kunthawat/ALwrity
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

AI Blogen - WIP - 0.0.00000.001

Browse Source
This commit is contained in:
AjaySi
2024-01-13 16:27:28 +05:30
parent b51e9a8c2f
commit fd7053fb4b
17 changed files with 1003 additions and 101 deletions
Download Patch File Download Diff File Expand all files Collapse all files

323
lib/arxiv_schlorly_research.py Normal file
View File

@@ -0,0 +1,323 @@
####################################################
#
# FIXME: Gotta use this lib: https://github.com/monk1337/resp/tree/main
# https://github.com/danielnsilva/semanticscholar
# https://github.com/shauryr/S2QA
#
####################################################
import os
import sys
import re
import pandas as pd
import arxiv
import PyPDF2
import io
import requests
from bs4 import BeautifulSoup
import urllib.parse
from loguru import logger
logger.remove()
logger.add(sys.stdout,
colorize=True,
format="<level>{level}</level>|<green>{file}:{line}:{function}</green>| {message}"
)
def fetch_arxiv_data(query, max_results=10):
try:
# Construct the default API client
client = arxiv.Client()
# Search for articles matching the keyword
search = arxiv.Search(
query=query,
max_results=max_results,
sort_by=arxiv.SortCriterion.SubmittedDate
)
# Fetching results
results = list(client.results(search))
# Extracting data
all_data = []
for result in results:
temp = [result.title, result.published, result.entry_id, result.summary, result.pdf_url]
all_data.append(temp)
return all_data
except Exception as e:
print("An error occurred while fetching data from arXiv:", e)
raise e
def create_dataframe(data, column_names):
try:
df = pd.DataFrame(data, columns=column_names)
return df
except Exception as e:
print("An error occurred while creating DataFrame:", e)
return pd.DataFrame()
def get_arxiv_main_content(url):
"""
Returns the main content of an arXiv paper.
Args:
url (str): The URL of the arXiv paper.
Returns:
str: The main content of the paper as a string.
"""
try:
# Send a GET request to the URL
response = requests.get(url)
response.raise_for_status() # Raise an exception for HTTP errors
# Parse the HTML content
soup = BeautifulSoup(response.content, "html.parser")
# Find the main content in 'ltx_page_content'
main_content = soup.find('div', class_='ltx_page_content')
if not main_content:
logger.warning("Main content not found in the page.")
return "Main content not found."
# Remove specific section with class 'package-alerts ltx_document'
alert_section = main_content.find('div', class_='package-alerts ltx_document')
if alert_section:
alert_section.decompose()
# Optional: Remove abstract and authors if present
for element_id in ["abs", "authors"]:
element = main_content.find(id=element_id)
if element:
element.decompose()
return main_content.text.strip()
# Could not access the arxiv HTML content, instead download pdf and read its content.
except Exception as html_error:
logger.warning(f"HTML content not accessible, trying PDF: {html_error}")
try:
# Extract arXiv ID from URL
arxiv_id = url.split('/')[-1]
# Fetch paper information using arXiv API
paper = next(arxiv.Client().results(arxiv.Search(id_list=[arxiv_id])))
pdf_filename = paper.download_pdf(filename=f"downloaded-paper-{arxiv_id}.pdf")
# Initialize an empty string to store the extracted text
pdf_text = ''
# Read the downloaded PDF
with open(pdf_filename, 'rb') as f:
pdf_reader = PyPDF2.PdfReader(f)
for page in pdf_reader.pages:
try:
# Attempt to extract text from the current page
page_text = page.extract_text()
# If text extraction is successful, add it to the cumulative text
if page_text:
pdf_text += page_text + '\n'
except UnicodeDecodeError as err:
# FIXME: Handle any UnicodeDecodeError that arises during text extraction
logger.error(f"UnicodeDecodeError that arises during text extraction: {err}")
pass
# Optionally, remove the downloaded PDF file
os.remove(pdf_filename)
# Pattern to match 'References' and everything that follows
pattern = r'References\s*.*'
pdf_text = re.sub(pattern, '', pdf_text, flags=re.IGNORECASE | re.DOTALL)
sections_to_remove = ['Acknowledgements', 'References', 'Bibliography']
for section in sections_to_remove:
# Pattern to match the section title and any text following it until the next big title or end of document
pattern = r'(' + re.escape(section) + r'\s*.*?)(?=\n[A-Z]{2,}|$)'
pdf_text = re.sub(pattern, '', pdf_text, flags=re.DOTALL | re.IGNORECASE)
return pdf_text
except Exception as pdf_error:
logger.error(f"Failed to process PDF: {pdf_error}")
