navodit17/Final_Assignment_Template

1from smolagents import Tool, tool2from youtube_transcript_api import YouTubeTranscriptApi3from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline4import torch5 6@tool7def fetch_webpage(url: str, convert_to_markdown: bool = True) -> str:8    """9    Visit a website / url and fetch the content of the webpage. 10    if markdown conversion is enabled, it will remove script and style and return the text content as markdown else return raw unfiltered HTML11    Args:12        url (str): The URL to fetch.13        convert_to_markdown (bool): If True, convert the HTML content to Markdown format. else return the raw HTML.14    Returns:15        str: The HTML content of the URL.16    """17    import requests18    from bs4 import BeautifulSoup19    from markdownify import markdownify as md20 21    content = None22    headers = {23        'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36'24    }25    response = requests.get(url, timeout=30, headers=headers)26    27    if (convert_to_markdown):28        soup = BeautifulSoup(response.text, "html.parser")29 30        # remove script and style tags31        for script in soup(["script", "style"]):32            script.extract()33 34        # for wikipedia only keep the main content35        if "wikipedia.org" in url:36 37            elements_to_remove = [38                # Navigation and reference elements39                {'class': 'navbox'},40                {'class': 'navbox-group'},41                {'class': 'reflist'},42                {'class': 'navigation-box'},43                {'class': 'sister-project'},44                {'class': 'metadata'},45                {'class': 'interlanguage-link'},46                {'class': 'catlinks'},47                {'id': 'References'},48                {'id': 'External_links'},49                {'id': 'Further_reading'},50                {'id': 'See_also'},51                {'id': 'Notes'},52            ]53            54            for selector in elements_to_remove:55                elements = soup.find_all(attrs=selector)56                for element in elements:57                    # For ID-based elements, remove the parent section58                    if 'id' in selector:59                        parent = element.parent60                        if parent and parent.name in ['h2', 'h3', 'h4']:61                            # Remove heading and all content until next heading62                            current = parent63                            while current and current.next_sibling:64                                next_elem = current.next_sibling65                                if (hasattr(next_elem, 'name') and 66                                    next_elem.name in ['h2', 'h3', 'h4']):67                                    break68                                if hasattr(next_elem, 'decompose'):69                                    next_elem.decompose()70                                else:71                                    current = next_elem72                            parent.decompose()73                    else:74                        element.decompose()75 76            main_content = soup.find("main",{"id":"content"})77            if main_content:78                content = md(str(main_content),strip=['script', 'style'], heading_style="ATX").strip()79            else:80                content = md(response.text,strip=['script', 'style'], heading_style="ATX").strip()81        else:82            # Fallback for all other sites - from chatgpt - not tested83            content = md(str(soup), strip=['script', 'style'], heading_style="ATX").strip()84    else:85        content = response.text86           87    return content88 89 90@tool91def read_file_tool(file_path: str) -> str:92    """93    Tool to read a file and return its content.94 95    Args:96        file_path (str): Path to the file to read.97 98    Returns:99        str: Content of the file or error message.100    """101    try:102        with open(file_path, "r") as file:103            return file.read()104    except Exception as e:105        return f"Error reading file: {str(e)}"106    107 108@tool109def get_youtube_transcript(video_id: str) -> str:110    """111    Fetches the transcript of a YouTube video given its video ID.    112    Args:113        video_id (str): The ID of the YouTube video. Pass in the video ID, NOT the video URL. For a video with the URL https://www.youtube.com/watch?v=12345 the ID is 12345.114    Returns:115        str: The transcript of the YouTube video. as a single string with each line separated by a newline character.116    """117    # Initialize the YouTubeTranscriptApi118    ytt_api = YouTubeTranscriptApi()119    fetched_transcript = ytt_api.fetch(video_id)120    raw_data = fetched_transcript.to_raw_data()121    # raw data is in the form of [{        'text': 'Hey there',        'start': 0.0,        'duration': 1.54    },    {        'text': 'how are you',,        'start': 1.54,        'duration': 4.16    },   ... ] we will return ony the text element as lines122    transcript = "\n".join([item['text'] for item in raw_data])123    return transcript124 125 126@tool127def transcribe_audio(audio_path: str) -> str:128    """129    Speech to Text - transcribes audio file and returns the text130 131    Args:132        audio_path (str): Local file path to the audio133 134    Returns:135        str: The transcript of the audio file136    """137    138    device = "cuda:0" if torch.cuda.is_available() else "cpu"139    torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32140    141    model_id = "openai/whisper-small"142    143    model = AutoModelForSpeechSeq2Seq.from_pretrained(144        model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True145    )146    model.to(device)147    148    processor = AutoProcessor.from_pretrained(model_id)149    150    pipe = pipeline(151        "automatic-speech-recognition",152        model=model,153        tokenizer=processor.tokenizer,154        feature_extractor=processor.feature_extractor,155        torch_dtype=torch_dtype,156        device=device,157        chunk_length_s=30,158    )159    160    result = pipe(audio_path)161    return result162