是一个包含所有活跃序列的列表
+
+        if sequences_needing_generation:
+            turn += 1
+            print(f'\n-------------- Turn {turn} --------------')
+            print(f"We have {len(sequences_needing_generation)} sequences needing generation...")
+            outputs = run_generation(sequences_needing_generation, max_tokens) # 根据prompt
+            print("Generation completed, processing outputs...")
+
+            # Initialize batch variables
+            batch_relevant_info = []
+            batch_original_questions = []
+            batch_prev_reasonings = []
+            batch_search_queries = []
+            batch_documents = []
+            batch_sequences = []
+
+            # Collect URLs to fetch across all sequences
+            all_urls_to_fetch = set() # 初始化一个集合 all_urls_to_fetch 用来收集所有需要获取的 URL
+            url_snippets = {}
+            url_sequence_map = {}  # Map URL to list of sequences needing it
+
+            # Process each sequence and collect URLs
+            for seq, out in zip(sequences_needing_generation, outputs): # 遍历需要生成新内容的序列，并生成新内容，同时收集需要获取的 URL
+                text = out.outputs[0].text # 将生成的文本添加到序列的历史记录、提示和输出中
+                seq['history'].append(text)
+                # Append generated text to prompt and output
+                seq['prompt'] += text
+                seq['output'] += text
+                seq['all_info'].append({f"turn_{turn}_reason": text})
+                # Extract search query
+                search_query = extract_between(text, BEGIN_SEARCH_QUERY, END_SEARCH_QUERY) # 提取搜索查询
+
+                # If a search query is present and needs to be executed
+                if search_query and seq['output'].rstrip().endswith(END_SEARCH_QUERY):
+                    if seq['search_count'] < MAX_SEARCH_LIMIT and search_query not in seq['executed_search_queries']:
+                        # Execute search, use cache if available
+                        if search_query in search_cache:
+                            results = search_cache[search_query] # 如果搜索查询结果在缓存中存在，则从缓存中获取结果
+                            print(f"Using cached search results for query: \"{search_query}\"")
+                        else:
+                            try:
+                                print(f"Execute and cache search for query: \"{search_query}\"")
+                                results = bing_web_search(search_query, bing_subscription_key, bing_endpoint, market='en-US', language='en') # 执行搜索
+                                search_cache[search_query] = results  # 将搜索结果添加到缓存中
+                                print(f"Executed and cached search for query: \"{search_query}\"")
+                            except Exception as e:
+                                print(f"Error during search query '{search_query}': {e}")
+                                search_cache[search_query] = {}
+                                results = {}
+
+                        # Extract relevant information from Bing search results
+                        relevant_info = extract_relevant_info(results)[:top_k] # 从搜索结果中提取出最相关的信息
+                        # relevant_info['search_query'] = search_query
+                        seq['relevant_info'] = relevant_info
+
+                        # Extract URLs and snippets
+                        urls_to_fetch = [it['url'] for it in relevant_info] # 从搜索结果中提取出所有 URL
+                        snippets = {info['url']: info['snippet'] for info in relevant_info if 'snippet' in info} # 创建一个字典 snippets，将 URL 映射到它们对应的片段（snippet）。如果 snippet 字段存在，则把它加入字典
+
+                        # Filter URLs that are not cached
+                        urls_to_fetch_filtered = [u for u in urls_to_fetch if u not in url_cache] # 筛选出所有没有被缓存的 UR
+                        cached_urls = [u for u in urls_to_fetch if u in url_cache] # 选出已经缓存的 URL（即在 url_cache 中存在的 URL）。这些 URL 已经被处理过，不需要再次请求
+
+                        # Store info for all_urls_to_fetch and url_snippets
+                        for url in urls_to_fetch_filtered:
+                            all_urls_to_fetch.add(url)
+                            url_snippets[url] = snippets.get(url, "") # 将每个 URL 对应的片段存储到 url_snippets 字典中
+
+                        all_reasoning_steps = seq['output']
+                        all_reasoning_steps = all_reasoning_steps.replace('\n\n', '\n').split("\n") # 将连续的空行（\n\n）替换为单个换行符（\n），然后按换行符拆分文本，得到每个推理步骤单独的一行
+
+                        truncated_prev_reasoning = ""
+                        for i, step in enumerate(all_reasoning_steps):
+                            truncated_prev_reasoning += f"Step {i + 1}: {step}\n\n" # 遍历 all_reasoning_steps 中的每个步骤，并将每个步骤编号和步骤内容格式化后，添加到 truncated_prev_reasoning 字符串中。这样生成一个包含步骤编号和内容的字符串
+
+                        prev_steps = truncated_prev_reasoning.split('\n\n') # 将推理步骤字符串 truncated_prev_reasoning 按照每两个换行符拆分成多个步骤
+                        if len(prev_steps) <= 5: # 如果步骤的数量不超过 5，直接保留所有步骤
+                            truncated_prev_reasoning = '\n\n'.join(prev_steps)
+                        else:
+                            truncated_prev_reasoning = ''
+                            for i, step in enumerate(prev_steps): # 如果步骤超过 5 个，进行截断。保留第一个步骤、最后四个步骤以及包含搜索查询或搜索结果的步骤。如果中间的步骤不重要，则用 ... 来表示省略
+                                if i == 0 or i >= len(prev_steps) - 4 or BEGIN_SEARCH_QUERY in step or BEGIN_SEARCH_RESULT in step:
+                                    truncated_prev_reasoning += step + '\n\n'
+                                else: # 如果步骤超过 5 个，进行截断。保留第一个步骤、最后四个步骤以及包含搜索查询或搜索结果的步骤。如果中间的步骤不重要，则用 ... 来表示省略
+                                    if truncated_prev_reasoning[-len('\n\n...\n\n'):] != '\n\n...\n\n':
+                                        truncated_prev_reasoning += '...\n\n'
+                        truncated_prev_reasoning = truncated_prev_reasoning.strip('\n')
+
+                        # Collect parameters for batch processing
+                        batch_relevant_info.append(relevant_info) # 搜索出来的信息
+                        batch_original_questions.append(seq['item']['Question']) # 原始问题
+                        batch_prev_reasonings.append(truncated_prev_reasoning) # 之前的推理步骤
+                        batch_search_queries.append(search_query) # 搜索查询
+                        batch_sequences.append(seq)
+
+                        # Update search count and executed queries
+                        seq['search_count'] += 1 # 更新搜索计数
+                        seq['executed_search_queries'].add(search_query) # 将已执行的搜索查询添加到集合中
+
+                    elif seq['search_count'] >= MAX_SEARCH_LIMIT: # 如果搜索次数达到或超过该限制，则返回一条消息，通知该查询无法再进行
+                        limit_message = f"\n{BEGIN_SEARCH_RESULT}\nThe maximum search limit is exceeded. You are not allowed to search.\n{END_SEARCH_RESULT}\n"
+                        seq['prompt'] += limit_message
+                        seq['output'] += limit_message
+                        seq['history'].append(limit_message)
+                        seq["all_info"].append({f"turn_{turn}_search_limited": limit_message})
+                        print(f"Search limit reached for query: \"{search_query}\"")
+
+                    elif search_query in seq['executed_search_queries']: # 如果当前查询已经执行过，则返回一个消息，提示用户查询已重复，并引导其查看之前的结果
+                        limit_message = f"\n{BEGIN_SEARCH_RESULT}\nYou have searched this query. Please refer to previous results.\n{END_SEARCH_RESULT}\n"
+                        seq['prompt'] += limit_message
+                        seq['output'] += limit_message
+                        seq['history'].append(limit_message)
+                        seq["all_info"].append({f"turn_{turn}_search_limited": limit_message})
+                        print(f"Repeated search for query: \"{search_query}\"")
+                        
+
+                else:  # 如果当前序列没有需要执行的搜索查询，则将该序列标记为完成，并打印提示信息
+                    # If no search query needs to be executed, mark the sequence as finished
+                    seq['finished'] = True
+                    print("Sequence marked as complete.")
+
+            print(f"all_urls_to_fetch len: {len(all_urls_to_fetch)}, url_cache len: {len(url_cache)}")
+            print(f"all_urls_to_fetch: {all_urls_to_fetch}")
+            # Batch fetch all URLs at once to optimize speed
+            
+            if all_urls_to_fetch:
+                print(f"Fetching {len(all_urls_to_fetch)} URLs...")
+                try:
+                    fetched_contents = fetch_page_content( # 一次性获取所有 URL 的中搜索出来的内容
+                        list(all_urls_to_fetch),
+                        use_jina=use_jina,
+                        jina_api_key=jina_api_key,
+                        # snippets=url_snippets  # Do not pass snippets when updating url_cache directly
+                    )
+                    print(f"Fetched {len(fetched_contents)} URLs successfully.")
+                except Exception as e:
+                    print(f"Error during batch URL fetching: {e}")
+                    fetched_contents = {url: f"Error fetching URL: {e}" for url in all_urls_to_fetch}
+                # Update cache with fetched contents
+                for url, content in fetched_contents.items(): # 将获取的内容添加到 url_cache 中
+                    url_cache[url] = content
+
+            # After fetching, prepare formatted documents for batch processing
+            for relevant_info in batch_relevant_info:
+                formatted_documents = "" # 初始化一个空字符串 formatted_documents，用于拼接本次要处理的所有网页信息。后面会将其添加到 batch_documents 列表中
+                for i, doc_info in enumerate(relevant_info):
+                    url = doc_info['url']
+                    raw_context = url_cache.get(url, "") # 获取 url 对应的内容
+                    doc_info['snippet'] = doc_info['snippet'].replace('<b>','').replace('</b>','')            
+                    success, filtered_context = extract_snippet_with_context(raw_context, doc_info['snippet'], context_chars=max_doc_len)
+                    if success:
+                        print("Extracted snippet with context successfully.")
+                        context = filtered_context
+                    else: # 否则就取 raw_context 的前 max_doc_len * 2 个字符，作为一个有限的片段，避免过长导致后续处理负担
+                        print("Failed to extract snippet with context, using raw context.")
+                        context = raw_context[:max_doc_len*2]
+
+                    doc_info['context'] = context
+                    formatted_documents += f"**Web Page {i + 1}:**\n"
+                    formatted_documents += json.dumps(doc_info, ensure_ascii=False, indent=2) + "\n"
+                    
+                batch_documents.append(formatted_documents) # 将本组搜索结果的所有信息（拼接成的字符串 formatted_documents）添加到 batch_documents 列表中
+
+            # After fetching, prepare for batch processing if there are any
+            if batch_sequences:
+                print(f"Batch processing {len(batch_sequences)} sequences with generate_webpage_to_reasonchain_batch...")
+                webpage_analyses = generate_webpage_to_reasonchain_batch( # 根据前面处理好的结果，生成新的推理
+                    original_questions=batch_original_questions,
+                    prev_reasonings=batch_prev_reasonings,
+                    search_queries=batch_search_queries,
+                    documents=batch_documents,
+                    dataset_name=dataset_name,
+                    batch_output_records=batch_output_records,  # Pass the collection list
+                    max_tokens=max_tokens,
+                )
+                print("Batch generation completed, assigning outputs to sequences...")
+
+                # 拼接同一个query总结的结果
+
+                for seq, analysis,doc in zip(batch_sequences, webpage_analyses, batch_documents): # 遍历批处理返回的 webpage_analyses，将处理结果与相应的序列 seq 进行一一对应
+                    if isinstance(analysis, str): # 判断 analysis 是否是纯字符串。如果是字符串，说明直接可以追加到序列的文本中
+                        append_text = f"\n\n{BEGIN_SEARCH_RESULT}{analysis}{END_SEARCH_RESULT}\n\n" # 封装处理结果，添加到序列的历史记录、提示和输出中
+                        seq['prompt'] += append_text
+                        seq['output'] += append_text
+                        seq['history'].append(append_text) # 存的是每一次的webpage_analyses
+                        seq['all_info'].extend([{f"turn_{turn}_search": doc}, {f"turn_{turn}_webpage_analyses": analysis}])
+                    else: # 如果 analysis 不是纯字符串，那么可能是一种特殊的结构，比如表示需要替换推理步骤的 dict 或其他格式
+                        append_text = replace_recent_steps(seq['output'], analysis)
+                        seq['prompt'] += append_text
+                        seq['output'] += append_text
+                        seq['history'].append(append_text)
+                        seq['all_info'].extend([{f"turn_{turn}_search": doc}, {f"turn_{turn}_webpage_analyses": analysis}])
+
+        # Check if all sequences are finished
+        # 保存active_sequences
+        active_sequences_part = [{ # 记录每个问题的搜索历史
+            'item': ele["item"],
+            'prompt': ele['prompt'],
+            'output': ele["output"],
+            'finished': ele["finished"], # 一开始均为未完成
+            'history':ele["history"],
+            'search_count': ele["search_count"],
+            'all_info': ele['all_info']
+        } for ele in active_sequences]
+        with open(os.path.join(output_dir, f"turn_{turn}.json"), 'w', encoding='utf-8') as f:
+            json.dump(active_sequences_part, f, ensure_ascii=False, indent=2)
+        unfinished = [seq for seq in active_sequences if not seq['finished']] # 是否结束是基于模型是否生成了新的搜索
+        if not unfinished:
+            break
+        else:
+            if turn >= MAX_TURN:
+                print(f"Maximum number of turns ({MAX_TURN}) reached, stopping.")
+                break
+
+    total_time = time.time() - start_time
+    print(f"Total time taken: {total_time} seconds")
+
+    # ---------------------- Save Batch Output Records to JSON File ----------------------
+    # Define output JSON file path
+    t = time.localtime()
+    batch_output_file = os.path.join(output_dir, f'{split}.{t.tm_mon}.{t.tm_mday},{t.tm_hour}:{t.tm_min}.info_extract.json')
+
+    # Save batch_output_records to JSON file
+    with open(batch_output_file, 'w', encoding='utf-8') as f: # 这里存的是webpage推理时的输入和输出和提取后的信息
+        json.dump(batch_output_records, f, ensure_ascii=False, indent=2)
+
+    print(f"Batch outputs saved to {batch_output_file}")
+
+    # Prepare output list for evaluation
+    output_list = [seq['output'] for seq in active_sequences]
+
+    # Run evaluation
+    run_evaluation(filtered_data, input_list, output_list, dataset_name, output_dir, total_time, split)
+
+    # 评测has answer信息
+    turn_files = os.listdir(output_dir)
+    turn_files = [file for file in turn_files if file.startswith("turn_")]
+    max_turn_file = max(turn_files, key=lambda x: int(re.search(r'turn_(\d+)', x).group(1)))
+
+    max_turn_file_path = os.path.join(output_dir, max_turn_file)
+    print(f"max_turn_file_path: {max_turn_file_path}")
+    add_eval(model_path, max_turn_file_path)
+
+    # ---------------------- Update Search and URL Cache ----------------------
+    print('Updating Search and URL Cache...')
+    # Load existing caches or initialize empty dictionaries
+    if os.path.exists(search_cache_path):
+        with open(search_cache_path, 'r', encoding='utf-8') as f:
+            search_cache_new = json.load(f)
+    else:
+        search_cache_new = {}
+
+    if os.path.exists(url_cache_path):
+        with open(url_cache_path, 'r', encoding='utf-8') as f:
+            url_cache_new = json.load(f)
+    else:
+        url_cache_new = {}
+
+    search_cache.update(search_cache_new)
+    url_cache.update(url_cache_new)
+
+    save_caches()
+
+    print("Process completed.")
+
+if __name__ == "__main__":
+    main()