return "Failed to retrieve content."
def download_image(image_url, base_url, folder="images"):
# Skip downloading if the image URL is a data URI
if image_url.startswith('data:image'):
print(f"Skipping download of data URI image: {image_url}")
return False
# Create the folder if it doesn't exist
if not os.path.exists(folder):
os.makedirs(folder)
# Form the absolute URL for image paths
if not urllib.parse.urlparse(image_url).scheme:
if not base_url.endswith('/'):
base_url += '/'
image_url = base_url + image_url
# Download and save the image
try:
response = requests.get(image_url)
response.raise_for_status()
image_name = image_url.split("/")[-1]
with open(os.path.join(folder, image_name), 'wb') as file:
file.write(response.content)
return True
except requests.RequestException as e:
print(f"Error downloading {image_url}: {str(e)}")
return False
def scrape_images_from_arxiv(url):
try:
response = requests.get(url)
response.raise_for_status()
soup = BeautifulSoup(response.text, 'html.parser')
images = soup.find_all('img')
image_urls = [img['src'] for img in images if 'src' in img.attrs]
return image_urls
except requests.RequestException as e:
print(f"Error fetching page {url}: {str(e)}")
return []
def arxiv_bibtex(arxiv_id):
"""
Get the BibTeX entry for an arXiv paper.
Args:
arxiv_id: The arXiv ID of the paper.
Returns:
A string containing the BibTeX entry.
"""
import urllib.request, xml.dom.minidom
# Download the XML
try:
usock = urllib.request.urlopen(f'http://export.arxiv.org/api/query?id_list={arxiv_id}')
xmldoc = xml.dom.minidom.parse(usock)
usock.close()
except Exception as e:
raise e
# Parse the XML
entry = xmldoc.getElementsByTagName("entry")[0]
date = entry.getElementsByTagName("updated")[0].firstChild.data
text_year = date[:4]
title = entry.getElementsByTagName("title")[0]
text_title = title.firstChild.data.strip()
authorlist = []
first = True
for person_name in entry.getElementsByTagName("author"):
# Get names
name = person_name.getElementsByTagName("name")[0]
text_name = name.firstChild.data
text_given_name = ' '.join(text_name.split()[:-1])
text_surname = text_name.split()[-1]
authorlist.append(f"{text_surname}, {text_given_name}")
# First author?
if first:
text_first_author_surname = text_surname
first = False
# Construct the BibTeX entry
bibtex = f"@MISC{{{text_first_author_surname}{text_year[-2:]},\n"
bibtex += f" author = {' and '.join(authorlist)},\n"
bibtex += f" title = {{{text_title}}},\n"
bibtex += f" year = {{{text_year}}},\n"
bibtex += f" eprint = {{{arxiv_id}}},\n"
bibtex += f" url = {{http://arxiv.org/abs/{arxiv_id}}}\n"
bibtex += "}"
return bibtex
#from serpapi import GoogleSearch
#params = {
# "api_key": "os.getenv(SERPER_API_KEY)",
# "engine": "google_scholar",
# "q": "llm",
# "hl": "en",
# "as_ylo": "2023",
# "as_yhi": "2024"
#}
#search = GoogleSearch(params)
#results = search.get_dict()
def extract_arxiv_ids_from_line(line):
"""
Extract the arXiv ID from a given line of text.
Args:
line (str): A line of text potentially containing an arXiv URL.
Returns:
str: The extracted arXiv ID, or None if not found.
"""
arxiv_id_pattern = re.compile(r'arxiv\.org\/abs\/(\d+\.\d+)(v\d+)?')
match = arxiv_id_pattern.search(line)
if match:
return match.group(1) + (match.group(2) if match.group(2) else '')
return None
def read_written_ids(file_path):
"""
Read already written arXiv IDs from a file.
Args:
file_path (str): Path to the file containing written IDs.
Returns:
set: A set of arXiv IDs.
"""
written_ids = set()
try:
with open(file_path, 'r') as file:
for line in file:
written_ids.add(line.strip())
except FileNotFoundError:
logger.error(f"File not found: {file_path}")
except Exception as e:
logger.error(f"Error while reading the file: {e}")
return written_ids
def append_id_to_file(arxiv_id, output_file_path):
"""
Append a single arXiv ID to a file. Checks if the file exists and creates it if not.
Args:
arxiv_id (str): The arXiv ID to append.
output_file_path (str): Path to the output file.
"""
try:
# Check if file exists
if not os.path.exists(output_file_path):
logger.info(f"File does not exist. Creating new file: {output_file_path}")
# Create a new file and append the ID
with open(output_file_path, 'a') as outfile:
outfile.write(arxiv_id + '\n')
else:
logger.info(f"Appending to existing file: {output_file_path}")
# File exists, append the ID
with open(output_file_path, 'a') as outfile:
outfile.write(arxiv_id + '\n')
except Exception as e:
logger.error(f"Error while appending to file: {e}")

