从最开始使用cursor（免费版）写函数，到现在利用本地模型做简单任务，有几点体会。
1 开源和免费系列差别不大
写代码方面大同小异，可能是因为主要训练素材来自流行的开源代码仓库。gemma3:4b综合表现还行。有不少独立开发者说claude code收费版很厉害，目前还没试过。
2 本地模型适合做编程助理
做安全很多年，不喜欢用云上的模型。简单使用了下程序员助理，问题不大。
我：写python脚本，把html转成word，保留样式
gemma3：
from bs4 import BeautifulSoup
from docx import Document
from docx.shared import Inches
def html_to_word(html_file_path, word_file_path):
"""
Converts an HTML file to a Word document, attempting to preserve styles.
Args:
html_file_path (str): Path to the HTML file.
word_file_path (str): Path to save the generated Word document.
"""
try:
with open(html_file_path, 'r', encoding='utf-8') as f:
html_content = f.read()
except FileNotFoundError:
print(f"Error: HTML file not found at {html_file_path}")
return
except Exception as e:
print(f"Error reading HTML file: {e}")
return
soup = BeautifulSoup(html_content, 'html.parser')
document = Document()
# Add styles based on HTML tags and attributes
for element in soup.find_all():
if element.name == 'h1':
document.add_heading(element.text, level=1)
elif element.name == 'h2':
document.add_heading(element.text, level=2)
elif element.name == 'h3':
document.add_heading(element.text, level=3)
elif element.name == 'p':
paragraph = document.add_paragraph(element.text)
# Add basic paragraph formatting (you can customize this further)
paragraph.paragraph_format.space_after = Inches(0) # Adjust spacing as needed
elif element.name == 'img':
# Add image (basic implementation - you might need to handle image paths and sizes better)
document.add_picture(element['src'], width=Inches(4)) # Adjust width as needed
elif element.name == 'a':
# Add hyperlink (basic implementation)
document.add_paragraph(element.text, style='list bullet')
document.paragraphs[-1].embed(element['href'], 'http', element['href'])
# Add other styles as needed (e.g., tables, lists, etc.)
document.save(word_file_path)
print(f"Word document created successfully at {word_file_path}")
# Example usage:
if __name__ == "__main__":
html_file = "your_html_file.html" # Replace with your HTML file path
word_file = "output.docx" # Replace with your desired Word file path
html_to_word(html_file, word_file)格式挺工整的，但是如果想要功能可用，还要考虑很多问题，比如图片链接等。需要来多几次对话，或者一次想清楚，然后交给AI处理。