deep_search/data_syn/search_o1_data_syn_sum_single_page.py

@@ -0,0 +1,953 @@
+# run_search_o1.py
+import os
+import json
+import time
+import re
+from tqdm import tqdm
+import numpy as np
+import torch
+import string
+from typing import Optional, Tuple, List, Dict
+import argparse
+
+from transformers import AutoTokenizer
+from vllm import LLM, SamplingParams
+
+from bing_search import (
+    bing_web_search, 
+    extract_relevant_info, 
+    fetch_page_content, 
+    extract_snippet_with_context
+)
+from evaluate import (
+    run_evaluation, 
+    extract_answer
+)
+from prompts import (
+    get_gpqa_search_o1_instruction, 
+    get_math_search_o1_instruction, 
+    get_code_search_o1_instruction, 
+    get_singleqa_search_o1_instruction, 
+    get_multiqa_search_o1_instruction, 
+    get_webpage_to_reasonchain_instruction,
+    get_task_instruction_openqa, 
+    get_task_instruction_math, 
+    get_task_instruction_multi_choice, 
+    get_task_instruction_code, 
+    get_singleqa_search_o1_instruction_1,
+    get_multiqa_search_o1_instruction_1,
+    get_webpage_to_reasonchain_instruction_1,
+    get_math_search_o1_instruction_1,
+)
+
+
+from openai import OpenAI
+
+from add_eval import add_eval
+# Modify OpenAI's API key and API base to use vLLM's API server.
+# 使用 vLLM 的 API 服务器需要修改 OpenAI 的 API 密钥和 API 库。
+
+# openai_api_key = "EMPTY"
+# openai_api_base = "http://localhost:8000/v1"
+# client = OpenAI(
+#     api_key=openai_api_key,
+#     base_url=openai_api_base,
+# )
+
+# Define special tokens
+BEGIN_SEARCH_QUERY = "<|begin_search_query|>"
+END_SEARCH_QUERY = "<|end_search_query|>"
+BEGIN_SEARCH_RESULT = "<|begin_search_result|>"
+END_SEARCH_RESULT = "<|end_search_result|>"
+
+
+# os.environ['http_proxy'] = 'http://127.0.0.1:7880'
+# os.environ['https_proxy'] = 'http://127.0.0.1:7880'
+
+# 增加了cache共享和has answer评测,truncate doc
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Run Search O1 for various datasets and models.")
+
+    # Dataset and split configuration
+    parser.add_argument(
+        '--dataset_name',
+        type=str,
+        required=True,
+        choices=['simpleqa', 'gpqa', 'math500', 'aime', 'amc', 'livecode', 'nq', 'triviaqa', 'hotpotqa', '2wiki', 'musique', 'bamboogle'],
+        help="Name of the dataset to use."
+    )
+
+    parser.add_argument(
+        '--split',
+        type=str,
+        required=True,
+        choices=['test', 'diamond', 'main', 'extended'],
+        help="Dataset split to use."
+    )
+
+    parser.add_argument(
+        '--subset_num',
+        type=int,
+        default=-1,
+        help="Number of examples to process. Defaults to all if not specified."
+    )
+
+    # Search and document retrieval configuration
+    parser.add_argument(
+        '--max_search_limit',
+        type=int,
+        default=10,
+        help="Maximum number of searches per question."
+    )
+
+    parser.add_argument(
+        '--max_turn',
+        type=int,
+        default=15,
+        help="Maximum number of turns."
+    )
+
+    parser.add_argument( # 使用搜索引擎时，返回的最大文档数
+        '--top_k',
+        type=int,
+        default=10,
+        help="Maximum number of search documents to return."
+    )
+
+    parser.add_argument(
+        '--max_doc_len',
+        type=int,
+        default=3000,
+        help="Maximum length of each searched document."
+    )
+
+    # parser.add_argument(
+    #     '--use_jina',
+    #     type=bool,
+    #     default=False,
+    #     help="Whether to use Jina API for document fetching."
+    # )
+    parser.add_argument(
+        '--use_jina',
+        action='store_true',
+        help="Whether to use Jina API for document fetching."
+    )
+
+    parser.add_argument(
+        '--jina_api_key',
+        type=str,
+        default='None',
+        help="Your Jina API Key to Fetch URL Content."
+    )
+
+    # Model configuration
+    parser.add_argument(
+        '--model_path',
+        type=str,
+        required=True,
+        help="Path to the pre-trained model."
+    )
+
+    # Sampling parameters
+    parser.add_argument(
+        '--temperature',
+        type=float,
+        default=0.7,
+        help="Sampling temperature."
+    )
+
+    parser.add_argument(
+        '--top_p',
+        type=float,
+        default=0.8,
+        help="Top-p sampling parameter."
+    )
+
+    parser.add_argument(
+        '--top_k_sampling',
+        type=int,
+        default=20,
+        help="Top-k sampling parameter."
+    )
+
+    parser.add_argument(
+        '--repetition_penalty',
+        type=float,
+        default=None,
+        help="Repetition penalty. If not set, defaults based on the model."
+    )
+
+    parser.add_argument(
+        '--max_tokens',
+        type=int,
+        default=32768,
+        help="Maximum number of tokens to generate. If not set, defaults based on the model and dataset."
+    )
+
+    # Bing API Configuration
+    parser.add_argument(
+        '--bing_subscription_key',
+        type=str,
+        required=True,
+        help="Bing Search API subscription key."
+    )
+
+    parser.add_argument(
+        '--bing_endpoint',
+        type=str,
+        default="https://api.bing.microsoft.com/v7.0/search",
+        help="Bing Search API endpoint."
+    )
+
+    parser.add_argument(
+        '--cache_dir_base',
+        type=str,
+        required=True,
+        help="cache path."
+    )
+
+    parser.add_argument(
+        '--output_dir_base',
+        type=str,
+        required=True,
+        help="output_dir"
+    )
+
+    # parser.add_argument(
+    #     '--model_doc_reason_path',
+    #     type=str,
+    #     required=True,
+    #     help="Path to the document reasoning model."
+    # )
+
+    # openai_api_base
+    # parser.add_argument(
+    #     '--openai_api_base',
+    #     type=str,
+    #     required=True,
+    #     help="openai_api_base"
+    # )
+    # parser.add_argument(
+    #     '--data_path',
+    #     type=str,
+    #     required=True,
+    #     help="Path to the document reasoning model."
+    # )
+    return parser.parse_args()
+
+def main():
+    args = parse_args()
+    print(f"args.use_jina: {args.use_jina}")
+    # Extract arguments
+    dataset_name = args.dataset_name
+    split = args.split
+    subset_num = args.subset_num
+    MAX_SEARCH_LIMIT = args.max_search_limit
+    MAX_TURN = args.max_turn
+    top_k = args.top_k
+    max_doc_len = args.max_doc_len
+    model_path = args.model_path
+    # model_doc_reason_path = args.model_doc_reason_path
+    temperature = args.temperature
+    top_p = args.top_p
+    top_k_sampling = args.top_k_sampling
+    repetition_penalty = args.repetition_penalty
+    max_tokens = args.max_tokens
+    bing_subscription_key = args.bing_subscription_key
+    bing_endpoint = args.bing_endpoint
+    use_jina = args.use_jina
+    jina_api_key = args.jina_api_key
+    cache_dir_base = args.cache_dir_base
+    output_dir_base = args.output_dir_base
+    # openai_api_base = args.openai_api_base
+    use_jina = False
+    print(f"use_jina: {use_jina}")
+
+    print(f"CUDA_VISIBLE_DEVICES is set to: {os.environ['CUDA_VISIBLE_DEVICES']}")
+
+    # openai_api_key = "EMPTY"
+    # openai_api_base = openai_api_base
+    # client = OpenAI(
+    #     api_key=openai_api_key,
+    #     base_url=openai_api_base,
+    # )
+
+    # Adjust parameters based on dataset
+    if dataset_name in ['nq', 'triviaqa', 'hotpotqa', 'musique', 'bamboogle', '2wiki', 'medmcqa', 'pubhealth']:
+        MAX_SEARCH_LIMIT = 5
+        if dataset_name in ['simpleqa' ,'hotpotqa', 'musique', 'bamboogle', '2wiki']:
+            MAX_SEARCH_LIMIT = 10
+            MAX_TURN = 15
+        top_k = 10
+        max_doc_len = 3000
+    
+    if args.jina_api_key == 'None':
+        jina_api_key = None
+
+    # Set default repetition_penalty if not provided
+    if repetition_penalty is None:
+        repetition_penalty = 1.05 if 'qwq' in model_path.lower() else 1.0
+
+    # Data paths based on dataset
+    if split == "test": # 测试用的数据集地址
+        data_path = f"./data/test/{dataset_name}.json"
+    else: # 训练用的数据集地址
+        if dataset_name == 'livecode':
+            data_path = f'./data/LiveCodeBench/{split}.json'
+        elif dataset_name in ['math500', 'gpqa', 'aime', 'amc']:
+            data_path = f'./data/{dataset_name.upper()}/{split}.json'
+        else:
+            data_path = f'./data/QA_Datasets/{dataset_name}.json'
+
+    print('-----------------------')
+    print(f'Using {dataset_name} {split} set.')
+    print('-----------------------')
+
+    # ---------------------- Caching Mechanism ----------------------
+    # Define cache directories and file paths
+    # cache_dir = './cache'
+    model_name = model_path.split('/')[-1].replace('-instruct', '')
+    # cache_dir = f'./{cache_dir_base}_{dataset_name}_{model_name}'
+    cache_dir = cache_dir_base
+    search_cache_path = os.path.join(cache_dir, 'search_cache.json')
+    url_cache_path = os.path.join(cache_dir, 'url_cache.json')
+
+    # Ensure cache directory exists
+    os.makedirs(cache_dir, exist_ok=True)
+
+    # Load existing caches or initialize empty dictionaries
+    if os.path.exists(search_cache_path):
+        with open(search_cache_path, 'r', encoding='utf-8') as f:
+            search_cache = json.load(f)
+    else:
+        search_cache = {}
+
+    if os.path.exists(url_cache_path):
+        with open(url_cache_path, 'r', encoding='utf-8') as f:
+            url_cache = json.load(f)
+    else:
+        url_cache = {}
+
+    # Function to save caches
+    def save_caches():
+        with open(search_cache_path, 'w', encoding='utf-8') as f:
+            json.dump(search_cache, f, ensure_ascii=False, indent=2)
+        with open(url_cache_path, 'w', encoding='utf-8') as f:
+            json.dump(url_cache, f, ensure_ascii=False, indent=2)
+
+    # ---------------------- Model Loading ----------------------
+    print(f"Loading tokenizer from {model_path}...")
+    tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
+    if tokenizer.pad_token is None:
+        tokenizer.pad_token = tokenizer.eos_token
+    tokenizer.padding_side = 'left' # 主要是左填充
+    print("Tokenizer loaded successfully.")
+
+    # Define output directory based on model and dataset
+  
+    
+    # if 'qwq' in model_path.lower():
+    #     if dataset_name in ['math500', 'gpqa', 'aime', 'amc', 'livecode']:
+    #         output_dir = f'./{output_dir_base}/{dataset_name}.qwq.search_o1'
+    #         if dataset_name == 'gpqa' and (MAX_SEARCH_LIMIT != 5 or top_k != 10):
+    #             output_dir = f'./{output_dir_base}/runs.analysis/{dataset_name}.qwq.search_o1.{MAX_SEARCH_LIMIT}.{top_k}'
+    #     else:
+    #         output_dir = f'./{output_dir_base}/runs.qa/{dataset_name}.qwq.search_o1'
+    # else:
+    #     model_short_name = model_path.split('/')[-1].lower().replace('-instruct', '')
+    #     output_dir = f'./{output_dir_base}/runs.baselines/{dataset_name}.{model_short_name}.search_o1'
+    output_dir = output_dir_base
+    os.makedirs(output_dir, exist_ok=True)
+
+    print(f"Loading model from {model_path}...")
+    print(f"device_count: {torch.cuda.device_count()}")
+    
+    # Initialize the LLM
+    llm = LLM(
+        model=model_path,
+        tensor_parallel_size=torch.cuda.device_count(),
+        gpu_memory_utilization=0.95,
+
+    )
+    print("Model loaded successfully.")
+
+    # # ----------------------Loading model to reason in document ----------------------
+
+    # print(f"Loading tokenizer_doc_reason from {model_doc_reason_path}...")
+    # tokenizer_doc_reason = AutoTokenizer.from_pretrained(model_doc_reason_path, trust_remote_code=True)
+    # if tokenizer_doc_reason.pad_token is None:
+    #     tokenizer_doc_reason.pad_token = tokenizer_doc_reason.eos_token
+    # tokenizer_doc_reason.padding_side = 'left' # 主要是左填充
+    # print("tokenizer_doc_reason loaded successfully.")
+
+    # print(f"Loading tokenizer_doc_reason from {model_doc_reason_path}...")
+    
+    # # Initialize the LLM
+    # # torch.cuda.set_device(6,7)
+    
+    # llm_doc_reason = LLM(
+    #     model=model_doc_reason_path,
+    #     tensor_parallel_size=2,
+    #     gpu_memory_utilization=0.95,
+
+    # )
+    # print("Model_doc_reason loaded successfully.")
+
+    # ---------------------- Data Loading ----------------------
+    print(f"Loading data from {data_path}...")
+    with open(data_path, 'r', encoding='utf-8') as json_file:
+        filtered_data = json.load(json_file)
+    print(f"Data loaded successfully. Total examples: {len(filtered_data)}")
+
+    # ---------------------- Batch Generation Function ----------------------
+    def generate_webpage_to_reasonchain_batch( # 模型基于网页内容生成推理，然后从模型的回复中提取答案
+            original_questions: List[str],
+            prev_reasonings: List[str],
+            search_queries: List[str],
+            documents: List[List[str]],
+            dataset_name: str,
+            batch_output_records: List[Dict],  # New parameter to collect outputs
+            max_tokens: int = 32768,
+            coherent: bool = False,
+        ) -> List[str]:
+
+            # if "Qwen2.5" in model_path:
+            #     max_tokens = 8192
+                
+            #     encode_docs = tokenizer(documents, truncation=True, max_length=20000, add_special_tokens=False)["input_ids"]
+            #     documents = tokenizer.batch_decode(encode_docs)
+
+            #     # 统计每个文档的长度
+            #     doc_lengths = [len(doc) for doc in encode_docs]
+
+            #     # # 打印每个文档的长度
+            #     # for i, length in enumerate(doc_lengths):
+            #     #     print(f"Document {i + 1}: {length} tokens")
+
+            #     # 如果需要返回长度列表，可以直接使用 doc_lengths
+            #     print(f"for {model_path}, set max_tokens={max_tokens} for doc gen, truncate documnets. ")
+            #     print("All document lengths:", doc_lengths)
+
+            user_prompts = []
+            assert len(original_questions) == len(prev_reasonings) == len(search_queries) == len(documents), "Input lists must have the same length"
+            
+            for i, doc_str_list in enumerate(documents):
+                assert len(doc_str_list) == top_k, f"Expected {top_k} documents, but got {len(doc_str_list)}"
+                for j, doc_str in enumerate(doc_str_list):
+                    r = prev_reasonings[i]
+                    sq = search_queries[i]
+                    user_prompts.append(get_webpage_to_reasonchain_instruction(r, sq, doc_str))
+
+            # user_prompts = [ # 根据之前的推理，搜索query和搜索到的doc生成用户提示
+            #     get_webpage_to_reasonchain_instruction(r, sq, doc)
+            #     for r, sq, doc in zip(prev_reasonings, search_queries, documents)
+            # ]
+
+            prompts = [{"role": "user", "content": up} for up in user_prompts]
+            prompts = [tokenizer.apply_chat_template([p], tokenize=False, add_generation_prompt=True) for p in prompts]
+
+            output = llm.generate( # 生成模型回复
+                prompts,
+                sampling_params=SamplingParams(
+                    max_tokens=max_tokens,
+                    temperature=0.7,
+                    top_p=0.8,
+                    top_k=20,
+                    repetition_penalty=1.05,
+                )
+            )
+
+            raw_outputs = [out.outputs[0].text for out in output]
+            extracted_infos = [extract_answer(raw, mode='infogen') for raw in raw_outputs] # 提取模型基于网页生成的推理
+
+            formatted_infos = []
+            formatted_outputs = []
+            formatted_prompts = []
+            formatted_info = ""
+            formatted_output = []
+            formatted_prompt = []
+            for i, (p, e, r) in enumerate(zip(prompts, extracted_infos, raw_outputs)):
+                if i % top_k == 0 and i != 0:
+                    formatted_infos.append(formatted_info)
+                    formatted_outputs.append(formatted_output)
+                    formatted_prompts.append(formatted_prompt)
+                    formatted_info = ""
+                    formatted_output = []
+                    formatted_prompt = []
+                    
+                formatted_info += f"Summary of Web Page {i % top_k + 1}:\n{e}\n\n"
+                formatted_output.append(r)
+                formatted_prompt.append(p)
+            
+            # 最后一组
+            formatted_infos.append(formatted_info)
+            formatted_outputs.append(formatted_output)
+            formatted_prompts.append(formatted_prompt)
+
+            assert len(formatted_infos) == len(formatted_outputs) == len(formatted_prompts) == len(original_questions), "Number of formatted_infos and formatted_outputs must match number of questions"
+            for i, (p, r, e) in enumerate(zip(formatted_prompts, formatted_outputs, formatted_infos)):
+                batch_output_records.append({
+                    'prompt': p,
+                    'raw_output': r,
+                    'extracted_info': e
+                })
+
+            return extracted_infos
+
+    # ---------------------- Preparation of Input Prompts ----------------------
+    input_list = []
+    for item in filtered_data: # 生成prompts
+        question = item['Question']
+
+        if dataset_name in ['nq', 'triviaqa', 'hotpotqa', 'musique', 'bamboogle', '2wiki']:
+            if dataset_name in ['nq', 'triviaqa']:
+                instruction = get_singleqa_search_o1_instruction_1(MAX_SEARCH_LIMIT)
+            elif dataset_name in ['hotpotqa', 'musique', 'bamboogle', '2wiki']:
+                instruction = get_multiqa_search_o1_instruction_1(MAX_SEARCH_LIMIT)
+            if 'qwq' in model_path.lower():
+                user_prompt = get_task_instruction_openqa(question, model_name='qwq')
+            else:
+                user_prompt = get_task_instruction_openqa(question)
+
+        elif dataset_name in ['math500', 'aime', 'amc']:
+            instruction = get_math_search_o1_instruction_1(MAX_SEARCH_LIMIT)
+            if 'qwq' in model_path.lower():
+                user_prompt = get_task_instruction_math(question, model_name='qwq')
+            else:
+                user_prompt = get_task_instruction_math(question)
+
+        elif dataset_name == 'gpqa':
+            instruction = get_gpqa_search_o1_instruction(MAX_SEARCH_LIMIT)
+            if 'qwq' in model_path.lower():
+                user_prompt = get_task_instruction_multi_choice(question, model_name='qwq')
+            elif 'llama' in model_path.lower():
+                user_prompt = get_task_instruction_multi_choice(question, model_name='llama')
+            else:
+                user_prompt = get_task_instruction_multi_choice(question)
+
+        elif dataset_name == 'livecode':
+            instruction = get_code_search_o1_instruction(MAX_SEARCH_LIMIT)
+            question_title = item.get('question_title', '')
+            if 'qwq' in model_path.lower():
+                user_prompt = get_task_instruction_code(question, question_title=question_title, model_name='qwq')
+            else:
+                user_prompt = get_task_instruction_code(question)
+        else:
+            user_prompt = ""  # Default to empty if dataset not matched
+
+        prompt = [{"role": "user", "content": instruction + user_prompt}] # instruction是告诉模型怎么进行搜索，user_prompt是用户具体问题
+        prompt = tokenizer.apply_chat_template(prompt, tokenize=False, add_generation_prompt=True)
+        input_list.append(prompt)
+
+    if subset_num != -1:
+        input_list = input_list[:subset_num]
+        filtered_data = filtered_data[:subset_num]
+
+    # Initialize active sequences
+    active_sequences = [{ # 记录每个问题的搜索历史
+        'item': item,
+        'prompt': prompt,
+        'output': '',
+        'finished': False, # 一开始均为未完成
+        'history': [],
+        'search_count': 0,
+        'executed_search_queries': set(),
+        'all_info': [],
+    } for item, prompt in zip(filtered_data, input_list)]
+
+    # ---------------------- Set Max Tokens ----------------------
+    # if 'qwq' in model_path.lower():
+    #     if dataset_name in ['aime', 'amc', 'livecode']:
+    #         max_tokens = 32768
+    #     else:
+    #         max_tokens = 20480
+    # else:
+    #     max_tokens = 8192
+    # max_tokens = 16384
+    if dataset_name in ['aime', 'amc', 'livecode']:
+        max_tokens = 32768
+    else:
+        max_tokens = 20480
+    # ---------------------- Generation Function ----------------------
+    def run_generation(sequences: List[Dict], max_tokens: int) -> List:
+        prompts = [s['prompt'] for s in sequences] # 提取prompt
+        sampling_params = SamplingParams(
+            max_tokens=max_tokens,
+            temperature=temperature,
+            top_p=top_p,
+            top_k=top_k_sampling,
+            repetition_penalty=repetition_penalty,
+            stop=[END_SEARCH_QUERY, tokenizer.eos_token],
+            include_stop_str_in_output=True,
+        )
+        output_list = llm.generate(prompts, sampling_params=sampling_params) # 模型根据prompt生成回答
+        print(f"run_generation completed {len(output_list)}")
+        return output_list
+
+    # Function to extract text between two tags 提取位于 start_tag 和 end_tag 之间的内容
+    def extract_between(text: str, start_tag: str, end_tag: str) -> Optional[str]:
+        pattern = re.escape(start_tag) + r"(.*?)" + re.escape(end_tag)
+        matches = re.findall(pattern, text, flags=re.DOTALL)
+        if matches:
+            return matches[-1].strip()
+        return None
+
+    def replace_recent_steps(origin_str, replace_str): # 使用replace_str更新origin_str
+        """
+        Replaces specific steps in the original reasoning steps with new steps.
+        If a replacement step contains "DELETE THIS STEP", that step is removed.
+
+        Parameters:
+        - origin_str (str): The original reasoning steps.
+        - replace_str (str): The steps to replace or delete.
+
+        Returns:
+        - str: The updated reasoning steps after applying replacements.
+        这个函数的主要功能是替换给定的推理步骤（reasoning steps），
+        并根据传入的 replace_str 进行更新。
+        如果新的步骤包含 "DELETE THIS STEP"，则删除该步骤
+        """
+
+        def parse_steps(text):
+            """
+            Parses the reasoning steps from a given text.
+
+            Parameters:
+            - text (str): The text containing reasoning steps.
+
+            Returns:
+            - dict: A dictionary mapping step numbers to their content.
+            """
+            step_pattern = re.compile(r"Step\s+(\d+):\s*") # 这个模式会匹配 "Step" 后面跟一个或多个空格，然后是一个数字（步骤编号），最后是冒号
+            steps = {}
+            current_step_num = None
+            current_content = []
+
+            for line in text.splitlines(): # 将输入的文本按行分割，并逐行遍历。每一行会被检查是否包含一个步骤
+                step_match = step_pattern.match(line)
+                if step_match: # 匹配到一个新的步骤
+                    # If there's an ongoing step, save its content，如果当前的步骤不为空，将其为上一个步骤，先将上一个步骤的内容（存在current_content中）保存，然后再更新current_step_num和current_content
+                    if current_step_num is not None:
+                        steps[current_step_num] = "\n".join(current_content).strip()
+                    current_step_num = int(step_match.group(1))
+                    content = line[step_match.end():].strip()
+                    current_content = [content] if content else []
+                else:
+                    if current_step_num is not None:
+                        current_content.append(line)
+            
+            # Save the last step if any
+            if current_step_num is not None: # 保存最后一个步骤
+                steps[current_step_num] = "\n".join(current_content).strip()
+            
+            return steps
+
+        # Parse the original and replacement steps
+        origin_steps = parse_steps(origin_str) # 解析原始的推理步骤
+        replace_steps = parse_steps(replace_str) # 解析要替换的推理步骤
+
+        # Apply replacements
+        for step_num, content in replace_steps.items(): # 遍历要替换的步骤
+            if "DELETE THIS STEP" in content: 
+                # Remove the step if it exists
+                if step_num in origin_steps: # 如果要删除的步骤在原始的推理步骤中存在，则删除该步骤
+                    del origin_steps[step_num]
+            else: # 如果要替换的步骤不是要删除的步骤，则替换该步骤
+                # Replace or add the step 
+                origin_steps[step_num] = content
+
+        # Sort the steps by step number
+        sorted_steps = sorted(origin_steps.items()) # 按照步骤编号对步骤进行排序
+
+        # Reconstruct the reasoning steps as a single string
+        new_reasoning_steps = "\n\n".join([f"{content}" for num, content in sorted_steps]) # 根据排序后的步骤构建新的推理步骤字符串，步骤之间以两个换行符分隔
+
+        return new_reasoning_steps
+
+    # ---------------------- Initialize Collection Structure ----------------------
+    # Initialize a list to collect batch outputs
+    batch_output_records = []
+
+    start_time = time.time()
+    turn = 0
+
+    # 流程
+    # 首先根据prompt让模型生成回复
+    # 从模型的回复中提取搜索查询
+    # 如果有（回复要以END_SEARCH_QUERY结尾）
+    #     根据搜索查询，从互联网上提取相关信息
+    #     处理查询的信息
+    #     让模型基于之前的步骤，检索query和查询到的信息生成新的推理，得到search result
+    #     然后回到第一步（这里模型就会根据前面的search result，再次生成新的回复
+    # 如果没有查询则该条问题结束
+
+    # Main loop until all sequences are finished or maximum turns reached
+    while True:
+        # Identify sequences that need generation
+        sequences_needing_generation = [seq for seq in active_sequences if not seq['finished']] # 筛选出需要生成的新内容的序列，active_sequences 是一个包含所有活跃序列的列表
+
+        if sequences_needing_generation:
+            turn += 1
+            print(f'\n-------------- Turn {turn} --------------')
+            print(f"We have {len(sequences_needing_generation)} sequences needing generation...")
+            outputs = run_generation(sequences_needing_generation, max_tokens) # 根据prompt
+            print("Generation completed, processing outputs...")
+
+            # Initialize batch variables
+            batch_relevant_info = []
+            batch_original_questions = []
+            batch_prev_reasonings = []
+            batch_search_queries = []
+            batch_documents = []
+            batch_sequences = []
+
+            # Collect URLs to fetch across all sequences
+            all_urls_to_fetch = set() # 初始化一个集合 all_urls_to_fetch 用来收集所有需要获取的 URL
+            url_snippets = {}
+            url_sequence_map = {}  # Map URL to list of sequences needing it
+
+            # Process each sequence and collect URLs
+            for seq, out in zip(sequences_needing_generation, outputs): # 遍历需要生成新内容的序列，并生成新内容，同时收集需要获取的 URL
+                text = out.outputs[0].text # 将生成的文本添加到序列的历史记录、提示和输出中
+                seq['history'].append(text)
+                # Append generated text to prompt and output
+                seq['prompt'] += text
+                seq['output'] += text
+                seq['all_info'].append({f"turn_{turn}_reason": text})
+                # Extract search query
+                search_query = extract_between(text, BEGIN_SEARCH_QUERY, END_SEARCH_QUERY) # 提取搜索查询
+
+                # If a search query is present and needs to be executed
+                if search_query and seq['output'].rstrip().endswith(END_SEARCH_QUERY):
+                    if seq['search_count'] < MAX_SEARCH_LIMIT and search_query not in seq['executed_search_queries']:
+                        # Execute search, use cache if available
+                        if search_query in search_cache:
+                            results = search_cache[search_query] # 如果搜索查询结果在缓存中存在，则从缓存中获取结果
+                            print(f"Using cached search results for query: \"{search_query}\"")
+                        else:
+                            try:
+                                print(f"Execute and cache search for query: \"{search_query}\"")
+                                results = bing_web_search(search_query, bing_subscription_key, bing_endpoint, market='en-US', language='en') # 执行搜索
+                                search_cache[search_query] = results  # 将搜索结果添加到缓存中
+                                print(f"Executed and cached search for query: \"{search_query}\"")
+                            except Exception as e:
+                                print(f"Error during search query '{search_query}': {e}")
+                                search_cache[search_query] = {}
+                                results = {}
+
+                        # Extract relevant information from Bing search results
+                        relevant_info = extract_relevant_info(results)[:top_k] # 从搜索结果中提取出最相关的信息
+                        seq['relevant_info'] = relevant_info
+
+                        # Extract URLs and snippets
+                        urls_to_fetch = [it['url'] for it in relevant_info] # 从搜索结果中提取出所有 URL
+                        snippets = {info['url']: info['snippet'] for info in relevant_info if 'snippet' in info} # 创建一个字典 snippets，将 URL 映射到它们对应的片段（snippet）。如果 snippet 字段存在，则把它加入字典
+
+                        # Filter URLs that are not cached
+                        urls_to_fetch_filtered = [u for u in urls_to_fetch if u not in url_cache] # 筛选出所有没有被缓存的 UR
+                        cached_urls = [u for u in urls_to_fetch if u in url_cache] # 选出已经缓存的 URL（即在 url_cache 中存在的 URL）。这些 URL 已经被处理过，不需要再次请求
+
+                        # Store info for all_urls_to_fetch and url_snippets
+                        for url in urls_to_fetch_filtered:
+                            all_urls_to_fetch.add(url)
+                            url_snippets[url] = snippets.get(url, "") # 将每个 URL 对应的片段存储到 url_snippets 字典中
+
+                        all_reasoning_steps = seq['output']
+                        all_reasoning_steps = all_reasoning_steps.replace('\n\n', '\n').split("\n") # 将连续的空行（\n\n）替换为单个换行符（\n），然后按换行符拆分文本，得到每个推理步骤单独的一行
+
+                        truncated_prev_reasoning = ""
+                        for i, step in enumerate(all_reasoning_steps):
+                            truncated_prev_reasoning += f"Step {i + 1}: {step}\n\n" # 遍历 all_reasoning_steps 中的每个步骤，并将每个步骤编号和步骤内容格式化后，添加到 truncated_prev_reasoning 字符串中。这样生成一个包含步骤编号和内容的字符串
+
+                        prev_steps = truncated_prev_reasoning.split('\n\n') # 将推理步骤字符串 truncated_prev_reasoning 按照每两个换行符拆分成多个步骤
+                        if len(prev_steps) <= 5: # 如果步骤的数量不超过 5，直接保留所有步骤
+                            truncated_prev_reasoning = '\n\n'.join(prev_steps)
+                        else:
+                            truncated_prev_reasoning = ''
+                            for i, step in enumerate(prev_steps): # 如果步骤超过 5 个，进行截断。保留第一个步骤、最后四个步骤以及包含搜索查询或搜索结果的步骤。如果中间的步骤不重要，则用 ... 来表示省略
+                                if i == 0 or i >= len(prev_steps) - 4 or BEGIN_SEARCH_QUERY in step or BEGIN_SEARCH_RESULT in step:
+                                    truncated_prev_reasoning += step + '\n\n'
+                                else: # 如果步骤超过 5 个，进行截断。保留第一个步骤、最后四个步骤以及包含搜索查询或搜索结果的步骤。如果中间的步骤不重要，则用 ... 来表示省略
+                                    if truncated_prev_reasoning[-len('\n\n...\n\n'):] != '\n\n...\n\n':
+                                        truncated_prev_reasoning += '...\n\n'
+                        truncated_prev_reasoning = truncated_prev_reasoning.strip('\n')
+
+                        # Collect parameters for batch processing
+                        batch_relevant_info.append(relevant_info) # 搜索出来的信息
+                        batch_original_questions.append(seq['item']['Question']) # 原始问题
+                        batch_prev_reasonings.append(truncated_prev_reasoning) # 之前的推理步骤
+                        batch_search_queries.append(search_query) # 搜索查询
+                        batch_sequences.append(seq)
+
+                        # Update search count and executed queries
+                        seq['search_count'] += 1 # 更新搜索计数
+                        seq['executed_search_queries'].add(search_query) # 将已执行的搜索查询添加到集合中
+
+                    elif seq['search_count'] >= MAX_SEARCH_LIMIT: # 如果搜索次数达到或超过该限制，则返回一条消息，通知该查询无法再进行
+                        limit_message = f"\n{BEGIN_SEARCH_RESULT}\nThe maximum search limit is exceeded. You are not allowed to search.\n{END_SEARCH_RESULT}\n"
+                        seq['prompt'] += limit_message
+                        seq['output'] += limit_message
+                        seq['history'].append(limit_message)
+                        seq["all_info"].append({f"turn_{turn}_search_limited": limit_message})
+                        print(f"Search limit reached for query: \"{search_query}\"")
+
+                    elif search_query in seq['executed_search_queries']: # 如果当前查询已经执行过，则返回一个消息，提示用户查询已重复，并引导其查看之前的结果
+                        limit_message = f"\n{BEGIN_SEARCH_RESULT}\nYou have searched this query. Please refer to previous results.\n{END_SEARCH_RESULT}\n"
+                        seq['prompt'] += limit_message
+                        seq['output'] += limit_message
+                        seq['history'].append(limit_message)
+                        seq["all_info"].append({f"turn_{turn}_search_limited": limit_message})
+                        print(f"Repeated search for query: \"{search_query}\"")
+                        
+
+                else:  # 如果当前序���没有需要执行的搜索查询，则将该序列标记为完成，并打印提示信息
+                    # If no search query needs to be executed, mark the sequence as finished
+                    seq['finished'] = True
+                    print("Sequence marked as complete.")
+
+            print(f"all_urls_to_fetch len: {len(all_urls_to_fetch)}, url_cache len: {len(url_cache)}")
+            print(f"all_urls_to_fetch: {all_urls_to_fetch}")
+            # Batch fetch all URLs at once to optimize speed
+            
+            if all_urls_to_fetch:
+                print(f"Fetching {len(all_urls_to_fetch)} URLs...")
+                try:
+                    fetched_contents = fetch_page_content( # 一次性获取所有 URL 的中搜索出来的内容
+                        list(all_urls_to_fetch),
+                        use_jina=use_jina,
+                        jina_api_key=jina_api_key,
+                        # snippets=url_snippets  # Do not pass snippets when updating url_cache directly
+                    )
+                    print(f"Fetched {len(fetched_contents)} URLs successfully.")
+                except Exception as e:
+                    print(f"Error during batch URL fetching: {e}")
+                    fetched_contents = {url: f"Error fetching URL: {e}" for url in all_urls_to_fetch}
+                # Update cache with fetched contents
+                for url, content in fetched_contents.items(): # 将获取的内容添加到 url_cache 中
+                    url_cache[url] = content
+
+            # After fetching, prepare formatted documents for batch processing
+            for relevant_info in batch_relevant_info:
+                formatted_documents = "" # 初始化一个空字符串 formatted_documents，用于拼接本次要处理的所有网页信息。后面会将其添加到 batch_documents 列表中
+                
+                doc_str_list = []
+                for i, doc_info in enumerate(relevant_info):
+                    url = doc_info['url']
+                    raw_context = url_cache.get(url, "") # 获取 url 对应的内容
+                    doc_info['snippet'] = doc_info['snippet'].replace('<b>','').replace('</b>','')            
+                    success, filtered_context = extract_snippet_with_context(raw_context, doc_info['snippet'], context_chars=max_doc_len)
+                    if success:
+                        context = filtered_context
+                    else: # 否则就取 raw_context 的前 max_doc_len * 2 个字符，作为一个有限的片段，避免过长导致后续处理负担
+                        context = raw_context[:max_doc_len*2]
+
+                    doc_info['context'] = context
+                    # formatted_documents += f"**Web Page {i + 1}:**\n"
+                    # formatted_documents += json.dumps(doc_info, ensure_ascii=False, indent=2) + "\n"
+                    single_doc = json.dump(doc_info, ensure_ascii=False, indent=2)
+                    doc_str_list.append(single_doc)
+                batch_documents.append(doc_str_list) # 将本组搜索结果的所有信息（拼接成的字符串 formatted_documents）添加到 batch_documents 列表中
+
+            # After fetching, prepare for batch processing if there are any
+            if batch_sequences:
+                print(f"Batch processing {len(batch_sequences)} sequences with generate_webpage_to_reasonchain_batch...")
+                webpage_analyses = generate_webpage_to_reasonchain_batch( # 根据前面处理好的结果，生成新的推理
+                    original_questions=batch_original_questions,
+                    prev_reasonings=batch_prev_reasonings,
+                    search_queries=batch_search_queries,
+                    documents=batch_documents,
+                    dataset_name=dataset_name,
+                    batch_output_records=batch_output_records,  # Pass the collection list
+                    max_tokens=max_tokens,
+                )
+                print("Batch generation completed, assigning outputs to sequences...")
+
+                for seq, analysis,doc in zip(batch_sequences, webpage_analyses, batch_documents): # 遍历批处理返回的 webpage_analyses，将处理结果与相应的序列 seq 进行一一对应
+                    if isinstance(analysis, str): # 判断 analysis 是否是纯字符串。如果是字符串，说明直接可以追加到序列的文本中
+                        append_text = f"\n\n{BEGIN_SEARCH_RESULT}{analysis}{END_SEARCH_RESULT}\n\n" # 封装处理结果，添加到序列的历史记录、提示和输出中
+                        seq['prompt'] += append_text
+                        seq['output'] += append_text
+                        seq['history'].append(append_text) # 存的是每一次的webpage_analyses
+                        seq['all_info'].extend([{f"turn_{turn}_search": doc}, {f"turn_{turn}_webpage_analyses": analysis}])
+                    else: # 如果 analysis 不是纯字符串，那么可能是一种特殊的结构，比如表示需要替换推��步骤的 dict 或其他格式
+                        append_text = replace_recent_steps(seq['output'], analysis)
+                        seq['prompt'] += append_text
+                        seq['output'] += append_text
+                        seq['history'].append(append_text)
+                        seq['all_info'].extend([{f"turn_{turn}_search": doc}, {f"turn_{turn}_webpage_analyses": analysis}])
+
+        # Check if all sequences are finished
+        # 保存active_sequences
+        active_sequences_part = [{ # 记录每个问题的搜索历史
+            'item': ele["item"],
+            'prompt': ele['prompt'],
+            'output': ele["output"],
+            'finished': ele["finished"], # 一开始均为未完成
+            'history':ele["history"],
+            'search_count': ele["search_count"],
+            'all_info': ele['all_info']
+        } for ele in active_sequences]
+        with open(os.path.join(output_dir, f"turn_{turn}.json"), 'w', encoding='utf-8') as f:
+            json.dump(active_sequences_part, f, ensure_ascii=False, indent=2)
+        unfinished = [seq for seq in active_sequences if not seq['finished']] # 是否结束是基于模型是否生成了新的搜索
+        if not unfinished:
+            break
+        else:
+            if turn >= MAX_TURN:
+                print(f"Maximum number of turns ({MAX_TURN}) reached, stopping.")
+                break
+
+    total_time = time.time() - start_time
+    print(f"Total time taken: {total_time} seconds")
+
+    # ---------------------- Save Batch Output Records to JSON File ----------------------
+    # Define output JSON file path
+    t = time.localtime()
+    batch_output_file = os.path.join(output_dir, f'{split}.{t.tm_mon}.{t.tm_mday},{t.tm_hour}:{t.tm_min}.info_extract.json')
+
+    # Save batch_output_records to JSON file
+    with open(batch_output_file, 'w', encoding='utf-8') as f: # 这里存的是webpage推理时的输入和输出和提取后的信息
+        json.dump(batch_output_records, f, ensure_ascii=False, indent=2)
+
+    print(f"Batch outputs saved to {batch_output_file}")
+
+    # Prepare output list for evaluation
+    output_list = [seq['output'] for seq in active_sequences]
+
+    # Run evaluation
+    run_evaluation(filtered_data, input_list, output_list, dataset_name, output_dir, total_time, split)
+
+    # 评测has answer信息
+    turn_files = os.listdir(output_dir)
+    turn_files = [file for file in turn_files if file.startswith("turn_")]
+    max_turn_file = max(turn_files, key=lambda x: int(re.search(r'turn_(\d+)', x).group(1)))
+
+    max_turn_file_path = os.path.join(output_dir, max_turn_file)
+    print(f"max_turn_file_path: {max_turn_file_path}")
+    add_eval(model_path, max_turn_file_path)
+
+    # ---------------------- Update Search and URL Cache ----------------------
+    print('Updating Search and URL Cache...')
+    # Load existing caches or initialize empty dictionaries
+    if os.path.exists(search_cache_path):
+        with open(search_cache_path, 'r', encoding='utf-8') as f:
+            search_cache_new = json.load(f)
+    else:
+        search_cache_new = {}
+
+    if os.path.exists(url_cache_path):
+        with open(url_cache_path, 'r', encoding='utf-8') as f:
+            url_cache_new = json.load(f)
+    else:
+        url_cache_new = {}
+
+    search_cache.update(search_cache_new)
+    url_cache.update(url_cache_new)
+
+    save_caches()
+
+    print("Process completed.")
+
+if __name__ == "__main__":
+    main()