54
lib/combine_research_and_blog.py
View File

@@ -1,13 +1,23 @@
import sys
from .gpt_providers.openai_chat_completion import openai_chatgpt
from .gpt_providers.gemini_pro_text import gemini_text_response
from loguru import logger
logger.remove()
logger.add(sys.stdout,
colorize=True,
format="<level>{level}</level>|<green>{file}:{line}:{function}</green>| {message}"
)
def blog_with_research(report, blog):
def blog_with_research(report, blog, gpt_providers="openai"):
"""Combine the given online research and gpt blog content"""
prompt = f"""
You are an expert copywriter specializing in content optimization for SEO.
I will provide you with a 'research report' and a 'blog content' on the same topic.
Your task is to transform and combine the given research and blog content into a well-structured, unique
Your task is to transform and combine the given research and blog content into a well-structured markdown, unique
and engaging blog article.
Your objectives include:
@@ -27,18 +37,40 @@ def blog_with_research(report, blog):
11. Ensure Uniqueness: Guarantee the article is plagiarism-free. Write in unique, informative style.
12. Punctuation: Use appropriate question marks at the end of questions.
13. Pass AI Detection Tools: Create content that easily passes AI plagiarism detection tools.
14. REMEMBER: Use the formatting style of given research report and include citations, referances in combined article.
14. REMEMBER: Use the formatting style of given research report and include highlights, citations, referances in combined article.
Follow these guidelines to combine and write a new, unique, and informative blog article
Follow these guidelines to combine and write a new, unique, and informative blog article
that will rank well in search engine results and engage readers effectively.
Create a blog post from the given research report and blog content below.
Create a blog post, in markdown, from the given research report and blog content below.
Research report: {report}
Blog content: {blog}
"""
try:
# TBD: Add logic for which_provider and which_model
response = openai_chatgpt(prompt)
return response
except Exception as err:
SystemError(f"Error in combining blog and research report.")
if 'gemini' in gpt_providers:
prompt = f"""You are an expert copywriter specializing in content optimization for SEO.
You are world famous writer, known for your originality and engaging content.
I will provide you with a 'research report' and a 'blog content' on the same topic.
Your task is to transform and combine the given research and blog content into a blog article.
Your blog should be highly detailed and well formatted.
Include a section in your blog on the highlights section of blog content.
Do not miss out any details from provided content. Always, include figures, data, results from given content.
It is important that your blog is original and unique. It should be highly readable and SEO optimized.
Research report: '{report}'
Blog content: '{blog}'
"""
try:
response = gemini_text_response(prompt)
return response
except Exception as err:
logger.error(f"Failed to get response from gemini: {err}")
raise err
elif 'openai' in gpt_providers:
try:
logger.info("Calling OpenAI LLM.")
response = openai_chatgpt(prompt)
return response
except Exception as err:
logger.error(f"failed to get response from Openai: {err}")
raise err

15
lib/convert_content_to_markdown.py
View File

@@ -4,12 +4,13 @@ from .gpt_providers.gemini_pro_text import gemini_text_response
def convert_tomarkdown_format(blog_content, gpt_provider="openai"):
""" Helper for converting content to markdown format for static sites. """
prompt = f"""
As an expert in markdown language format and font matter,
prompt = f"""As an expert in markdown language format and font matter,
I will provide you with a blog post.
Your task is to improve formatting of given blog post.
Your task is to only Improve the formatting and structure of a blog post to enhance readability, visual appeal, and overall user experience. Do not alter the content of the provided blog. Modify only for the formatting.
Dont provide explanations, just your final response.
Use below guidelines to do formatting, structuring to make it highly readable:
Guidelines to do formatting:
1. **Headings for Structure:**
- Use # for the main title of the blog post.
- Use ## for subheadings that divide the post into clear sections.
@@ -54,8 +55,6 @@ def convert_tomarkdown_format(blog_content, gpt_provider="openai"):
- Keep the blog post organized and easy to navigate.
- Use a consistent formatting style throughout the post.
Dont provide explanations, just your final response.
Convert the given blog post in well organised markdown content:\n
Blog Post: '{blog_content}'"""
if 'openai' in gpt_provider:
@@ -65,6 +64,10 @@ def convert_tomarkdown_format(blog_content, gpt_provider="openai"):
except Exception as err:
SystemError(f"Openai Error in converting to Markdown format.")
elif 'gemini' in gpt_provider:
prompt = f""" Convert the given blog post into well structured MARKDOWN content.
Do not alter the given blog post.
blog post: "{blog_content}" """
try:
response = gemini_text_response(prompt)
return response

