import gradio as gr import logging import os import torch from transformers import AutoTokenizer, AutoModelForCausalLM import whisper from pydub import AudioSegment import requests from bs4 import BeautifulSoup from typing import Optional, Dict, Any from dataclasses import dataclass logging.basicConfig( level=logging.INFO, format='%(asctime)s - %(levelname)s) %(message)s' ) logger = logging.getLogger(__name__) @dataclass class NewsConfig: model_name: str = "meta-llama/Llama-2-3b-chat-hf" max_tokens: int = 256 temperature: float = 0.7 top_p: float = 0.95 class NewsGenerator: def __init__(self): self.config = NewsConfig() self.tokenizer = None self.model = None self.whisper_model = None self._initialize_models() def _initialize_models(self): """Initialize models with efficient settings""" try: if not self.tokenizer: self.tokenizer = AutoTokenizer.from_pretrained( self.config.model_name, use_fast=True, model_max_length=self.config.max_tokens ) self.tokenizer.pad_token = self.tokenizer.eos_token if not self.model: self.model = AutoModelForCausalLM.from_pretrained( self.config.model_name, device_map="auto", torch_dtype=torch.float16, low_cpu_mem_usage=True, use_safetensors=True ) if not self.whisper_model: self.whisper_model = whisper.load_model( "tiny", device="cuda" if torch.cuda.is_available() else "cpu" ) except Exception as e: logger.error(f"Error initializing models: {str(e)}") raise def transcribe_audio(self, audio_file: str) -> str: """Transcribe audio file with improved error handling""" try: if not audio_file: return "Error: No audio file provided" result = self.whisper_model.transcribe(audio_file) return result.get("text", "Transcription failed") except Exception as e: logger.error(f"Audio transcription error: {str(e)}") return f"Error transcribing audio: {str(e)}" def generate_news(self, prompt: str) -> str: """Generate news article with optimized parameters""" try: with torch.inference_mode(): outputs = self.model.generate( inputs=self.tokenizer(prompt, return_tensors="pt").input_ids, max_new_tokens=self.config.max_tokens, temperature=self.config.temperature, top_p=self.config.top_p, do_sample=True, early_stopping=True ) return self.tokenizer.decode(outputs[0], skip_special_tokens=True) except Exception as e: logger.error(f"News generation error: {str(e)}") return f"Error generating news: {str(e)}" def read_document(document_path: str) -> str: """Read document content with better error handling""" try: if document_path.endswith(".pdf"): with fitz.open(document_path) as doc: return "\n".join(page.get_text() for page in doc) elif document_path.endswith((".docx", ".xlsx", ".csv")): content = "" if document_path.endswith(".docx"): import docx doc = docx.Document(document_path) content = "\n".join(p.text for p in doc.paragraphs) elif document_path.endswith(".xlsx"): import pandas as pd content = pd.read_excel(document_path).to_string() elif document_path.endswith(".csv"): import pandas as pd content = pd.read_csv(document_path).to_string() return content return "Unsupported file type" except Exception as e: logger.error(f"Document reading error: {str(e)}") return f"Error reading document: {str(e)}" def read_url(url: str) -> str: """Read URL content with better handling""" try: response = requests.get(url, timeout=10) response.raise_for_status() return BeautifulSoup(response.content, 'html.parser').get_text() except Exception as e: logger.error(f"URL reading error: {str(e)}") return f"Error reading URL: {str(e)}" def process_social_media(url: str) -> Dict[str, Any]: """Process social media content with improved handling""" try: text = read_url(url) return {"text": text, "video": None} except Exception as e: logger.error(f"Social media processing error: {str(e)}") return {"text": None, "video": None} def main(): """Main function to create and run the Gradio app""" news_generator = NewsGenerator() with gr.Blocks() as demo: gr.Markdown("# Generador de Noticias Optimizado") with gr.Row(): instrucciones = gr.Textbox(label="Instrucciones", lines=2) hechos = gr.Textbox(label="Hechos", lines=4) tamaño = gr.Number(label="Tamaño (palabras)", value=100) tono = gr.Dropdown(label="Tono", choices=["serio", "neutral", "divertido"], value="neutral") with gr.Row(): documento = gr.File(label="Documento", file_types=["pdf", "docx", "xlsx", "csv"]) audio = gr.File(label="Audio/Video", file_types=["audio", "video"]) url = gr.Textbox(label="URL") social_url = gr.Textbox(label="URL de red social") with gr.Row(): generar = gr.Button("Generar Noticia") noticia = gr.Textbox(label="Noticia Generada", lines=20) transcripciones = gr.Textbox(label="Transcripciones", lines=10) def generate_news_output( instrucciones: str, hechos: str, tamaño: int, tono: str, documento: Optional[gr.File], audio: Optional[gr.File], url: Optional[str], social_url: Optional[str] ): try: # Process document if documento: doc_content = read_document(documento.name) else: doc_content = "" # Process audio if audio: audio_content = news_generator.transcribe_audio(audio.name) else: audio_content = "" # Process URL if url: url_content = read_url(url) else: url_content = "" # Process social media if social_url: social_content = process_social_media(social_url) else: social_content = {"text": "", "video": ""} # Generate prompt prompt = f"""[INST] Escribe una noticia basada en la siguiente información: Instrucciones: {instrucciones} Hechos: {hechos} Documento: {doc_content} Audio: {audio_content} URL: {url_content} Red Social: {social_content['text']} Video: {social_content['video'] if social_content else ''} Parámetros: - Tamaño: {tamaño} palabras - Tono: {tono} - Incluye: Título, gancho, cuerpo, 5W - Estilo periodístico [/INST]""" # Generate news news = news_generator.generate_news(prompt) return news, f"Transcripciones generadas correctamente" except Exception as e: return f"Error generando noticia: {str(e)}", f"Error: {str(e)}" generate_news_output( instrucciones, hechos, tamaño, tono, documento, audio, url, social_url )(generar, [noticia, transcripciones]) if __name__ == "__main__": demo.launch() if __name__ == "__main__": main()