deep_search/data_syn/select_data.py

@@ -0,0 +1,233 @@
+import json
+from collections import defaultdict,OrderedDict, Counter
+from tqdm import tqdm
+import random
+import matplotlib.pyplot as plt
+
+
+
+def analyze_and_visualize_sources(final_dataset, output_image_path="source_distribution_selected_data.png", output_json_path="source_counts.json"):
+    """
+    统计 final_dataset 中 source 的分布，并绘制饼图和保存为 JSON 文件。
+    
+    :param final_dataset: 输入的数据列表，每个元素是一个字典
+    :param output_image_path: 饼图保存路径（默认为 "source_distribution.png"）
+    :param output_json_path: JSON 文件保存路径（默认为 "source_counts.json"）
+    """
+    # 提取所有包含 source 的项
+    sources = [item["source"] for item in final_dataset if "source" in item]
+    
+    # 统计 source 的分布
+    source_counts = Counter(sources)
+    
+    # 将统计结果保存为 JSON 文件
+    with open(output_json_path, "w", encoding="utf-8") as f:
+        json.dump(source_counts, f, ensure_ascii=False, indent=4)
+    print(f"Source 分布已保存到 {output_json_path}")
+    
+    # 绘制饼图
+    labels = list(source_counts.keys())
+    counts = list(source_counts.values())
+    
+    plt.figure(figsize=(8, 8))
+    plt.pie(counts, labels=labels, autopct='%1.1f%%', startangle=140)
+    plt.title("Source Distribution")
+    plt.axis('equal')  # 确保饼图为正圆
+    
+    # 保存饼图为图片文件
+    plt.savefig(output_image_path)
+    plt.close()
+    print(f"Source 分布饼图已保存到 {output_image_path}")
+
+# 读取原始数据
+with open('/share/project/sunshuang/deep_search/data_for_rl/tagged_domain_keypoints/merged_data_tagged_domain_keypoints_keywords_count.json') as f:
+    data = json.load(f)
+
+for idx, item in enumerate(data): # 增加Question key
+    if "question" in item:
+        item["Question"] = item["question"]
+    
+
+# step 1 预处理 Domain 分类
+# 统计原始 domain 分布
+print("step 1: preprocss domain data")
+domain_counter = defaultdict(int)
+for item in data:
+    domain = item["domain_keypoints"]["domain"]
+    domain_counter[domain] += 1
+
+# 确定有效 domain（>=100 条）
+# valid_domains = {d for d, cnt in domain_counter.items() if cnt >= 100}
+# other_domains = [d for d, cnt in domain_counter.items() if cnt < 100]
+valid_domains = {d for d, cnt in domain_counter.items() if cnt >= 500}
+other_domains = [d for d, cnt in domain_counter.items() if cnt < 500]
+print(f"valid_domains: {len(valid_domains)}")
+print(f"other_domains: {len(other_domains)}")
+
+# 重构数据集（合并 small domains 到 other）
+processed_data = []
+for item in data:
+    original_domain = item["domain_keypoints"]["domain"]
+    if original_domain in valid_domains:
+        new_domain = original_domain
+    else:
+        new_domain = "other"
+    
+    new_item = {
+        **item,
+        "domain_keypoints": {
+            **item["domain_keypoints"],
+            "domain": new_domain
+        }
+    }
+    processed_data.append(new_item)
+
+
+
+# step 2 分配样本配额
+# 计算新 domain 分布
+print("step 2: calculate domain distribution")
+new_domain_counts = defaultdict(int)
+for item in processed_data:
+    domain = item["domain_keypoints"]["domain"]
+    new_domain_counts[domain] += 1
+
+
+# 确定最终 domain 列表
+final_domains = list(valid_domains)
+if new_domain_counts.get("other", 0) > 0:
+    final_domains.append("other")
+
+print(f"new_domain_counts: {len(new_domain_counts)}")
+print(f"final_domains: {len(final_domains)}")
+
+# 分配 2000 条样本配额
+total_samples = 3000
+num_domains = len(final_domains)
+base_quota = total_samples // num_domains
+remainder = total_samples % num_domains
+print(f"base_quota: {base_quota}")
+print(f"remainder: {remainder}")
+
+domain_quotas = {}
+# cnttt = 0
+for idx, domain in enumerate(final_domains):
+    domain_quotas[domain] = base_quota + (1 if idx < remainder else 0)
+    # cnttt += domain_quotas[domain]
+
+# print(f"cnttt: {cnttt}")
+
+
+# step 3 统计 Keypoints 分布
+print("step 3 calculating keypoints distribution")
+domain_keypoint_dist = defaultdict(lambda: defaultdict(int))
+
+for item in tqdm(processed_data, desc="Calculating keypoints distribution"):
+    domain = item["domain_keypoints"]["domain"]
+    keypoints = item["domain_keypoints"]["key_points"]
+    for kp in keypoints:
+        domain_keypoint_dist[domain][kp] += 1
+
+# domain内部按照key points降序排列
+for domain, keypoint_counts in domain_keypoint_dist.items():
+    sorted_keypoints = sorted(keypoint_counts.items(), key=lambda x: x[1], reverse=True)
+    domain_keypoint_dist[domain] = OrderedDict(sorted_keypoints)
+
+
+# 保存分布文件
+with open("domain_keypoints_distribution_without_remove_dup.json", "w") as f:
+    json.dump(domain_keypoint_dist, f, indent=4,ensure_ascii=False)
+
+
+# step 4 分层抽样与去重
+print("step 4: select data")
+final_dataset = []
+question_set = set()
+
+real_selected_cnt = 0
+for domain in tqdm(final_domains, desc="Processing domains"):
+    # 获取当前 domain ��所有数据
+    domain_data = [item for item in processed_data if item["domain_keypoints"]["domain"] == domain]
+    print(f"-------------- process {domain}")
+    print(f"data: {len(domain_data)}")
+    # 计算 keypoint 配额
+    keypoints = list(domain_keypoint_dist[domain].keys())
+    # random.shuffle(keypoints) # 随机打乱顺序
+
+    print(f"key points {len(keypoints)}")
+    print(f"domain_quotas: {domain_quotas[domain]}")
+    # kp_base = domain_quotas[domain] // len(keypoints)
+    # print(f"kp_base: {kp_base}")
+    # kp_remainder = domain_quotas[domain] % len(keypoints)
+    # print(f"kp_remainder: {kp_remainder}")
+    # 因为domain_quotas远远小于keypoint的数目，改为将所有的数据按照total降序排序，然后从上往下选，重复上面的操作，直到选完
+    
+    sorted_domain_data = sorted(domain_data, key=lambda x: x["keywords_count"]["total"], reverse=True)
+
+    selected_cnt = 0
+    cycle_cnt = 0
+    selected_index = [] # 已选择的index
+    while selected_cnt < domain_quotas[domain]: # 当前选择的数目小于domain_quotas[domain]
+        cycle_cnt += 1
+        # print(f"cycle_cnt: {cycle_cnt}")
+        selected_keypoints = set() # 每次循环都清零
+        for idx, item in enumerate(sorted_domain_data): # 遍历所有的数据
+            if selected_cnt >= domain_quotas[domain]:
+                break
+            if idx not in selected_index: # 当前数据未被选择
+                dup_keypoints = False
+                for key_point in item["domain_keypoints"]["key_points"]: # 确保数据的key point不包含在已经选择的key point里
+                    if key_point in selected_keypoints:
+                        dup_keypoints = True
+                        break
+                if dup_keypoints: # 重复key points
+                    continue
+                if item["Question"] not in question_set:
+                    final_dataset.append(item)
+                    question_set.add(item["Question"])
+                    selected_keypoints.update(item["domain_keypoints"]["key_points"])
+                    selected_index.append(idx)
+                    selected_cnt += 1
+    real_selected_cnt  += selected_cnt
+    print(f"cycle_cnt: {cycle_cnt}")
+
+print(f"real_selected_cnt: {real_selected_cnt}")              
+
+
+    # # 按 keypoint 分层抽样
+    # selected = []
+    # for kp_idx, kp in enumerate(keypoints):
+    #     # 分配配额
+    #     kp_quota = kp_base + (1 if kp_idx < kp_remainder else 0)
+    #     # print(f"kp_quota: {kp_quota}")
+    #     # 获取包含该 keypoint 的问题
+    #     kp_items = [item for item in domain_data if kp in item["domain_keypoints"]["key_points"]]
+        
+    #     # 按 total 降序排序并选择
+    #     sorted_items = sorted(kp_items, key=lambda x: x["keywords_count"]["total"], reverse=True)
+    #     selected.extend(sorted_items[:kp_quota])
+    
+    # # 去重并保留顺序，这一步导致选不够
+    # dup_cnt = 0
+    # for item in selected:
+    #     q = item["Question"]
+    #     if q not in question_set:
+    #         final_dataset.append(item)
+    #         question_set.add(q)
+    #     else:
+    #         dup_cnt += 1
+    # print(f"dup_cnt: {dup_cnt}")
+
+# 精确控制最终数量
+
+
+# final_dataset = sorted(final_dataset, key=lambda x: x["idx"])
+print(f"final_dataset len: {len(final_dataset)}")
+# final_dataset = final_dataset
+
+analyze_and_visualize_sources(final_dataset)
+
+
+# step 5 保存结果
+with open("final_selected_dataset.json", "w") as f:
+    json.dump(final_dataset, f, indent=4, ensure_ascii=False)

deep_search/data_syn/select_remain_data.py

@@ -0,0 +1,58 @@
+# run_search_o1.py
+import os
+import json
+import time
+import re
+from tqdm import tqdm
+import numpy as np
+import torch
+import string
+from typing import Optional, Tuple, List, Dict
+import argparse
+
+
+
+def load_json(file_path):
+    with open(file_path, "r", encoding="utf-8") as file:
+        data = json.load(file)
+    print(f"Loaded {len(data)} items from {file_path}")
+    return data
+
+def save_json(data, file_path):
+    with open(file_path, "w", encoding="utf-8") as file:
+        json.dump(data, file, ensure_ascii=False, indent=4)
+    print(f"Saved {len(data)} items to {file_path}")
+
+
+file_1 = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/final_dataset_new/final_selected_dataset.json"
+file_2 = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data/final_selected_dataset.json"
+output_file = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data.json"
+data_1 = load_json(file_1)
+data_2 = load_json(file_2)
+
+id_1 = [item["idx"] for item in data_1]
+
+in_id_1 = 0
+not_in_id_1 = 0
+
+in_id_1_id = []
+
+dup_id = []
+
+remain_data = []
+for item in data_2:
+    if item["idx"] not in id_1:
+        remain_data.append(item)
+        not_in_id_1 += 1
+    else:
+        if item["idx"] in in_id_1_id:
+            dup_id.append(item["idx"])
+        else:
+            
+            in_id_1_id.append(item["idx"])
+        in_id_1 += 1
+
+
+print(f"in_id_1: {in_id_1}, not_in_id_1: {not_in_id_1}")
+print(f"dup_id: {len(dup_id)}")
+# save_json(remain_data, output_file)

deep_search/data_syn/source_stats.py

@@ -0,0 +1,57 @@
+import json
+from collections import defaultdict,OrderedDict, Counter
+from tqdm import tqdm
+import random
+import matplotlib.pyplot as plt
+
+
+
+def save_to_json(data, filename):
+    """保存数据到 JSON 文件"""
+    with open(filename, 'w', encoding='utf-8') as f:
+        json.dump(data, f, ensure_ascii=False, indent=4)
+    print(f"Saved to {filename}, data length: {len(data)}")
+
+def load_json(file_path):
+    """从 JSON 文件加载数据"""
+    with open(file_path, "r", encoding="utf-8") as f:
+        data = json.load(f)
+    print(f"Loaded from {file_path}, data length: {len(data)}")
+    return data
+def analyze_and_visualize_sources(final_dataset, output_image_path="source_distribution_selected_data.png", output_json_path="source_counts.json"):
+    """
+    统计 final_dataset 中 source 的分布，并绘制饼图和保存为 JSON 文件。
+    
+    :param final_dataset: 输入的数据列表，每个元素是一个字典
+    :param output_image_path: 饼图保存路径（默认为 "source_distribution.png"）
+    :param output_json_path: JSON 文件保存路径（默认为 "source_counts.json"）
+    """
+    # 提取所有包含 source 的项
+    sources = [item["source"] for item in final_dataset if "source" in item]
+    
+    # 统计 source 的分布
+    source_counts = Counter(sources)
+    
+    # 将统计结果保存为 JSON 文件
+    with open(output_json_path, "w", encoding="utf-8") as f:
+        json.dump(source_counts, f, ensure_ascii=False, indent=4)
+    print(f"Source 分布已保存到 {output_json_path}")
+    
+    # 绘制饼图
+    labels = list(source_counts.keys())
+    counts = list(source_counts.values())
+    
+    plt.figure(figsize=(8, 8))
+    plt.pie(counts, labels=labels, autopct='%1.1f%%', startangle=140)
+    plt.title("Source Distribution")
+    plt.axis('equal')  # 确保饼图为正圆
+    
+    # 保存饼图为图片文件
+    plt.savefig(output_image_path)
+    plt.close()
+    print(f"Source 分布饼图已保存到 {output_image_path}")
+
+
+input_file = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/mixed_data_all.json"
+data = load_json(input_file)
+analyze_and_visualize_sources(data)

deep_search/data_syn/test.py

@@ -0,0 +1,69 @@
+
+
+# user_prompt = f"""You are an advanced semantic analyzer. For the given question, perform the following tasks step-by-step:
+
+# 1. **Domain Identification**:
+# - Determine the broad subject category (domain) this question belongs to.
+# - Examples: film, history, biology, geography, politics, technology, etc (or any other suitable domain)
+
+# 2. **Key Point Extraction**:
+# - Identify 2-4 core semantic components that are crucial for answering
+# - Include:
+# • Key entities (e.g., films, people, locations)
+# • Critical attributes (e.g., age, duration, population)
+# • Core relationships (e.g., comparison, causality)
+# • Measurement dimensions (e.g., time, quantity)
+# - Exclude filler words and non-essential descriptors
+
+# **Output Requirements**:
+# - Use JSON format: {{"domain": "...", "key_points": [...]}}
+# - Keep key_points concise (1-2 words each)
+# - Use lowercase for all outputs
+# - Separate multiple key_points with commas
+
+# **Examples**:
+# Question: "Which film whose director is younger, Charge It To Me or Danger: Diabolik?"
+# Output: {{"domain": "film", "key_points": ["director", "age"]}}
+
+# **Now process this question:**
+# {{Question}}"""
+
+# print(user_prompt.replace('{Question}', 'Which film whose director is younger, Charge It To Me or Danger: Diabolik?'))
+
+
+
+# import json
+
+# txt = "{'domain': 'film', 'key_points': ['directors', 'country', 'same']}"
+# print(json.loads(txt))
+
+
+# 使用ast.literal_eval解析Python风格字符串（需import ast）
+# import ast
+# py_style_txt = "{'domain': 'film', 'key_points': ['directors', 'country', 'same']}"
+# print(ast.literal_eval(py_style_txt))  # 输出解析后的字典
+import re
+def extract_last_braced_content(s):
+    """
+    提取字符串中被 {} 包裹的内容，如果有多个则返回最后一个。
+    
+    :param s: 输入字符串
+    :return: 最后一个被 {} 包裹的内容，如果没有则返回 None
+    """
+    # 使用正则表达式匹配所有被 {} 包裹的内容
+    # matches = re.findall(r'\{(.*?)\}', s)
+    matches = re.findall(r'\{.*?\}', s)
+    
+    # 如果有匹配的内容，返回最后一个；否则返回 None
+    return matches[-1] if matches else None
+
+# import json
+
+# input_file_path = "/share/project/sunshuang/deep_search/data_syn/data/mixed_data/splits/tagged_domain_keypoints/final_selected_dataset.json"
+# with open(input_file_path, 'r') as f:
+#     data = json.load(f)
+
+# print(len(data))
+
+text = '{"domain": "development", "key_points": ["human development index", "adopted", "time", "employer"]}  \n*Note: The key point "human development index" exceeds the 1-2 word limit. To comply strictly, "hdi" (abbreviation) could be used instead for conciseness, but the original term is more precise. Adjusting for the requirement:*  \n{"domain": "development", "key_points": ["hdi", "adopted", "time", "employer"]}'
+print(extract_last_braced_content(text))