26
lib/get_blog_metadata.py Normal file
View File

@@ -0,0 +1,26 @@
import sys
from .get_blog_meta_desc import generate_blog_description
from .get_tags import get_blog_tags
from .get_blog_category import get_blog_categories
from .get_blog_title import generate_blog_title
from loguru import logger
logger.remove()
logger.add(sys.stdout,
colorize=True,
format="<level>{level}</level>|<green>{file}:{line}:{function}</green>| {message}"
)
def blog_metadata(blog_content, gpt_providers="openai"):
""" Common function to get blog metadata """
blog_title = generate_blog_title(blog_content, gpt_providers)
blog_meta_desc = generate_blog_description(blog_content, gpt_providers)
logger.info(f"The blog meta description is: {blog_meta_desc}\n")
blog_tags = get_blog_tags(blog_content, gpt_providers)
logger.info(f"Blog tags for generated content: {blog_tags}")
blog_categories = get_blog_categories(blog_content, gpt_providers)
logger.info(f"Generated blog categories: {blog_categories}\n")
return(blog_title, blog_meta_desc, blog_tags, blog_categories)

2
lib/get_blog_title.py
View File

@@ -36,4 +36,4 @@ def generate_blog_title(blog_article, gpt_providers="openai"):
response = openai_chatgpt(prompt)
return response
except Exception as err:
SystemError(f"Error in generating blog summary: {err}")
SystemError(f"Failed to get response from Openai: {err}")

79
lib/gpt_providers/gemini_arvix_image_details.py Normal file
View File

@@ -0,0 +1,79 @@
"""
"""
import os
import logging
from pathlib import Path
import google.generativeai as genai
logging.basicConfig(level=logging.INFO, format='%(asctime)s-%(levelname)s-%(module)s-%(lineno)d-%(message)s')
from dotenv import load_dotenv
load_dotenv(Path('../../.env'))
from tenacity import (
retry,
stop_after_attempt,
wait_random_exponential,
) # for exponential backoff
@retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
def gemini_arxiv_img_info(img_path):
""" Get image details from arxiv papers. """
try:
genai.configure(api_key=os.getenv("API_KEY"))
except Exception as e:
logging.error(f"Could not load gemini API key: {e}")
raise e
# Set up the model
generation_config = {
"temperature": 0.9,
"top_p": 1,
"top_k": 1,
"max_output_tokens": 1096,
}
safety_settings = [{
"category": "HARM_CATEGORY_HARASSMENT",
"threshold": "BLOCK_MEDIUM_AND_ABOVE"
},
{
"category": "HARM_CATEGORY_HATE_SPEECH",
"threshold": "BLOCK_MEDIUM_AND_ABOVE"
},
{
"category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
"threshold": "BLOCK_MEDIUM_AND_ABOVE"
},
{
"category": "HARM_CATEGORY_DANGEROUS_CONTENT",
"threshold": "BLOCK_MEDIUM_AND_ABOVE"
},]
try:
model = genai.GenerativeModel(model_name="gemini-pro-vision",
generation_config=generation_config,
safety_settings=safety_settings)
except Exception as e:
logging.error(f"Could not create GenerativeModel: {e}")
raise e
# Validate that an image is present
if not (img := Path(img_path)).exists():
raise FileNotFoundError(f"Could not find image: {img}")
image_parts = [{
"mime_type": "image/png",
"data": Path(img_path).read_bytes()
},]
prompt_parts = [
"As scholar on evaluating research papers, I will provide you with an image from a research paper. Your task is to explain the image in details so that I can use it in a blog article. Explain the key findings and conclusions from the image. Your description should be in simple terms to explain to a wider audience. Explain key findings from the given image.",
image_parts[0],]
try:
response = model.generate_content(prompt_parts)
return response.text
except Exception as e:
logging.error(f"Could not generate gemini content: {e}")
raise e

14
lib/gpt_providers/gemini_pro_text.py
View File

@@ -7,6 +7,7 @@ import google.generativeai as genai
logging.basicConfig(level=logging.INFO, format='%(asctime)s-%(levelname)s-%(module)s-%(lineno)d-%(message)s')
from dotenv import load_dotenv
load_dotenv(Path('../../.env'))
from .mistral_chat_completion import mistral_text_response
from tenacity import (
retry,
@@ -17,7 +18,7 @@ from tenacity import (
@retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
def gemini_text_response(prompt):
""" Provide a programming blog and get code exmaples."""
""" Common functiont to get response from gemini pro Text. """
genai.configure(api_key=os.getenv('GEMINI_API_KEY'))
# Set up the model
@@ -25,14 +26,17 @@ def gemini_text_response(prompt):
"temperature": 1,
"top_p": 1,
"top_k": 1,
"max_output_tokens": 4096,
"max_output_tokens": 6096,
}
model = genai.GenerativeModel(model_name="gemini-pro", generation_config=generation_config)
try:
response = model.generate_content(prompt)
except Exception as err:
logger.error(f"Failed to get response from Gemini: {err}. Retrying..")
gemini_research_report(query)
except Exception as err:
logger.error(f"Failed to get response from Gemini: {err}. Retrying.")
# Try with minstral.
print(f"\n\n\n--MINSTRAL--\n\n\n\n")
response = mistral_text_response(prompt)
return response
return response.text