deep_search/data_syn/testttt.py

@@ -0,0 +1,2 @@
+import matplotlib
+print(matplotlib.matplotlib_fname())

deep_search/search_o1/google_search.py

@@ -0,0 +1,246 @@
+import os
+import json
+import requests
+from requests.exceptions import Timeout
+from bs4 import BeautifulSoup
+from tqdm import tqdm
+import time
+import concurrent
+from concurrent.futures import ThreadPoolExecutor
+# import pdfplumber
+from io import BytesIO
+import re
+import string
+from typing import Optional, Tuple
+from nltk.tokenize import sent_tokenize
+
+# os.environ['http_proxy'] = 'http://127.0.0.1:7890'
+# os.environ['https_proxy'] = 'http://127.0.0.1:7890'
+
+
+# ----------------------- Custom Headers -----------------------
+headers = {
+    'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) '
+                  'AppleWebKit/537.36 (KHTML, like Gecko) '
+                  'Chrome/58.0.3029.110 Safari/537.36',
+    'Referer': 'https://www.google.com/',
+    'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8',
+    'Accept-Language': 'en-US,en;q=0.5',
+    'Connection': 'keep-alive',
+    'Upgrade-Insecure-Requests': '1'
+}
+
+# Initialize session
+session = requests.Session()
+session.headers.update(headers)
+
+proxies = {
+    "http": "http://127.0.0.1:7880",
+    "https": "http://127.0.0.1:7880"
+}
+
+# def bing_web_search(query, subscription_key, endpoint, market='en-US', language='en', timeout=20):
+#     """
+#     Perform a search using the Bing Web Search API with a set timeout.
+
+#     Args:
+#         query (str): Search query.
+#         subscription_key (str): Subscription key for the Bing Search API.
+#         endpoint (str): Endpoint for the Bing Search API.
+#         market (str): Market, e.g., "en-US" or "zh-CN".
+#         language (str): Language of the results, e.g., "en".
+#         timeout (int or float or tuple): Request timeout in seconds.
+#                                          Can be a float representing the total timeout,
+#                                          or a tuple (connect timeout, read timeout).
+
+#     Returns:
+#         dict: JSON response of the search results. Returns None or raises an exception if the request times out.
+#     函数的目标是使用 Bing Web Search API 执行搜索，并返回 JSON 格式的结果。
+#     如果请求超时或出现其他问题，返回空字典（{}）或抛出异常
+#     """
+#     # query = f"{query} -site:en.wikipedia.org"  # Add site:wikipedia.org to the query to restrict results to Wikipedia
+#     print(query)
+#     payload = json.dumps({
+#         "q": query,  # 设置查询内容
+#         "mkt": market,  # 设置市场
+#         "setLang": language,  # 设置语言
+#         "textDecorations": True,  # 启用文本装饰
+#         "textFormat": "HTML"  # 设置文本格式
+#     })
+
+#     headers = {
+#         'X-API-KEY': subscription_key,
+#         'Content-Type': 'application/json'
+#     }
+
+#     try:
+#         # 发送POST请求
+#         response = requests.request("POST", endpoint, headers=headers, data=payload)
+#         response.raise_for_status()  # Raise exception if the request failed 检查响应的状态码。如果返回的状态码是 4xx 或 5xx（表示客户端或服务器错误），它将引发 requests.exceptions.HTTPError 异常
+#         search_results = response.json() #
+#         return search_results
+#     except Timeout:
+#         print(f"Bing Web Search request timed out ({timeout} seconds) for query: {query}")
+#         return {}  # Or you can choose to raise an exception
+#     except requests.exceptions.RequestException as e:
+#         print(f"Error occurred during Bing Web Search request: {e}")
+#         return {}
+
+# def bing_web_search(query, subscription_key, endpoint, market='en-US', language='en', timeout=2000):
+#     """
+#     Perform a search using the Bing Web Search API with a set timeout.
+
+#     Args:
+#         query (str): Search query.
+#         subscription_key (str): Subscription key for the Bing Search API.
+#         endpoint (str): Endpoint for the Bing Search API.
+#         market (str): Market, e.g., "en-US" or "zh-CN".
+#         language (str): Language of the results, e.g., "en".
+#         timeout (int or float or tuple): Request timeout in seconds.
+#                                          Can be a float representing the total timeout,
+#                                          or a tuple (connect timeout, read timeout).
+
+#     Returns:
+#         dict: JSON response of the search results. Returns None or raises an exception if the request times out.
+#     函数的目标是使用 Bing Web Search API 执行搜索，并返回 JSON 格式的结果。
+#     如果请求超时或出现其他问题，返回空字典（{}）或抛出异常
+#     """
+#     payload = json.dumps({
+#         "q": query,  # 设置查询内容
+#         "num": 11,
+#         "mkt": market,  # 设置市场
+#         "setLang": language,  # 设置语言
+#         "textDecorations": True,  # 启用文本装饰
+#         "textFormat": "HTML",  # 设置文本格式
+#     })
+
+#     headers = {
+#         'X-API-KEY': subscription_key,
+#         'Content-Type': 'application/json'
+#     }
+#     error_cnt = 0
+#     while True:
+#         if error_cnt == 20:
+#             print(f"qery: {query} has tried {error_cnt} times without success, just skip it.")
+#             break
+#         try:
+#             # 发送POST请求
+#             response = requests.request("POST", endpoint, headers=headers, data=payload,  proxies=proxies,timeout=timeout)
+#             response.raise_for_status()  # Raise exception if the request failed 检查响应的状态码。如果返回的状态码是 4xx 或 5xx（表示客户端或服务器错误），它将引发 requests.exceptions.HTTPError 异常
+#             search_results = response.json() #
+#             return search_results
+#         except Timeout:
+#             error_cnt += 1
+#             print(f"error_cnt: {error_cnt}, Bing Web Search request timed out ({timeout} seconds) for query: {query}")
+#             time.sleep(5)
+#             # return {}  # Or you can choose to raise an exception
+#         except requests.exceptions.RequestException as e:
+#             error_cnt += 1
+#             print(f"error_cnt: {error_cnt}, Error occurred during Bing Web Search request: {e}, payload: {payload}")
+#             time.sleep(5)
+#             return {}
+
+
+
+
+def bing_web_search(query, subscription_key, endpoint, market='en-US', language='en', exclude_urls=[],timeout=2000):
+    """
+    Perform a search using the Bing Web Search API with a set timeout.
+
+    Args:
+        query (str): Search query.
+        subscription_key (str): Subscription key for the Bing Search API.
+        endpoint (str): Endpoint for the Bing Search API.
+        market (str): Market, e.g., "en-US" or "zh-CN".
+        language (str): Language of the results, e.g., "en".
+        timeout (int or float or tuple): Request timeout in seconds.
+                                         Can be a float representing the total timeout,
+                                         or a tuple (connect timeout, read timeout).
+
+    Returns:
+        dict: JSON response of the search results. Returns None or raises an exception if the request times out.
+    函数的目标是使用 Bing Web Search API 执行搜索，并返回 JSON 格式的结果。
+    如果请求超时或出现其他问题，返回空字典（{}）或抛出异常
+    """
+    exclude_urls = ['https://greysanatomy.fandom.com/wiki/Season_14_(Grey%27s_Anatomy)', "https://abc.com/news/9fce69b2-3b93-40f3-9cd3-d75b1d90bb05/category/738075"]
+    # if exclude_urls:
+    #     for site in exclude_urls:
+    #         query += f" -site:{site}"
+
+    payload = json.dumps({
+        "q": query,  # 设置查询内容
+        "num": 11,
+        "mkt": market,  # 设置市场
+        "setLang": language,  # 设置语言
+        "textDecorations": True,  # 启用文本装饰
+        "textFormat": "HTML"  # 设置文本格式
+    })
+
+    headers = {
+        'X-API-KEY': subscription_key,
+        'Content-Type': 'application/json'
+    }
+    error_cnt = 0
+    while True:
+        if error_cnt == 20:
+            print(f"qery: {query} has tried {error_cnt} times without success, just skip it.")
+            break
+        try:
+            # 发送POST请求
+            response = requests.request("POST", endpoint, headers=headers, data=payload, timeout=timeout)
+            response.raise_for_status()  # Raise exception if the request failed 检查响应的状态码。如果返回的状态码是 4xx 或 5xx（表示客户端或服务器错误），它将引发 requests.exceptions.HTTPError 异常
+            search_results = response.json() #
+            return search_results
+        except Timeout:
+            error_cnt += 1
+            print(f"error_cnt: {error_cnt}, Bing Web Search request timed out ({timeout} seconds) for query: {query}")
+            time.sleep(5)
+            # return {}  # Or you can choose to raise an exception
+        except requests.exceptions.RequestException as e:
+            error_cnt += 1
+            print(f"error_cnt: {error_cnt}, Error occurred during Bing Web Search request: {e}, payload: {payload}")
+            time.sleep(5)
+            # return {}
+
+# url = "https://google.serper.dev/search"
+
+# payload = json.dumps({
+#   "q": "apple inc"
+# })
+# headers = {
+#   'X-API-KEY': 'cb0d28279a826d7e5cf22d71f683c77ffd4ba27d',
+#   'Content-Type': 'application/json'
+# }
+
+# response = requests.request("POST", url, headers=headers, data=payload)
+
+# print(response.text)
+
+# result = google_web_search("apple inc", 'cb0d28279a826d7e5cf22d71f683c77ffd4ba27d')
+# print(result)
+
+
+def extract_text_from_url(url: str) -> Optional[str]:
+    response = session.get(url, timeout=20)
+    response.raise_for_status()
+    content_type = response.headers.get('Content-Type', '').lower()
+    try:
+        soup = BeautifulSoup(response.content, 'lxml')
+        # print(soup)
+    except Exception:
+        print("lxml parser not found or failed, falling back to html.parser")
+        soup = BeautifulSoup(response.text, 'html.parser')
+    text = soup.get_text(separator=' ', strip=True)
+    return text
+def save_json(data, file_path):
+    with open(file_path, "w", encoding="utf-8") as file:
+        json.dump(data, file, ensure_ascii=False, indent=4)
+    print(f"Saved {len(data)} items to {file_path}")
+
+result = bing_web_search("when does season 14 of grey's anatomy come out", 'cb0d28279a826d7e5cf22d71f683c77ffd4ba27d', 'https://google.serper.dev/search')
+
+print(result)
+save_json(result, "/share/project/sunshuang/deep_search/search_o1/search_ex.json")
+
+# text = extract_text_from_url("https://greysanatomy.fandom.com/wiki/Season_14_(Grey%27s_Anatomy)")
+# print(text)

deep_search/search_o1/infer_wo_search.sh

@@ -0,0 +1,59 @@
+
+export CUDA_VISIBLE_DEVICES=4,5
+python scripts/inference_wo_search.py \
+    --model_name QwQ-32B \
+    --model_path /share/project/zhipengchen/model/QwQ-32B \
+    --dataset_name eval \
+    --input_file /share/project/sunshuang/deep_search/search_o1/data/eval/eval.json 
+
+
+python scripts/inference_wo_search.py \
+    --model_name DeepSeek-R1-Distill-Qwen-32B \
+    --model_path /share/project/zhipengchen/model/DeepSeek-R1-Distill-Qwen-32B \
+    --dataset_name eval \
+    --input_file /share/project/sunshuang/deep_search/search_o1/data/eval/eval.json 
+
+
+python scripts/inference_wo_search.py \
+    --model_name Qwen2.5-7B-Instruct \
+    --model_path /share/project/zhipengchen/model/Qwen2.5-7B-Instruct \
+    --dataset_name eval \
+    --input_file /share/project/sunshuang/deep_search/search_o1/data/eval/eval.json 
+
+python scripts/inference_wo_search.py \
+    --model_name Qwen2.5-32B-Instruct \
+    --model_path /share/project/zhipengchen/model/Qwen2.5-32B-Instruct \
+    --dataset_name eval \
+    --input_file /share/project/sunshuang/deep_search/search_o1/data/eval/eval.json 
+
+# ############################
+# # musique
+
+
+# export CUDA_VISIBLE_DEVICES=6.7
+# python scripts/inference_wo_search.py \
+#     --model_name qwq \
+#     --model_path /share/project/zhipengchen/model/QwQ-32B \
+#     --dataset_name musique_syn \
+#     --input_file /share/project/sunshuang/deep_search/search_o1/data/test/musique_syn.json 
+
+
+# python scripts/inference_wo_search.py \
+#     --model_name DeepSeek-R1-Distill-Qwen-32B \
+#     --model_path /share/project/zhipengchen/model/DeepSeek-R1-Distill-Qwen-32B \
+#     --dataset_name musique_syn \
+#     --input_file /share/project/sunshuang/deep_search/search_o1/data/test/musique_syn.json
+
+
+# python scripts/inference_wo_search.py \
+#     --model_name Qwen2.5-7B-Instruct \
+#     --model_path /share/project/zhipengchen/model/Qwen2.5-7B-Instruct \
+#     --dataset_name musique_syn \
+#     --input_file /share/project/sunshuang/deep_search/search_o1/data/test/musique_syn.json
+
+# python scripts/inference_wo_search.py \
+#     --model_name Qwen2.5-32B-Instruct \
+#     --model_path /share/project/zhipengchen/model/Qwen2.5-32B-Instruct \
+#     --dataset_name musique_syn \
+#     --input_file /share/project/sunshuang/deep_search/search_o1/data/test/musique_syn.json
+

deep_search/search_o1/llm_as_judge_source.sh

@@ -0,0 +1,67 @@
+
+# export CUDA_VISIBLE_DEVICES=4,5
+# nohup python -u scripts/llm_as_judge_w_source.py \
+#     --model_path /share/project/zhipengchen/model/QwQ-32B/ \
+#     --input_file /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/73/musique_syn/turn_13.json \
+#     --dataset musique_syn \
+#     --language zh > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/73/musique_syn/llm_inf.log 2>&1 &
+
+
+# export CUDA_VISIBLE_DEVICES=4,5
+# nohup python -u scripts/llm_as_judge_w_source.py \
+#     --model_path /share/project/zhipengchen/model/QwQ-32B/ \
+#     --input_file /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/musique_syn/turn_13.json \
+#     --dataset musique_syn \
+#     --language zh > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/musique_syn/llm_inf.log 2>&1 &
+
+
+# export CUDA_VISIBLE_DEVICES=4,5
+# nohup python -u scripts/llm_as_judge_w_source.py \
+#     --model_path /share/project/zhipengchen/model/QwQ-32B/ \
+#     --input_file /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:15076#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/91/musique_syn/turn_4.json \
+#     --dataset musique_syn \
+#     --language zh > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:15076#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/91/musique_syn/llm_inf.log 2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+python -u scripts/llm_as_judge_w_source.py \
+    --model_path /share/project/zhipengchen/model/QwQ-32B/ \
+    --input_file /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/73/musique_syn/turn_13.json \
+    --dataset musique_syn \
+    --language zh > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/73/musique_syn/llm_inf.log 2>&1 
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+python -u scripts/llm_as_judge_w_source.py \
+    --model_path /share/project/zhipengchen/model/QwQ-32B/ \
+    --input_file /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/musique_syn/turn_13.json \
+    --dataset musique_syn \
+    --language zh > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/musique_syn/llm_inf.log 2>&1 
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+python -u scripts/llm_as_judge_w_source.py \
+    --model_path /share/project/zhipengchen/model/QwQ-32B/ \
+    --input_file /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:15076#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/91/musique_syn/turn_4.json \
+    --dataset musique_syn \
+    --language zh > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:15076#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/91/musique_syn/llm_inf.log 2>&1 
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+python -u scripts/llm_as_judge_w_source.py \
+    --model_path /share/project/zhipengchen/model/QwQ-32B/ \
+    --input_file /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_setting_musique_syn/turn_13.json \
+    --dataset musique_syn \
+    --language zh > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_setting_musique_syn/llm_inf.log 2>&1 
+
+
+# /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/syn_zh/turn_10.json
+
+# export CUDA_VISIBLE_DEVICES=2,3
+# nohup python -u scripts/llm_as_judge_w_source.py \
+#     --model_path /share/project/zhipengchen/model/QwQ-32B/ \
+#     --input_file /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/syn_zh/turn_10.json \
+#     --dataset musique_syn \
+#     --language zh > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/syn_zh/llm_inf.log 2>&1 &

deep_search/search_o1/process_analyses_data.py

@@ -0,0 +1,65 @@
+import json
+import os
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+from tqdm import tqdm
+os.environ["CUDA_VISIBLE_DEVICES"] = '6'
+
+
+def save_to_json(data, filename):
+    with open(filename, 'w', encoding='utf-8') as f:
+        json.dump(data, f, ensure_ascii=False, indent=4)
+    print(f"save to {filename}, data len: {len(data)}")
+def load_json(file_path):
+    with open(file_path, "r", encoding="utf-8") as f:
+        data = json.load(f)
+    print(f"load from {file_path}, data len: {len(data)}")
+    return data
+
+
+if __name__=="__main__":
+
+    file_path = "/opt/aps/workdir/sunshuang/search_o1/outputs_gen_data_hotpot_qa/runs.baselines/hotpotqa.deepseek-r1-distill-qwen-32.search_o1/batch_0_back/test.2.23,14:26.info_extract.json"
+    file_output_path = "/opt/aps/workdir/sunshuang/search_o1/outputs_gen_data_hotpot_qa/runs.baselines/hotpotqa.deepseek-r1-distill-qwen-32.search_o1/batch_0_back/test.2.23,14:26.info_extract_remove_special.json"
+    data = load_json(file_path)
+
+    model_path = "/capacity/userdata/models/DeepSeek-R1-Distill-Qwen-32"
+    tokenizer = AutoTokenizer.from_pretrained(model_path)
+
+    print([data[0]["prompt"]])
+    data_prompts = [item["prompt"] for item in data]
+    prefix_len = len("<｜begin▁of▁sentence｜><｜User｜>")
+    suffix_len = len("\n<｜Assistant｜><think>\n")
+    
+    no_help = 0
+    data_selected = []
+    for item in data:
+        if item["extracted_info"] == "No helpful information found.":
+            no_help += 1
+            continue
+        prompt = item["prompt"]
+        prompt_remove_special_tokens = prompt[prefix_len:-suffix_len]
+        item["prompt"] = prompt_remove_special_tokens
+        data_selected.append(item)
+
+    print(f"no_help: {no_help}, percent: {no_help / len(data)}")
+    print(f"selected data: {len(data_selected)}, percent: {len(data_selected) / len(data)}")
+    save_to_json(data_selected, file_output_path)
+    # print(len("\n<｜Assistant｜><think>\n"))
+    # tokenized_prompts = tokenizer(data_prompts, return_tensors="pt", padding="longest", add_special_tokens=False)["input_ids"]
+    
+    # print(f"tokenized_prompts size: {tokenized_prompts.size()}")
+
+    # decoded_prompts = tokenizer.batch_decode(tokenized_prompts, skip_special_tokens=False)
+
+    # processed_data = []
+    
+    # print(type(decoded_prompts))
+    # print(decoded_prompts[0])
+    # for idx, (item, prompt) in tqdm(enumerate(zip(data, decoded_prompts))):
+    #     if idx == 3:
+    #         break
+    #     item["prompt"] = prompt
+    #     processed_data.append(item)
+
+    # save_to_json(processed_data, file_output_path)

deep_search/search_o1/reason_two_model_1.sh

@@ -0,0 +1,60 @@
+export http_proxy=http://127.0.0.1:7880
+export https_proxy=http://127.0.0.1:7880
+
+# export https_proxy=http://127.0.0.1:7890
+# export http_proxy=http://127.0.0.1:7890
+# export all_proxy=socks5://127.0.0.1:7891
+export CUDA_VISIBLE_DEVICES=4,5
+
+model_name=JOB9986:LR1e-5:BASEQwen2.5-32B-Instruct:TOKENDeepSeek-R1-Distill-Qwen-32:BSZ1:ACC8/checkpoint-38
+
+python scripts/reason_two_model_2.py \
+    --dataset_name hotpotqa \
+    --cache_dir_base cache_reason_two_model//eval_reason_two_model_r1_32b_doc_by_qwen7b_inst \
+    --output_dir_base outputs_reason_two_model/eval_reason_two_model_r1_32b_doc_by_qwen7b_inst \
+    --split test \
+    --max_search_limit 5 \
+    --max_turn 10 \
+    --top_k 5 \
+    --max_doc_len 3000 \
+    --subset_num 500 \
+    --model_path "/capacity/userdata/models/DeepSeek-R1-Distill-Qwen-32" \
+    --model_doc_reason_path "/opt/aps/workdir/home/.cache/modelscope/hub/models/qwen/Qwen2.5-7B-Instruct" \
+    --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" \
+    --openai_api_base "http://localhost:8000/v1"
+
+
+python scripts/reason_two_model_2.py \
+    --dataset_name hotpotqa \
+    --cache_dir_base cache_reason_two_model/eval_reason_two_model_r1_32b_doc_by_qwen32b_inst \
+    --output_dir_base outputs_reason_two_model/eval_reason_two_model_r1_32b_doc_by_qwen32b_inst \
+    --split test \
+    --max_search_limit 5 \
+    --max_turn 10 \
+    --top_k 5 \
+    --max_doc_len 3000 \
+    --subset_num 500 \
+    --model_path "/capacity/userdata/models/DeepSeek-R1-Distill-Qwen-32" \
+    --model_doc_reason_path "/capacity/userdata/models/Qwen2.5-32B-Instruct" \
+    --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" \
+    --openai_api_base "http://localhost:8001/v1"
+
+# python scripts/reason_two_model_2.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache_reason_two_model/eval_qwen7b_inst_qwq \
+#     --output_dir_base outputs_reason_two_model/eval_qwen7b_inst_qwq \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/opt/aps/workdir/home/.cache/modelscope/hub/models/qwen/Qwen2.5-7B-Instruct" \
+#     --model_doc_reason_path "/capacity/userdata/models/QwQ-32B-Preview" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"

deep_search/search_o1/requirements.txt

@@ -0,0 +1,7 @@
+torch==2.5.1
+transformers==4.46.1
+sentencepiece==0.2.0
+vllm==0.6.4
+tqdm==4.67.0
+nltk==3.9.1
+pyext==0.7

deep_search/search_o1/run_eval_1.sh

@@ -0,0 +1,91 @@
+export http_proxy=http://127.0.0.1:7880
+export https_proxy=http://127.0.0.1:7880
+# export https_proxy=http://127.0.0.1:7890
+# export http_proxy=http://127.0.0.1:7890
+# export all_proxy=socks5://127.0.0.1:7891
+export CUDA_VISIBLE_DEVICES=2,3
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/reason_two_model_1.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache/eval \
+#     --output_dir_base outputs/eval_qwen_2_5_7b \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/opt/aps/workdir/output/checkpoint/qwen_2_5_7b_test" \
+#     --model_doc_reason_path "/capacity/userdata/models/DeepSeek-R1-Distill-Qwen-32" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+# 评测1217条数据训练前和训练后的qwen2.5 7B Instruct model
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache/eval \
+#     --output_dir_base outputs/eval_qwen_7b_inst_data_1217 \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/opt/aps/workdir/output/checkpoint/qwen_7b_inst_data_1217" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache/eval \
+#     --output_dir_base outputs/eval_original_qwen_7b_inst \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/opt/aps/workdir/home/.cache/modelscope/hub/models/qwen/Qwen2.5-7B-Instruct" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+# /opt/aps/workdir/output/checkpoint/qwen_7b_original_tokenizer_inst_data_1217
+
+# 将tokenizer换回自己的
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache/eval \
+#     --output_dir_base outputs/eval_qwen_7b_original_tokenizer_inst_data_1217 \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/opt/aps/workdir/output/checkpoint/qwen_7b_original_tokenizer_inst_data_1217" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+
+# /opt/aps/workdir/output/sft_use_original_tokenizer/qwen_7b_original_tokenizer_inst_data_1217
+
+/opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/reason_two_model_4.py \
+    --dataset_name hotpotqa \
+    --cache_dir_base cache_real_change_tokenizer/eval_test_wheel \
+    --output_dir_base outputs_real_change_tokenizer/eval_qwen_7b_original_tokenizer_inst_data_1217_test_wheel \
+    --split test \
+    --max_search_limit 5 \
+    --max_turn 10 \
+    --top_k 5 \
+    --max_doc_len 3000 \
+    --subset_num 3 \
+    --model_path "/opt/aps/workdir/output/sft_use_original_tokenizer/qwen_7b_original_tokenizer_inst_data_1217" \
+    --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --model_doc_reason_path "/capacity/userdata/models/Qwen2.5-32B-Instruct" \
+    --bing_endpoint "https://google.serper.dev/search" \
+    --openai_api_base "http://localhost:8001/v1"

deep_search/search_o1/run_eval_2.sh

@@ -0,0 +1,52 @@
+export http_proxy=http://127.0.0.1:7880
+export https_proxy=http://127.0.0.1:7880
+
+export CUDA_VISIBLE_DEVICES=5
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/reason_two_model_1.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache/eval \
+#     --output_dir_base outputs/eval_qwen_2_5_7b \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/opt/aps/workdir/output/checkpoint/qwen_2_5_7b_test" \
+#     --model_doc_reason_path "/capacity/userdata/models/DeepSeek-R1-Distill-Qwen-32" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+# 测试一下vllm serve+离线推理的代码
+/opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/reason_two_model_2.py \
+    --dataset_name hotpotqa \
+    --cache_dir_base cache/test_reason_two_model_2 \
+    --output_dir_base outputs/test_reason_two_model_2 \
+    --split test \
+    --max_search_limit 5 \
+    --max_turn 10 \
+    --top_k 5 \
+    --max_doc_len 3000 \
+    --subset_num 5 \
+    --model_path "/opt/aps/workdir/output/checkpoint/qwen_7b_inst_data_1217" \
+    --model_doc_reason_path "/opt/aps/workdir/home/.cache/modelscope/hub/models/qwen/Qwen2.5-7B-Instruct" \
+    --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search"
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache/eval \
+#     --output_dir_base outputs/eval_original_qwen_7b_inst \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/opt/aps/workdir/home/.cache/modelscope/hub/models/qwen/Qwen2.5-7B-Instruct" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"

deep_search/search_o1/run_eval_3.sh

@@ -0,0 +1,220 @@
+export http_proxy=http://127.0.0.1:7880
+export https_proxy=http://127.0.0.1:7880
+
+export CUDA_VISIBLE_DEVICES=2,3
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/reason_two_model_1.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache/eval \
+#     --output_dir_base outputs/eval_qwen_2_5_7b \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/opt/aps/workdir/output/checkpoint/qwen_2_5_7b_test" \
+#     --model_doc_reason_path "/capacity/userdata/models/DeepSeek-R1-Distill-Qwen-32" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache_modify_extract_answer/eval \
+#     --output_dir_base outputs_modify_extract_answer/eval_qwen_2_5_32b_data_1217 \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/opt/aps/workdir/output/checkpoint/qwen_2_5_32b_data_1217" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache_modify_extract_answer/eval \
+#     --output_dir_base outputs_modify_extract_answer/eval_original_qwen_2_5_32b \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/capacity/userdata/models/Qwen2.5-32B-Instruct" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache_modify_extract_answer/eval \
+#     --output_dir_base outputs_modify_extract_answer/eval_qwen_7b_inst_data_1217 \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/opt/aps/workdir/output/checkpoint/qwen_7b_inst_data_1217" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache_modify_extract_answer/eval \
+#     --output_dir_base outputs_modify_extract_answer/eval_original_qwen_7b_inst \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/opt/aps/workdir/home/.cache/modelscope/hub/models/qwen/Qwen2.5-7B-Instruct" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+# qwen32b
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache_eval/eval_hotpotqa_dev_500 \
+#     --output_dir_base output_eval/eval_hopotqa_dev_500_original_qwen_32b_inst \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 500 \
+#     --model_path "/capacity/userdata/models/Qwen2.5-32B-Instruct" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+
+datasets=('2wiki' 'hotpotqa' 'musique')
+
+
+for dataset in "${datasets[@]}"; do
+    echo "Running evaluation for dataset: $dataset"
+
+    cache_dir_base="cache_eval/eval_${dataset}_dev_500_1"
+    output_dir_base="output_eval/eval_${dataset}_dev_500_qwen_32b_original_tokenizer_inst_data_1217_1"
+
+    mkdir -p "$cache_dir_base"
+    mkdir -p "$output_dir_base"
+
+    /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+        --dataset_name "$dataset" \
+        --cache_dir_base "$cache_dir_base" \
+        --output_dir_base "$output_dir_base" \
+        --split test \
+        --max_search_limit 5 \
+        --max_turn 10 \
+        --top_k 5 \
+        --max_doc_len 3000 \
+        --subset_num 100 \
+        --model_path "/opt/aps/workdir/output/sft_use_original_tokenizer/qwen_32b_original_tokenizer_inst_data_1217" \
+        --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+        --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+        --bing_endpoint "https://google.serper.dev/search"
+done
+
+
+/opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+    --dataset_name hotpotqa \
+    --cache_dir_base cache_eval/eval_hotpotqa_dev_500_1 \
+    --output_dir_base output_eval/eval_hopotqa_dev_500_qwen_32b_original_tokenizer_inst_data_1217_1 \
+    --split test \
+    --max_search_limit 5 \
+    --max_turn 10 \
+    --top_k 5 \
+    --max_doc_len 3000 \
+    --subset_num 500 \
+    --model_path "/opt/aps/workdir/output/sft_use_original_tokenizer/qwen_32b_original_tokenizer_inst_data_1217" \
+    --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search"
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache_eval/eval_hotpotqa_dev_500 \
+#     --output_dir_base output_eval/eval_hopotqa_dev_500_original_r1_32b \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 500 \
+#     --model_path "/capacity/userdata/models/DeepSeek-R1-Distill-Qwen-32" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache_eval/eval_hotpotqa_dev_500 \
+#     --output_dir_base output_eval/eval_hopotqa_dev_500_original_qwen_7b_inst \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 500 \
+#     --model_path "/opt/aps/workdir/home/.cache/modelscope/hub/models/qwen/Qwen2.5-7B-Instruct" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache_modify_extract_answer/eval \
+#     --output_dir_base outputs_modify_extract_answer/eval_original_qwq \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/capacity/userdata/models/QwQ-32B-Preview" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache/eval \
+#     --output_dir_base outputs/eval_llama3_8b \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/opt/aps/workdir/output/checkpoint/llama3_8b_data_1217" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_eval.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache/eval \
+#     --output_dir_base outputs/eval_original_llama3_8b \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --subset_num 100 \
+#     --model_path "/capacity/userdata/models/QwQ-32B-Preview" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+

deep_search/search_o1/run_eval_benchmark.sh

@@ -0,0 +1,691 @@
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_ckpt176 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/search_o1/model/checkpoint-176" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_ckpt176.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwq_temp_0 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_qwq_temp_0.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwq_temp_0 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_qwq_temp_0.log 2>&1 &
+
+
+# /share/project/sunshuang/deep_search/output/JOB:16816#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_correct_solutions_sft_add_prompt.json/checkpoint-138
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwq_sft_1482_ckpt138 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:16816#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_correct_solutions_sft_add_prompt.json/checkpoint-138" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_qwq_sft_1482_ckpt138.log 2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwq_sft_1482_ckpt138 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:16816#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_correct_solutions_sft_add_prompt.json/checkpoint-138" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_qwq_sft_1482_ckpt138.log 2>&1 &
+
+
+# /share/project/sunshuang/deep_search/output/JOB:20762#LR:1e-5#BASE:DeepSeek-R1-Distill-Qwen-32B#TOKEN:DeepSeek-R1-Distill-Qwen-32B#BSZ:2#ACC:4_correct_solutions_sft_add_prompt.json/checkpoint-138
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_r132b_sft_1478_ckpt138 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:20762#LR:1e-5#BASE:DeepSeek-R1-Distill-Qwen-32B#TOKEN:DeepSeek-R1-Distill-Qwen-32B#BSZ:2#ACC:4_correct_solutions_sft_add_prompt.json/checkpoint-138" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_r132b_sft_1478_ckpt138.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_r132b \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/DeepSeek-R1-Distill-Qwen-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_r132b.log 2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwq_difficult_rollout_data_sft_add_prompt_ckpt123 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:25969#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_difficult_rollout_data_sft_add_prompt.json/checkpoint-123" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_qwq_difficult_rollout_data_sft_add_prompt_ckpt123.log 2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwq_difficult_rollout_data_sft_add_prompt_ckpt246 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:25969#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_difficult_rollout_data_sft_add_prompt.json/checkpoint-246" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_qwq_difficult_rollout_data_sft_add_prompt_ckpt246.log 2>&1 &
+
+
+
+######################
+# 使用最新的配置测试模型
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwq_new_setting \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_qwq_new_setting.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_r1_32b_new_setting \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/DeepSeek-R1-Distill-Qwen-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_r1_32b_new_setting.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwen32b_inst_new_setting \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-32B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_qwen32b_inst_new_setting.log 2>&1 &
+
+
+# /share/project/sunshuang/deep_search/output/JOB:16816#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_correct_solutions_sft_add_prompt.json/checkpoint-138
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:16816#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_correct_solutions_sft_add_prompt.json/138/eval_new_setting \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:16816#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_correct_solutions_sft_add_prompt.json/checkpoint-138" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/logs/log_eval/JOB:16816#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_correct_solutions_sft_add_prompt.json/138/eval_new_setting/inf.log 2>&1 &
+
+# /share/project/sunshuang/deep_search/search_o1/model/checkpoint-176
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_ckpt176_new_setting \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/search_o1/model/checkpoint-176" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/logs/log_eval/sum_all_webpage_ckpt176_new_setting.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwen7b_inst_new_setting_eval \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-7B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > output/output_eval/outputs_sum_all_webpage_qwen7b_inst_new_setting_eval/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+/share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_demo.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/eval_wiki \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-7B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/eval_wiki/inf.log  2>&1
+
+##################################################
+# musique_syn
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name musique_syn \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/testt \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-7B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/eval_wiki/test.log  2>&1 &
+
+
+##########################
+# aime
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name aime \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:7405#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft/73/aime \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:7405#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft/checkpoint-73" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/logs/log_eval/JOB:7405#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft/73/aime/inf.log 2>&1 &
+
+
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name aime \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:5303#LR:1e-5#BASE:Qwen2.5-32B-Instruct#TOKEN:Qwen2.5-32B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft/91/aime \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:5303#LR:1e-5#BASE:Qwen2.5-32B-Instruct#TOKEN:Qwen2.5-32B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft/checkpoint-91" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/logs/log_eval/JOB:5303#LR:1e-5#BASE:Qwen2.5-32B-Instruct#TOKEN:Qwen2.5-32B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft/91/aime/inf.log  2>&1 &
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name aime \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:11354#LR:1e-5#BASE:DeepSeek-R1-Distill-Qwen-32B#TOKEN:DeepSeek-R1-Distill-Qwen-32B#BSZ:2#ACC:4_new_instruction_2k_sft/91/aime \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:11354#LR:1e-5#BASE:DeepSeek-R1-Distill-Qwen-32B#TOKEN:DeepSeek-R1-Distill-Qwen-32B#BSZ:2#ACC:4_new_instruction_2k_sft/checkpoint-91" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/logs/log_eval/JOB:11354#LR:1e-5#BASE:DeepSeek-R1-Distill-Qwen-32B#TOKEN:DeepSeek-R1-Distill-Qwen-32B#BSZ:2#ACC:4_new_instruction_2k_sft/91/aime/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name aime \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_setting_aime \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/logs/log_eval/sum_all_webpage_qwq_new_setting_qwq.log  2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name aime \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_setting_aime \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/logs/log_eval/sum_all_webpage_qwq_new_setting_qwq.log  2>&1 &
+
+
+
+# /share/project/sunshuang/deep_search/output/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/checkpoint-73
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name aime \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/154/aime \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/checkpoint-154" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/154/aime/inf.log  2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name aime \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:15076#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/91/aime \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:15076#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/checkpoint-91" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:15076#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/91/aime/inf.log  2>&1 &
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name aime \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwen7b_inst_new_setting_aime \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-7B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > output/output_eval/outputs_sum_all_webpage_qwen7b_inst_new_setting_aime/inf.log  2>&1 &
+
+
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name aime \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/aime \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/checkpoint-228" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/aime/inf.log  2>&1 &
+
+
+######################
+# 测试中文
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name new \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/73/new \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/checkpoint-73" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/73/new/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name new \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwq_new_setting_new \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_qwq_new_setting_new.log  2>&1 &
+
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name new \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/154/new \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/checkpoint-154" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/logs/log_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/154/new/inf.log  2>&1 &
+
+
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name new \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwen32b_inst_new_setting_new \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-32B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_qwen32b_inst_new_setting_new.log  2>&1 &
+
+
+
+#############
+# 人工合成的数据
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name syn_zh \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/73/syn_zh \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/checkpoint-73" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/73/syn_zh/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name syn_zh \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwq_new_setting_syn_zh \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_qwq_new_setting_syn_zh.log  2>&1 &
+
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name syn_zh \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/154/syn_zh \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/checkpoint-154" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/logs/log_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/154/syn_zh/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name syn_zh \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/syn_zh \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/checkpoint-228" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/syn_zh/inf.log  2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/eval_1 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/checkpoint-228" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/eval_1/inf.log  2>&1 &
+
+
+
+# /share/project/sunshuang/deep_search/output/JOB:29319#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/checkpoint-96
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:29319#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/96/eval \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:29319#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/checkpoint-96" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:29319#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/96/eval/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name realqa \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:29319#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/96/eval \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:29319#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/checkpoint-96" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:29319#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/96/eval/inf.log  2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen7b_inst_new_prompt_eval \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-7B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen7b_inst_new_prompt_eval/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen32b_inst_new_prompt_eval \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-32B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen32b_inst_new_prompt_eval/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_prompt_eval \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_prompt_eval/inf.log  2>&1 &
+
+
+#############################
+# date
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_date.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage_date \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_prompt_eval_date \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_prompt_eval_date/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage_date \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen32b_inst_new_prompt_eval_date \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-32B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen32b_inst_new_prompt_eval_date/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_date.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage_date \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:11783#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_selected_data_1174_sft/91/eval_date \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:11783#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_selected_data_1174_sft/checkpoint-91" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:11783#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_selected_data_1174_sft/91/eval_date/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_date.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage_date \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:27033#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_syn_long_359_sft_1533/144/eval_date \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:27033#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_syn_long_359_sft_1533/checkpoint-144" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:27033#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_syn_long_359_sft_1533/144/eval_date/inf.log  2>&1 &
+