40
lib/gpt_providers/mistral_chat_completion.py Normal file
View File

@@ -0,0 +1,40 @@
import os
import logging
from pathlib import Path
from mistralai.client import MistralClient
from mistralai.models.chat_completion import ChatMessage
logging.basicConfig(level=logging.INFO, format='%(asctime)s-%(levelname)s-%(module)s-%(lineno)d-%(message)s')
from dotenv import load_dotenv
load_dotenv(Path('../../.env'))
from tenacity import (
retry,
stop_after_attempt,
wait_random_exponential,
) # for exponential backoff
@retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
def mistral_text_response(prompt):
""" Common function to get text response from minstral. """
api_key = os.environ["MISTRAL_API_KEY"]
model = "mistral-medium"
client = MistralClient(api_key=api_key)
messages = [
ChatMessage(role="user", content=prompt)
]
# No streaming
chat_response = client.chat(
model=model,
messages=messages,
)
print(chat_response)
# With streaming
for chunk in client.chat_stream(model=model, messages=messages):
print(chunk)

209
lib/main_arxiv_to_blog.py Normal file
View File

@@ -0,0 +1,209 @@
import sys
import os
import datetime
import tiktoken
from .arxiv_schlorly_research import fetch_arxiv_data, create_dataframe, get_arxiv_main_content
from .arxiv_schlorly_research import arxiv_bibtex, scrape_images_from_arxiv, download_image
from .arxiv_schlorly_research import read_written_ids, extract_arxiv_ids_from_line, append_id_to_file
from .write_research_review_blog import review_research_paper
from .combine_research_and_blog import blog_with_research
from .write_blog_scholar_paper import write_blog_from_paper
from .gpt_providers.gemini_pro_text import gemini_text_response
from .generate_image_from_prompt import generate_image
from .convert_content_to_markdown import convert_tomarkdown_format
from .get_blog_metadata import blog_metadata
from .get_code_examples import gemini_get_code_samples
from .save_blog_to_file import save_blog_to_file
from .take_url_screenshot import screenshot_api
from loguru import logger
logger.remove()
logger.add(sys.stdout,
colorize=True,
format="<level>{level}</level>|<green>{file}:{line}:{function}</green>| {message}"
)
def blog_arxiv_keyword(query):
""" Write blog on given arxiv paper."""
arxiv_id = None
arxiv_url = None
bibtex = None
research_review = None
column_names = ['Title', 'Date', 'Id', 'Summary', 'PDF URL']
papers = fetch_arxiv_data(query)
df = create_dataframe(papers, column_names)
for paper in papers:
# Extracting the arxiv_id
arxiv_id = paper[2].split('/')[-1]
arxiv_url = "https://browse.arxiv.org/html/" + arxiv_id
bibtex = arxiv_bibtex(arxiv_id)
logger.info(f"Get research paper text from the url: {arxiv_url}")
research_content = get_arxiv_main_content(arxiv_url)
num_tokens = num_tokens_from_string(research_content, "cl100k_base")
logger.info(f"Number of tokens sent: {num_tokens}")
# If the number of tokens is below the threshold, process and print the review
if 1000 < num_tokens < 30000:
logger.info(f"Writing research review on {paper[0]}")
research_review = review_research_paper(research_content)
research_review = f"\n{research_review}\n\n" + f"```{bibtex}```"
#research_review = research_review + "\n\n\n" + f"{df.to_markdown()}"
research_review = convert_tomarkdown_format(research_review, "gemini")
break
else:
# Skip to the next iteration if the condition is not met
continue
logger.info(f"Final scholar article: \n\n{research_review}\n")
# TBD: Scrape images from research reports and pass to vision to get conclusions out of it.
#image_urls = scrape_images_from_arxiv(arxiv_url)
#print("Downloading images found on the page:")
#for img_url in image_urls:
# download_image(img_url, arxiv_url)
try:
blog_postprocessing(arxiv_id, research_review)
except Exception as err:
logger.error(f"Failed in blog post processing: {err}")
sys.exit(1)
logger.info(f"\n\n ################ Finished writing Blog for : #################### \n")
def blog_arxiv_url_list(file_path):
""" Write blogs on all the arxiv links given in a file. """
extracted_ids = []
try:
with open(file_path, 'r') as file:
for line in file:
arxiv_id = extract_arxiv_ids_from_line(line)
if arxiv_id:
extracted_ids.append(arxiv_id)
except FileNotFoundError:
logger.error(f"File not found: {file_path}")
raise FileNotFoundError
except Exception as e:
logger.error(f"Error while reading the file: {e}")
raise e