deep_search/search_o1/run_eval_benchmark_date.sh

@@ -0,0 +1,62 @@
+
+#############################
+# date
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_date.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage_date \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_prompt_eval_date \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_prompt_eval_date/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage_date \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen32b_inst_new_prompt_eval_date \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-32B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen32b_inst_new_prompt_eval_date/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_date.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage_date \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:11783#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_selected_data_1174_sft/91/eval_date \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:11783#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_selected_data_1174_sft/checkpoint-91" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:11783#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_selected_data_1174_sft/91/eval_date/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_date.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage_date \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:27033#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_syn_long_359_sft_1533/144/eval_date \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:27033#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_syn_long_359_sft_1533/checkpoint-144" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:27033#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_syn_long_359_sft_1533/144/eval_date/inf.log  2>&1 &
+

deep_search/search_o1/run_eval_benchmark_musique_syn.sh

@@ -0,0 +1,108 @@
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name musique_syn \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/musique_syn \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/checkpoint-228" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/musique_syn/inf.log  2>&1 &
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name musique_syn \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/73/musique_syn \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/checkpoint-73" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:9578#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/73/musique_syn/inf.log  2>&1 &
+
+
+
+# /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:15076#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/91/eval_wiki
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name musique_syn \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:15076#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/91/musique_syn \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:15076#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/checkpoint-91" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:15076#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_new_instruction_2k_sft_remove_exp/91/musique_syn/inf.log  2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name musique_syn \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_setting_musique_syn \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_setting_musique_syn/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name musique_syn \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen7b_inst_new_setting_musique_syn \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-7B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen7b_inst_new_setting_musique_syn/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name musique_syn \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:17054#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_2462_nq_237_total_2699/211/musique_syn \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:17054#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_2462_nq_237_total_2699/checkpoint-211" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:17054#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_2462_nq_237_total_2699/211/musique_syn/inf.log  2>&1 &
+
+
+
+# musique_syn
+# export CUDA_VISIBLE_DEVICES=6,7
+# nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name musique_syn \
+#     --cache_dir_base cache_eval_sum_all_webpage \
+#     --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/testt \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --model_path "/share/project/zhipengchen/model/Qwen2.5-7B-Instruct" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/eval_wiki/test.log  2>&1 &
+

deep_search/search_o1/run_eval_benchmark_realqa.sh

@@ -0,0 +1,129 @@
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name realqa \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_setting_realqa \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_setting_realqa/inf.log  2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name realqa \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen7b_inst_new_setting_realqa \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-7B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen7b_inst_new_setting_realqa/inf.log  2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name realqa \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen32b_inst_new_setting_realqa \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/Qwen2.5-32B-Instruct" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwen32b_inst_new_setting_realqa/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name realqa \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:13192#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_strict_selected_1526_sft/144/realqa \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:13192#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_strict_selected_1526_sft/checkpoint-144" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:13192#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_strict_selected_1526_sft/144/realqa/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name realqa \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:2934#LR:1e-5#BASE:Qwen2.5-32B-Instruct#TOKEN:Qwen2.5-32B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/144/realqa \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:2934#LR:1e-5#BASE:Qwen2.5-32B-Instruct#TOKEN:Qwen2.5-32B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/checkpoint-144" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:2934#LR:1e-5#BASE:Qwen2.5-32B-Instruct#TOKEN:Qwen2.5-32B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/144/realqa/inf.log  2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name realqa \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:29319#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/144/realqa \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:29319#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/checkpoint-144" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:29319#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_strict_selected_1526_sft/144/realqa/inf.log  2>&1 &
+
+
+
+
+
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name realqa \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/realqa \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/checkpoint-228" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:12665#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_new_instruction_1161_1301_sft_remove_exp_2462/228/realqa/inf.log  2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name realqa \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:17054#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_2462_nq_237_total_2699/211/realqa \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:17054#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_2462_nq_237_total_2699/checkpoint-211" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/JOB:17054#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_2462_nq_237_total_2699/211/realqa/inf.log  2>&1 &
+

deep_search/search_o1/run_eval_benchmark_rl.sh

@@ -0,0 +1,14 @@
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name simpleqa \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_prompt_eval \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval/outputs_sum_all_webpage_qwq_new_prompt_eval/inf.log  2>&1 &
+

deep_search/search_o1/run_eval_sft_1.sh

@@ -0,0 +1,89 @@
+# export CUDA_VISIBLE_DEVICES=2,3
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name eval \
+#     --cache_dir_base cache_eval_sum_all_webpage \
+#     --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_1.1k_91/eval \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --model_path "/share/project/sunshuang/deep_search/output/JOB:11783#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_selected_data_1174_sft/checkpoint-91" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_1.1k_91/eval/inf.log 2>&1 
+
+
+# export CUDA_VISIBLE_DEVICES=2,3
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name eval \
+#     --cache_dir_base cache_eval_sum_all_webpage \
+#     --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_syn_long_1533/eval \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --model_path "/share/project/sunshuang/deep_search/output/JOB:27033#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_syn_long_359_sft_1533/checkpoint-144" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_syn_long_1533/eval/inf.log 2>&1
+
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+/share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name gaia \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_1.5k/context_8k/gaia \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:13192#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_strict_selected_1526_sft/checkpoint-144" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_1.5k/context_8k/gaia/inf.log 2>&1
+
+export CUDA_VISIBLE_DEVICES=2,3
+/share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name gaia \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_2.7k/context_8k/gaia \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:17054#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_2462_nq_237_total_2699/checkpoint-211" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_2.7k/context_8k/gaia/inf.log 2>&1
+
+
+# export CUDA_VISIBLE_DEVICES=2,3
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name eval \
+#     --cache_dir_base cache_eval_sum_all_webpage \
+#     --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_1.5k/context_8k/eval \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --model_path "/share/project/sunshuang/deep_search/output/JOB:13192#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_strict_selected_1526_sft/checkpoint-144" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_1.5k/context_8k/eval/inf.log 2>&1 
+
+
+# export CUDA_VISIBLE_DEVICES=2,3
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name frames \
+#     --cache_dir_base cache_eval_sum_all_webpage \
+#     --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_1.5k/context_8k/frames \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --model_path "/share/project/sunshuang/deep_search/output/JOB:13192#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_strict_selected_1526_sft/checkpoint-144" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_1.5k/context_8k/frames/inf.log 2>&1
+

deep_search/search_o1/run_eval_sft_2.sh

@@ -0,0 +1,106 @@
+
+# export CUDA_VISIBLE_DEVICES=4,5
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name frames \
+#     --cache_dir_base cache_eval_sum_all_webpage \
+#     --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_1.1k_91/frames \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --model_path "/share/project/sunshuang/deep_search/output/JOB:11783#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_selected_data_1174_sft/checkpoint-91" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_1.1k_91/frames/inf.log 2>&1 
+
+
+# export CUDA_VISIBLE_DEVICES=4,5
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name frames \
+#     --cache_dir_base cache_eval_sum_all_webpage \
+#     --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_syn_long_1533/frames \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --model_path "/share/project/sunshuang/deep_search/output/JOB:27033#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_syn_long_359_sft_1533/checkpoint-144" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_syn_long_1533/frames/inf.log 2>&1 
+
+
+
+# export CUDA_VISIBLE_DEVICES=0,1
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name eval \
+#     --cache_dir_base cache_eval_sum_all_webpage \
+#     --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_1.1k_91/eval \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --model_path "/share/project/sunshuang/deep_search/output/JOB:11783#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_selected_data_1174_sft/checkpoint-91" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_1.1k_91/eval/inf.log 2>&1 
+
+
+# export CUDA_VISIBLE_DEVICES=0,1
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name eval \
+#     --cache_dir_base cache_eval_sum_all_webpage \
+#     --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_syn_long_1533/eval \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --model_path "/share/project/sunshuang/deep_search/output/JOB:27033#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_syn_long_359_sft_1533/checkpoint-144" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_syn_long_1533/eval/inf.log 2>&1
+
+
+
+export CUDA_VISIBLE_DEVICES=0,1
+/share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name gaia \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_2.7k/context_8k/gaia \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:17054#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_2462_nq_237_total_2699/checkpoint-211" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_2.7k/context_8k/gaia/inf.log 2>&1
+
+# export CUDA_VISIBLE_DEVICES=0,1
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name eval \
+#     --cache_dir_base cache_eval_sum_all_webpage \
+#     --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_2.7k/context_8k/eval \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --model_path "/share/project/sunshuang/deep_search/output/JOB:17054#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_2462_nq_237_total_2699/checkpoint-211" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_2.7k/context_8k/eval/inf.log 2>&1 
+
+
+# export CUDA_VISIBLE_DEVICES=0,1
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name frames \
+#     --cache_dir_base cache_eval_sum_all_webpage \
+#     --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_2.7k/context_8k/frames \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --model_path "/share/project/sunshuang/deep_search/output/JOB:17054#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_merged_2462_nq_237_total_2699/checkpoint-211" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_sft_2.7k/context_8k/frames/inf.log 2>&1
+

deep_search/search_o1/run_eval_sft_3.sh

@@ -0,0 +1,30 @@
+
+export CUDA_VISIBLE_DEVICES=6,7
+/share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name gaia \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_1.1k_91/gaia \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:11783#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_selected_data_1174_sft/checkpoint-91" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_1.1k_91/gaia/inf.log 2>&1
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+/share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name gaia \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_syn_long_1533/gaia \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/output/JOB:11783#LR:1e-5#BASE:QwQ-32B#TOKEN:QwQ-32B#BSZ:2#ACC:4_selected_data_1174_sft/checkpoint-91" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwq_syn_long_1533/gaia/inf.log 2>&1
+

deep_search/search_o1/run_eval_sft_sh.sh

@@ -0,0 +1,5 @@
+nohup bash /share/project/sunshuang/deep_search/search_o1/run_eval_sft_1.sh > /share/project/sunshuang/deep_search/search_o1/run_eval_sft_1.log 2>&1 &
+
+nohup bash /share/project/sunshuang/deep_search/search_o1/run_eval_sft_2.sh > /share/project/sunshuang/deep_search/search_o1/run_eval_sft_2.log 2>&1 &
+
+nohup bash /share/project/sunshuang/deep_search/search_o1/run_eval_sft_3.sh > /share/project/sunshuang/deep_search/search_o1/run_eval_sft_3.log 2>&1 &

deep_search/search_o1/run_eval_sft_train.sh

@@ -0,0 +1,13 @@
+export CUDA_VISIBLE_DEVICES=0,1,6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name no_error_data_871 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwen7b_inst_sft_871/78/no_error_data_871 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/q1/JOB:2569#LR:1e-5#BASE:Qwen2.5-7B-Instruct#TOKEN:Qwen2.5-7B-Instruct#BSZ:2#ACC:4_new_800/checkpoint-41" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > /share/project/sunshuang/deep_search/search_o1/output/output_eval_sft/qwen7b_inst_sft_871/78/no_error_data_871/inf.log 2>&1 &

deep_search/search_o1/run_gen_data_1.sh

@@ -0,0 +1,36 @@
+export http_proxy=http://127.0.0.1:7880
+export https_proxy=http://127.0.0.1:7880
+
+export CUDA_VISIBLE_DEVICES=4,5
+
+/opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_5.py \
+    --dataset_name hotpotqa \
+    --cache_dir_base cache_eval_hotpotqa \
+    --output_dir_base outputs_eval_hotpotqa \
+    --split test \
+    --max_search_limit 5 \
+    --max_turn 10 \
+    --top_k 5 \
+    --max_doc_len 3000 \
+    --batch_size 5000 \
+    --subset_num 100 \
+    --model_path "/opt/aps/workdir/output/checkpoint/debug_1" \
+    --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search"
+
+# /opt/aps/workdir/miniforge3/envs/search_o1/bin/python scripts/run_search_o1_5.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache_eval_hotpotqa_1 \
+#     --output_dir_base outputs_eval_hotpotqa_1 \
+#     --split test \
+#     --max_search_limit 5 \
+#     --max_turn 10 \
+#     --top_k 5 \
+#     --max_doc_len 3000 \
+#     --batch_size 5000 \
+#     --subset_num 500 \
+#     --model_path "/capacity/userdata/models/Qwen2.5-7B" \
+#     --jina_api_key "jina_04d684ee4cc54d2ebe7c43bb7ad4aff0qlkdZGwm14NFBtm5BDkgK9KNf6vQ" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"

deep_search/search_o1/run_gen_data_for_simple_qa.sh

@@ -0,0 +1,379 @@
+export http_proxy=http://127.0.0.1:7880
+export https_proxy=http://127.0.0.1:7880
+
+export CUDA_VISIBLE_DEVICES=2,3
+
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_single_page.py \
+#     --dataset_name 2wiki \
+#     --cache_dir_base cache \
+#     --output_dir_base output/outputs_sum_single_page_2wiki_ckpt176 \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --subset_num 100 \
+#     --model_path "/share/project/sunshuang/deep_research/search_o1/model/checkpoint-176" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+/share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name bamboogle \
+    --cache_dir_base cache \
+    --output_dir_base output/outputs_sum_all_webpage_bamboogle_ckpt176 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --subset_num 200 \
+    --model_path "/share/project/sunshuang/deep_research/search_o1/model/checkpoint-176" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search"
+
+
+
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name 2wiki \
+#     --cache_dir_base cache \
+#     --output_dir_base output/outputs_sum_all_webpage_2wiki_ckpt176 \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --subset_num 100 \
+#     --model_path "/share/project/sunshuang/deep_research/search_o1/model/checkpoint-176" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+
+
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_single_page.py \
+#     --dataset_name simpleqa \
+#     --cache_dir_base cache \
+#     --output_dir_base output/outputs_sum_single_page_simpleqa \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --subset_num 100 \
+#     --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name simpleqa \
+#     --cache_dir_base cache \
+#     --output_dir_base output/outputs_sum_all_webpage_simpleqa_ex_urls_qwq \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --subset_num 100 \
+#     --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+
+
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_for_rag_rl.py \
+#     --dataset_name 2wiki \
+#     --cache_dir_base cache_rag_rl \
+#     --output_dir_base output/outputs_sum_all_webpage_simpleqa_rag_rl \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --subset_num 100 \
+#     --model_path "/share/project/zhipengchen/model/Qwen-2.5-7B-base-RAG-RL" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+
+/share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name simpleqa \
+    --cache_dir_base cache \
+    --output_dir_base output/outputs_sum_all_webpage_simpleqa500_qwq \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --subset_num 500 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search"
+
+export CUDA_VISIBLE_DEVICES=2,3
+/share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name simpleqa \
+    --cache_dir_base cache_simpleqa_exurls_qwq \
+    --output_dir_base output/outputs_sum_all_webpage_simpleqa500_ex_urls_qwq \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --subset_num 500 \
+    --is_exclude_urls \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search"
+
+export CUDA_VISIBLE_DEVICES=4,5
+/share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name hotpotqa \
+    --cache_dir_base cache \
+    --output_dir_base output/outputs_sum_all_webpage_hotpotqa500_qwq \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --subset_num 500 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search"
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+/share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name simpleqa \
+    --cache_dir_base cache \
+    --output_dir_base output/outputs_sum_all_webpage_simpleqa500_ckpt176 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --subset_num 500 \
+    --model_path "/share/project/sunshuang/deep_research/search_o1/model/checkpoint-176" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search"
+
+
+
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name simpleqa \
+    --cache_dir_base cache \
+    --output_dir_base output/outputs_sum_all_webpage_simpleqa500_r1_32b \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --subset_num 500 \
+    --model_path "/share/project/zhipengchen/model/DeepSeek-R1-Distill-Qwen-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/simpleqa500_r1_32b.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_test_new_prompt.py \
+    --dataset_name simpleqa \
+    --cache_dir_base cache \
+    --output_dir_base output/output_test_prompt/outputs_sum_all_webpage_simpleqa100_new_prompt_qwq \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --subset_num 100 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/simpleqa100_new_prompt_qwq.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name simpleqa \
+    --cache_dir_base cache \
+    --output_dir_base output/output_test_prompt/outputs_sum_all_webpage_simpleqa100_old_prompt_qwq \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --subset_num 100 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/simpleqa100_old_prompt_qwq.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_test_new_prompt.py \
+    --dataset_name hotpotqa \
+    --cache_dir_base cache \
+    --output_dir_base output/output_test_prompt/outputs_sum_all_webpage_hotpotqa100_new_prompt_qwq \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --subset_num 100 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/hotpotqa100_new_prompt_qwq.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name hotpotqa \
+    --cache_dir_base cache \
+    --output_dir_base output/output_test_prompt/outputs_sum_all_webpage_hotpotqa100_old_prompt_qwq \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --subset_num 100 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/hotpotqa100_old_prompt_qwq.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_single_page.py \
+    --dataset_name simpleqa \
+    --cache_dir_base cache_single_web_analysis \
+    --output_dir_base output/output_single_web_analysis/outputs_sum_single_page_gen_data_simpleqa \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/Simpleqa_sum_single_page_gen_data_simpleqa.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_single_page.py \
+    --dataset_name hotpotqa \
+    --cache_dir_base cache_single_web_analysis \
+    --output_dir_base output/output_single_web_analysis/outputs_sum_single_page_test_hotpotqa100 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --subset_num 100 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/Simpleqa_sum_single_page_test_hotpotqa100.log 2>&1 &
+
+
+# 测试eval数据集
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_qwq \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_qwq.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_single_page.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_single_page_test_maxworker_64 \
+    --output_dir_base output/output_eval/outputs_sum_single_page_qwq_test_maxworker_64 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_single_page_qwq_test_maxworker_64.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_test_new_prompt.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_test_new_prompt_qwq \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_test_new_prompt_qwq.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name eval \
+    --cache_dir_base cache_test \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_test \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --subset_num 5 \
+    --rollout_num 5 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/test_1.log 2>&1 &
+
+
+
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name simpleqa \
+#     --cache_dir_base cache_simpleqa_exurls_ckpt176 \
+#     --output_dir_base output/outputs_sum_all_webpage_simpleqa500_ex_urls_ckpt176 \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --subset_num 500 \
+#     --is_exclude_urls \
+#     --model_path "/share/project/sunshuang/deep_research/search_o1/model/checkpoint-176" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name hotpotqa \
+#     --cache_dir_base cache \
+#     --output_dir_base output/outputs_sum_all_webpage_hotpotqa500_ckpt176 \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --subset_num 500 \
+#     --model_path "/share/project/sunshuang/deep_research/search_o1/model/checkpoint-176" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search"
+
+
+
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name eval \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_eval/outputs_sum_all_webpage_ckpt176 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/sunshuang/deep_search/search_o1/model/checkpoint-176" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_eval/sum_all_webpage_ckpt176.log 2>&1 &

deep_search/search_o1/run_gen_data_for_simple_qa_ex_urls.sh

@@ -0,0 +1,36 @@
+export http_proxy=http://127.0.0.1:7880
+export https_proxy=http://127.0.0.1:7880
+
+export CUDA_VISIBLE_DEVICES=4,5,6,7
+
+
+/share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+    --dataset_name simpleqa \
+    --cache_dir_base cache_simpleqa_real_exurls_ckpt176 \
+    --output_dir_base output/outputs_sum_all_webpage_simpleqa500_real_exurls_ckpt176 \
+    --split test \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --max_doc_len 5000 \
+    --subset_num 500 \
+    --is_exclude_urls \
+    --model_path "/share/project/sunshuang/deep_research/search_o1/model/checkpoint-176" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/simpleqa500_real_exurls_ckpt176.log 2>&1 
+
+
+# /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage.py \
+#     --dataset_name simpleqa \
+#     --cache_dir_base cache_simpleqa_real_exurls_qwq \
+#     --output_dir_base output/outputs_sum_all_webpage_simpleqa500_real_exurls_qwq \
+#     --split test \
+#     --max_search_limit 10 \
+#     --max_turn 10 \
+#     --top_k 10 \
+#     --max_doc_len 5000 \
+#     --subset_num 500 \
+#     --is_exclude_urls \
+#     --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+#     --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+#     --bing_endpoint "https://google.serper.dev/search" > logs/simpleqa500_real_exurls_qwq.log 2>&1 

deep_search/search_o1/run_gen_data_mix.sh

@@ -0,0 +1,246 @@
+# worker 1
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_1 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_1 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_1.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_2 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_2 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_2.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_3 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_3 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_3.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_4 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_4 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_4.log 2>&1 &
+
+# worker 2
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_5 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_5 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_5.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_6 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_6 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_6.log 2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_7 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_7 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_7.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=6,7 
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_8 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_8 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_8.log 2>&1 &
+
+# worker 3
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_9 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_9 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_9.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_10 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_10 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_10.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_11 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_11 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_11.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_12 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_12 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_12.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_13 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_13 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_13.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_14 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_14 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 9 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_14.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_15 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_15 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 9 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_15.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name split_16 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_3200_split_16 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/3200_split_16.log 2>&1 &

deep_search/search_o1/run_gen_data_mix_last_8000.sh

@@ -0,0 +1,260 @@
+# worker 1
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name mixed_data_left_last_8000_split_1 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_mixed_data_left_last_8000_split_1 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 3 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/mixed_data_left_last_8000_split_1.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name mixed_data_left_last_8000_split_2 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_mixed_data_left_last_8000_split_2 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 1 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/mixed_data_left_last_8000_split_2.log 2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name mixed_data_left_last_8000_split_3 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_mixed_data_left_last_8000_split_3 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 3 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/mixed_data_left_last_8000_split_3.log 2>&1 &
+
+
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name mixed_data_left_last_8000_split_4 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_mixed_data_left_last_8000_split_4 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 3 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/mixed_data_left_last_8000_split_4.log 2>&1 &
+
+
+# worker 2
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name mixed_data_left_last_8000_split_5 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_mixed_data_left_last_8000_split_5 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 3 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/mixed_data_left_last_8000_split_5.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name mixed_data_left_last_8000_split_6 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_mixed_data_left_last_8000_split_6 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 1 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/mixed_data_left_last_8000_split_6.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name mixed_data_left_last_8000_split_7 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_mixed_data_left_last_8000_split_7 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 3 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/mixed_data_left_last_8000_split_7.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name mixed_data_left_last_8000_split_8 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_mixed_data_left_last_8000_split_8 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 3 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/mixed_data_left_last_8000_split_8.log 2>&1 &
+
+
+# worker 3
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name mixed_data_left_last_8000_split_9 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_mixed_data_left_last_8000_split_9 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 3 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/mixed_data_left_last_8000_split_9.log 2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name mixed_data_left_last_8000_split_10 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_mixed_data_left_last_8000_split_10 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 3 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/mixed_data_left_last_8000_split_10.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name mixed_data_left_last_8000_split_11 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_mixed_data_left_last_8000_split_11 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 3 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/mixed_data_left_last_8000_split_11.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name mixed_data_left_last_8000_split_12 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_mixed_data_left_last_8000_split_12 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 3 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/mixed_data_left_last_8000_split_12.log 2>&1 &
+
+
+
+# gen nq random select 500
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name nq_500_split_0 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_nq_500_split_0 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 5 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/nq_500_split_0.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name nq_500_split_1 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_nq_500_split_1 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 5 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/nq_500_split_1.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name nq_500_split_2 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_nq_500_split_2 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 5 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/nq_500_split_2.log 2>&1 &
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name nq_500_split_3 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_nq_500_split_3 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 5 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/nq_500_split_3.log 2>&1 &

deep_search/search_o1/run_gen_data_selected_remove_2k.sh

@@ -0,0 +1,201 @@
+# worker 1
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_1 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_1 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_1.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_2 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_2 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_2.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_3 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_3 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_3.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_4 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_4 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_4.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_5 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_5 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_5.log 2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_6 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_6 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_6.log 2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_7 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_7 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_7.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_8 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_8 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_8.log 2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=0,1
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_9 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_9 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_9.log 2>&1 &
+
+
+
+
+
+
+export CUDA_VISIBLE_DEVICES=2,3
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_10 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_10 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_10.log 2>&1 &
+
+
+
+export CUDA_VISIBLE_DEVICES=4,5
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_11 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_11 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_11.log 2>&1 &
+
+
+export CUDA_VISIBLE_DEVICES=6,7
+nohup /share/project/miniconda/envs/search_o1/bin/python -u scripts/search_o1_sum_all_webpage_rollout.py \
+    --dataset_name merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_12 \
+    --cache_dir_base cache_eval_sum_all_webpage \
+    --output_dir_base output/output_sum_all_webpage_gen_data/outputs_merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_12 \
+    --split gen \
+    --max_search_limit 10 \
+    --max_turn 10 \
+    --top_k 10 \
+    --rollout_num 10 \
+    --max_doc_len 5000 \
+    --model_path "/share/project/zhipengchen/model/QwQ-32B" \
+    --bing_subscription_key "cb0d28279a826d7e5cf22d71f683c77ffd4ba27d" \
+    --bing_endpoint "https://google.serper.dev/search" > logs/log_sum_all_webpage_gen_data/merged_tagged_domain_keypoints_keywords_count_hop_remove_2k_data_split_12.log 2>&1 &
+
+

deep_search/search_o1/search.json

@@ -0,0 +1,172 @@
+{
+    "searchParameters": {
+        "q": "when does season 14 of grey's anatomy come out",
+        "type": "search",
+        "num": 11,
+        "mkt": "en-US",
+        "setLang": "en",
+        "engine": "google"
+    },
+    "answerBox": {
+        "snippet": "The fourteenth season of the American television medical drama Grey's Anatomy was ordered on February 10, 2017, by American Broadcasting Company (ABC), and premiered on September 28, 2017 with a special 2-hour premiere.",
+        "snippetHighlighted": [
+            "September 28, 2017"
+        ],
+        "title": "Grey's Anatomy season 14 - Wikipedia",
+        "link": "https://en.wikipedia.org/wiki/Grey%27s_Anatomy_season_14"
+    },
+    "organic": [
+        {
+            "title": "Season 14 (Grey's Anatomy)",
+            "link": "https://greysanatomy.fandom.com/wiki/Season_14_(Grey%27s_Anatomy)",
+            "snippet": "The fourteenth season of Grey's Anatomy premiered September 28, 2017, with a two-hour premiere. It was broadcast on Thursday nights at 8 PM on ABC.",
+            "position": 1
+        },
+        {
+            "title": "Grey's Anatomy Season 14 Premiere Date Announced - ABC",
+            "link": "https://abc.com/news/9fce69b2-3b93-40f3-9cd3-d75b1d90bb05/category/738075",
+            "snippet": "Grey's Anatomy fans, get ready for the Season 14 two-hour premiere on Thursday, September 28 at 8|7c!",
+            "date": "Jul 31, 2017",
+            "position": 2
+        },
+        {
+            "title": "Grey's Anatomy Season 14 [DVD] [2018] - Amazon.com",
+            "link": "https://www.amazon.com/Greys-Anatomy-Season-14-DVD/dp/B07H2VPXY1",
+            "snippet": "Grey's Anatomy Season 14 [DVD] [2018] ; Language, ‎English (Dolby Digital 5.1) ; Aspect Ratio, ‎Unknown ; Release date, ‎October 22, 2018 ; ASIN, ‎B07H2VPXY1.",
+            "rating": 3.9,
+            "ratingCount": 1134,
+            "position": 3
+        },
+        {
+            "title": "Season 14 – Grey's Anatomy - Rotten Tomatoes",
+            "link": "https://www.rottentomatoes.com/tv/greys_anatomy/s14",
+            "snippet": "Watch Grey's Anatomy — Season 14 with a subscription on Disney+, Hulu, Netflix, or buy it on Fandango at Home. Grey's Anatomy — Season 14. What to Know.",
+            "position": 4
+        },
+        {
+            "title": "Grey's Anatomy (TV Series 2005– ) - Episode list - IMDb",
+            "link": "https://www.imdb.com/title/tt0413573/episodes/?season=14",
+            "snippet": "S14.E1 ∙ Break Down the House. Thu, Sep 28, 2017.",
+            "sitelinks": [
+                {
+                    "title": "Break Down the House",
+                    "link": "https://www.imdb.com/title/tt6523880/"
+                },
+                {
+                    "title": "Fight For Your Mind",
+                    "link": "https://www.imdb.com/title/tt7044330/"
+                },
+                {
+                    "title": "1-800-799-7233",
+                    "link": "https://www.imdb.com/title/tt7043736/"
+                },
+                {
+                    "title": "9.1",
+                    "link": "https://www.imdb.com/title/tt7043724/"
+                }
+            ],
+            "position": 5
+        },
+        {
+            "title": "Grey's Anatomy - Wikipedia",
+            "link": "https://en.wikipedia.org/wiki/Grey%27s_Anatomy",
+            "snippet": "The series premiered on March 27, 2005, on ABC as a mid-season replacement. The show's title is a reference to Gray's Anatomy, a classic human anatomy textbook.",
+            "sitelinks": [
+                {
+                    "title": "Season 14",
+                    "link": "https://en.wikipedia.org/wiki/Grey%27s_Anatomy_season_14"
+                },
+                {
+                    "title": "List of Grey's Anatomy episodes",
+                    "link": "https://en.wikipedia.org/wiki/List_of_Grey%27s_Anatomy_episodes"
+                },
+                {
+                    "title": "Season 1",
+                    "link": "https://en.wikipedia.org/wiki/Grey%27s_Anatomy_season_1"
+                },
+                {
+                    "title": "Season 15",
+                    "link": "https://en.wikipedia.org/wiki/Grey%27s_Anatomy_season_15"
+                }
+            ],
+            "position": 6
+        },
+        {
+            "title": "1-800-799-7233 | Grey's Anatomy Universe Wiki | Fandom",
+            "link": "https://greysanatomy.fandom.com/wiki/1-800-799-7233",
+            "snippet": "1-800-799-7233 is the ninth episode and mid-season premiere of the fourteenth season and the 302nd overall episode of Grey's Anatomy.",
+            "date": "Jan 18, 2018",
+            "position": 7
+        },
+        {
+            "title": "Grey's Anatomy Season 14 \"Most Addictive Drama\" Promo (HD)",
+            "link": "https://www.youtube.com/watch?v=4fdiQqDp6BY",
+            "snippet": "Grey's Anatomy Season 14 premieres thursday September 28th on ABC! Meredith and the team are focused on helping Owen's sister after her ...",
+            "date": "Sep 21, 2017",
+            "position": 8
+        },
+        {
+            "title": "Grey's Anatomy Season 14 Air Dates & Countdown - EpisoDate.com",
+            "link": "https://www.episodate.com/tv-show/grey-s-anatomy?season=14",
+            "snippet": "Grey's Anatomy Season 14 Air Dates ; Thursday Sep 28, 2017 · S14E01 - Break Down the House ; Thursday Sep 28, 2017 · S14E02 - Get Off on the Pain ; Thursday Oct 05, ...",
+            "rating": 9.5,
+            "ratingMax": 10,
+            "ratingCount": 109,
+            "position": 9
+        },
+        {
+            "title": "'Grey's Anatomy' Is Obviously Returning For Season 14 - Bustle",
+            "link": "https://www.bustle.com/p/when-does-greys-anatomy-season-14-premiere-the-show-will-go-on-as-long-as-meredith-does-57795",
+            "snippet": "UPDATE: ABC announced that Grey's Anatomy Season 14 will have a two-hour season premiere on Sept. 28. EARLIER: Grey's Anatomy is set to air on ...",
+            "date": "May 18, 2017",
+            "position": 10
+        }
+    ],
+    "peopleAlsoAsk": [
+        {
+            "question": "When did season 14 of Grey's Anatomy come out?",
+            "snippet": "The fourteenth season of Grey's Anatomy premiered September 28, 2017, with a two-hour premiere.",
+            "title": "Season 14 (Grey's Anatomy)",
+            "link": "https://greysanatomy.fandom.com/wiki/Season_14_(Grey%27s_Anatomy)"
+        },
+        {
+            "question": "When did Grey's Anatomy season 15 come out?",
+            "snippet": "The fifteenth season of the American television medical drama Grey's Anatomy was ordered on April 20, 2018, by American Broadcasting Company (ABC). The season premiered on September 27, 2018, with a special 2-hour premiere. The episode count for the season consists in 25 episodes.",
+            "title": "Grey's Anatomy season 15 - Wikipedia",
+            "link": "https://en.wikipedia.org/wiki/Grey%27s_Anatomy_season_15"
+        },
+        {
+            "question": "Is Grey's Anatomy season 19 out?",
+            "snippet": "It premiered on October 6, 2022, for the 2022–23 broadcast television season. Krista Vernoff returned for the season as executive producer and showrunner alongside her production company Trip the Light Productions.",
+            "title": "Grey's Anatomy season 19 - Wikipedia",
+            "link": "https://en.wikipedia.org/wiki/Grey%27s_Anatomy_season_19"
+        }
+    ],
+    "relatedSearches": [
+        {
+            "query": "Grey's Anatomy season 14 cast Guest Stars"
+        },
+        {
+            "query": "Grey's Anatomy season 14 cast interns"
+        },
+        {
+            "query": "Grey's Anatomy Season 15 summary"
+        },
+        {
+            "query": "How many Episodes in Grey's Anatomy season 15"
+        },
+        {
+            "query": "When did season 15 of Grey's Anatomy come out"
+        },
+        {
+            "query": "Grey's Anatomy season 14 Episode 11 full cast"
+        },
+        {
+            "query": "Grey's Anatomy season 14 Episode 16 cast astronaut"
+        },
+        {
+            "query": "Who dies in season 15 of Grey's Anatomy"
+        }
+    ],
+    "credits": 2
+}