# Read already written IDs
written_ids = read_written_ids('papers_already_written_on.txt')
# Write blogs on each of arxiv_id from the file.
for arxiv_id in extracted_ids:
# Check if we have already written on this research_paper. For this, all arxiv ids are written in
# a file called 'papers_already_written_on.txt'. If arxiv ID is found in this file, skip writing again.
# YUP, use a DB. KISS for now.
written_ids = read_written_ids('papers_already_written_on.txt')
# Loop through extracted IDs
for arxiv_id in extracted_ids:
if arxiv_id not in written_ids:
# This ID has not been written on yet
arxiv_url = "https://browse.arxiv.org/html/" + arxiv_id
logger.info(f"Get research paper text from the url: {arxiv_url}")
research_content = get_arxiv_main_content(arxiv_url)
try:
num_tokens = num_tokens_from_string(research_content, "cl100k_base")
except Exception as err:
logger.error(f"Failed in counting tokens: {err}")
sys.exit(1)
logger.info(f"Number of tokens sent: {num_tokens}")
# If the number of tokens is below the threshold, process and print the review
# FIXME: Docs over 30k tokens, need to be chunked and summarized.
if 1000 < num_tokens < 30000:
try:
logger.info(f"Getting bibtex for arxiv ID: {arxiv_id}")
bibtex = arxiv_bibtex(arxiv_id)
except Exception as err:
logger.error(f"Failed to get Bibtex: {err}")
try:
logger.info(f"Writing a research review..")
research_review = review_research_paper(research_content, "gemini")
logger.info(f"Research Review: \n{research_review}\n\n")
except Exception as err:
logger.error(f"Failed to write review on research paper: {arxiv_id}{err}")
research_blog = write_blog_from_paper(research_content, "gemini")
logger.info(f"\n\nResearch Blog: {research_blog}\n\n")
research_blog = f"\n{research_review}\n\n" + f"```\n{bibtex}\n```"
#research_review = blog_with_research(research_review, research_blog, "gemini")
#logger.info(f"\n\n\nBLOG_WITH_RESEARCh: {research_review}\n\n\n")
research_review = convert_tomarkdown_format(research_review, "gemini")
research_review = f"\n{research_review}\n\n" + f"```{bibtex}```"
logger.info(f"Final blog from research paper: \n\n{research_review}\n\n\n")
try:
blog_postprocessing(arxiv_id, research_review)
except Exception as err:
logger.error(f"Failed in blog post processing: {err}")
sys.exit(1)
logger.info(f"\n\n ################ Finished writing Blog for : #################### \n")
else:
# Skip to the next iteration if the condition is not met
logger.error("FIXME: Docs over 30k tokens, need to be chunked and summarized.")
continue
else:
logger.warning(f"Already written, skip writing on Arxiv paper ID: {arxiv_id}")
def blog_postprocessing(arxiv_id, research_review):
""" Common function to do blog postprocessing. """
try:
append_id_to_file(arxiv_id, "papers_already_written_on.txt")
except Exception as err:
logger.error(f"Failed to write/append ID to papers_already_written_on.txt: {err}")
raise err
try:
blog_title, blog_meta_desc, blog_tags, blog_categories = blog_metadata(research_review, "gemini")
except Exception as err:
logger.error(f"Failed to get blog metadata: {err}")
raise err
try:
arxiv_url_scrnsht = f"https://arxiv.org/abs/{arxiv_id}"
generated_image_filepath = take_paper_screenshot(arxiv_url_scrnsht)
except Exception as err:
logger.error(f"Failed to tsk paper screenshot: {err}")
raise err
try:
save_blog_to_file(research_review, blog_title, blog_meta_desc, blog_tags,\
blog_categories, generated_image_filepath)
except Exception as err:
logger.__repr__ror(f"Failed to save blog to a file: {err}")
raise err
def take_paper_screenshot(arxiv_url):
""" Common function to take paper screenshot. """
# fixme: Remove the hardcoding, need add another option OR in config ?
image_dir = os.path.join(os.getcwd(), "blog_images")
generated_image_name = f"generated_image_{datetime.datetime.now():%Y-%m-%d-%H-%M-%S}.png"
generated_image_filepath = os.path.join(image_dir, generated_image_name)
if arxiv_url:
try:
generated_image_filepath = screenshot_api(arxiv_url, generated_image_filepath)
except Exception as err:
logger.error(f"Failed in taking url screenshot: {err}")
return generated_image_filepath
def num_tokens_from_string(string, encoding_name):
"""Returns the number of tokens in a text string."""
try:
encoding = tiktoken.get_encoding(encoding_name)
num_tokens = len(encoding.encode(string))
return num_tokens
except Exception as err:
logger.error(f"Failed to count tokens: {err}")
sys.exit(1)

24
lib/save_blog_to_file.py
View File

@@ -2,6 +2,8 @@ import sys
import os
import re
import datetime
import random
from dateutil.relativedelta import relativedelta
from textwrap import dedent
import logging
from zoneinfo import ZoneInfo
@@ -22,6 +24,18 @@ output_path = "blogs"
output_path = os.path.join(os.getcwd(), output_path)
def random_date_last_three_months():
current_date = datetime.datetime.now(ZoneInfo('Asia/Kolkata'))
three_months_ago = current_date - relativedelta(months=3)
# Generate a random date between three_months_ago and current_date
random_date = three_months_ago + datetime.timedelta(
seconds=random.randint(0, int((current_date - three_months_ago).total_seconds()))
)
return random_date.strftime('%Y-%m-%d %H:%M:%S %z')
def save_blog_to_file(blog_content, blog_title, blog_meta_desc, blog_tags, blog_categories, main_img_path=None, file_type="md"):
"""
Saves the provided blog content to a file in the specified format.
@@ -60,9 +74,11 @@ def save_blog_to_file(blog_content, blog_title, blog_meta_desc, blog_tags, blog_
# Handle Markdown file type
if file_type == "md":
logger.info("Writing/Saving the resultant blog content in Markdown format.")
dtobj = datetime.datetime.now(ZoneInfo('Asia/Kolkata'))
formatted_date = dtobj.strftime('%Y-%m-%d %H:%M:%S %z')
blog_title = blog_title.replace(":", "-").replace('"', '')
# Hmmmm, bulk generation will benefit from randomizing publishing dates.
#dtobj = datetime.datetime.now(ZoneInfo('Asia/Kolkata'))
#formatted_date = dtobj.strftime('%Y-%m-%d %H:%M:%S %z')
formatted_date = random_date_last_three_months()
blog_title = blog_title.replace(":", "-").replace('"', '').replace('**', '')
if main_img_path:
blog_frontmatter = dedent(f"""\
---
@@ -70,7 +86,7 @@ def save_blog_to_file(blog_content, blog_title, blog_meta_desc, blog_tags, blog_
date: {formatted_date}
categories: [{blog_categories}]
tags: [{blog_tags}]
description: {blog_meta_desc.replace(":", "-")}
description: {blog_meta_desc.replace(":", "-").replace('**', '')}
img_path: '/assets/'
image:
path: {os.path.basename(main_img_path)}

18
lib/take_url_screenshot.py
View File

@@ -44,27 +44,15 @@ def screenshot_api(url, generated_image_filepath):
return generated_image_filepath
def take_screenshot(url, generated_image_filepath, full_screenshot):
def take_screenshot(url, generated_image_filepath):
# Create a webdriver instance
driver = webdriver.Chrome()
# Navigate to the given url
driver.get(url)
# Get the height of the webpage
page_height = driver.execute_script("return document.body.scrollHeight")
# Scroll down to the bottom of the webpage
for i in range(0, page_height, 100):
driver.execute_script(f"window.scrollTo(0, {i})")
# Get the total height of the webpage
total_height = driver.execute_script("return document.body.scrollHeight")
# Resize the webdriver window to the height of the webpage
if full_screenshot:
driver.set_window_size(800, total_height)
# Set a fixed window size (you can adjust this as needed)
driver.set_window_size(800, 600)
# Take a screenshot of the webpage
screenshot = driver.get_screenshot_as_png()

49
lib/write_blog_scholar_paper.py Normal file
View File

@@ -0,0 +1,49 @@
import sys
from .gpt_providers.openai_chat_completion import openai_chatgpt
from .gpt_providers.gemini_pro_text import gemini_text_response
from loguru import logger
logger.remove()
logger.add(sys.stdout,
colorize=True,
format="<level>{level}</level>|<green>{file}:{line}:{function}</green>| {message}"
)
def write_blog_from_paper(paper_content, gpt_providers="openai"):
""" Write blog from given paper url. """
prompt = f"""As an expert in NLP and AI, I will provide you with a content of a research paper.
Your task is to write a highly detailed blog(at least 2000 words), breaking down complex concepts for beginners.
Take your time and do not rush to respond.
Do not provide explanations, suggestions in your response.
Include the below section in your blog:
Highlights: Include a list of 5 most important and unique claims of the given research paper.
Abstract: Start by reading the abstract, which provides a concise summary of the research, including its purpose, methodology, and key findings.
Introduction: This section will give you background information and set the context for the research. It often ends with a statement of the research question or hypothesis.
Methodology: Include description of how authors conducted the research. This can include data sources, experimental setup, analytical techniques, etc.
Results: This section presents the data or findings of the research. Pay attention to figures, tables, and any statistical analysis provided.
Discussion/Analysis: In this section, Explain how research paper answers the research questions or how they fit with existing knowledge.
Conclusion: This part summarizes the main findings and their implications. It might also suggest areas for further research.
References: The cited works can provide additional context or background reading.
Remember, Please use MLA format and markdown syntax.
Do not provide description, explanations for your response.
Take your time in crafting your blog content, do not rush to give the response.
Using the blog structure above, please write a detailed and original blog on given research paper: \n'{paper_content}'\n\n"""
if 'gemini' in gpt_providers:
try:
response = gemini_text_response(prompt)
return response
except Exception as err:
logger.error(f"Failed to get response from gemini: {err}")
raise err
elif 'openai' in gpt_providers:
try:
logger.info("Calling OpenAI LLM.")
response = openai_chatgpt(prompt)
return response
except Exception as err:
logger.error(f"failed to get response from Openai: {err}")
raise err

89
lib/write_research_review_blog.py Normal file
View File

@@ -0,0 +1,89 @@
import sys
from .gpt_providers.openai_chat_completion import openai_chatgpt
from .gpt_providers.gemini_pro_text import gemini_text_response
from .gpt_providers.mistral_chat_completion import mistral_text_response
from loguru import logger
logger.remove()
logger.add(sys.stdout,
colorize=True,
format="<level>{level}</level>|<green>{file}:{line}:{function}</green>| {message}"
)
def review_research_paper(research_blog, gpt_providers="openai"):
""" """
prompt = f"""As world's top researcher and academician, I will provide you with research paper.
Your task is to write a highly detailed review report.
Important, your report should be factual, original and demostrate your expertise.
Review guidelines:
1). Read the Abstract and Introduction Carefully:
Begin by thoroughly reading the abstract and introduction of the paper.
Try to understand the research question, the objectives, and the background information.
Identify the central argument or hypothesis that the study is examining.
2). Examine the Methodology and Methods:
Read closely at the research design, whether it is experimental, observational, qualitative, or a combination of methods.
Check the sampling strategy and the size of the sample.
Review the methods of data collection and the instruments used for this purpose.
Think about any ethical issues and possible biases in the study.
3). Analyze the Results and Discussion:
Review how the results are presented, including any tables, graphs, and statistical analysis.
Evaluate the findings' validity and reliability.
Analyze whether the results support or contradict the research question and hypothesis.
Read the discussion section where the authors interpret their findings and their significance.
4). Consider the Limitations and Strengths:
Spot any limitations or potential weaknesses in the study.
Evaluate the strengths and contributions that the research makes.
Think about how generalizable the findings are to other populations or situations.
5). Assess the Writing and Organization:
Judge the clarity and structure of the report.
Consider the use of language, grammar, and the overall formatting.
Assess how well the arguments are logically organized and how coherent the report is.
6). Evaluate the Literature Review:
Examine how comprehensive and relevant the literature review is.
Consider how the study adds to or builds upon existing research.
Evaluate the timeliness and quality of the sources cited in the research.
7). Review the Conclusion and Implications:
Look at the conclusions drawn from the study and how well they align with the findings.
Think about the practical implications and potential applications of the research.
Evaluate the suggestions for further research or policy actions.
8). Overall Assessment:
Formulate an overall opinion about the research report's quality and thoroughness.
Consider the significance and impact of the findings.
Evaluate how the study contributes to its field of research.
9). Provide Constructive Feedback:
Offer constructive criticism and suggestions for improvement, where necessary.
Think about possible biases or alternative ways to interpret the findings.
Suggest ideas for future research or for replicating the study.
Do not provide description, explanations for your response.
Using the above review guidelines, write a detailed review report on the below research paper.
Research Paper: '{research_blog}'
"""
if 'gemini' in gpt_providers:
try:
response = gemini_text_response(prompt)
return response
except Exception as err:
logger.error(f"Failed to get response from gemini: {err}")
response = mistral_text_response(prompt)
return response
elif 'openai' in gpt_providers:
try:
logger.info("Calling OpenAI LLM.")
response = openai_chatgpt(prompt)
return response
except Exception as err:
SystemError(f"Failed to get response from Openai: {err}")