import gradio as gr from transformers import AutoTokenizer, AutoModelForCausalLM import spacy from sumy.parsers.plaintext import PlaintextParser from sumy.nlp.tokenizers import Tokenizer from sumy.summarizers.text_rank import TextRankSummarizer from detoxify import Detoxify from textblob import TextBlob from gramformer import Gramformer # Grammar correction import torch # Replace with your Hugging Face token # Load GPT-NeoX model and tokenizer tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neo-1.3B") model = AutoModelForCausalLM.from_pretrained("EleutherAI/gpt-neo-1.3B") # Initialize the sentiment analysis pipeline with DistilBERT sentiment_model = "distilbert-base-uncased" sentiment_analyzer = pipeline("sentiment-analysis", model=sentiment_model) nlp = spacy.load("en_core_web_sm") # Load Grammar Correction Model gf = Gramformer(models=1, use_gpu=torch.cuda.is_available()) # Bias Terms List (Can be expanded) BIAS_TERMS = ["he/she", "rich/poor", "race", "gender", "disabled", "poverty"] # Advanced Refinement - TextRank Summarization def summarize_text(text): parser = PlaintextParser.from_string(text, Tokenizer("english")) summarizer = TextRankSummarizer() summary = summarizer(parser.document, 3) # Extract top 3 sentences return " ".join(str(sentence) for sentence in summary) # Grammar Correction using Gramformer def correct_grammar(text): corrected = gf.correct(text, max_candidates=1) return list(corrected)[0] if corrected else text # Detect Toxic Content (Bias and Offensive Language) def detect_bias(text): toxic_score = Detoxify("original").predict(text) bias_flag = any(term in text.lower() for term in BIAS_TERMS) if toxic_score["toxicity"] > 0.3 or bias_flag: return "âš ī¸ Potential bias/toxicity detected!" return "✅ Content is neutral and inclusive." # Sentiment Analysis for Content Evaluation using DistilBERT for sentiment analysis def analyze_sentiment(text): sentiment = sentiment_analyzer(text) label = sentiment[0]['label'] if label == "POSITIVE": return "🙂 Positive" elif label == "NEGATIVE": return "â˜šī¸ Negative" else: return "😐 Neutral" # Content Generation Pipeline class TextGenerationPipeline: def __init__(self): self.tokenizer = tokenizer self.model = model self.nlp = nlp def generate(self, prompt): """Generates educational content based on user input.""" inputs = self.tokenizer(prompt, return_tensors="pt") outputs = self.model.generate(inputs.input_ids, max_length=250, num_return_sequences=1) generated_text = self.tokenizer.decode(outputs[0], skip_special_tokens=True) return generated_text def refine(self, text): """Refines content using summarization, grammar correction, and readability fixes.""" summarized = summarize_text(text) corrected = correct_grammar(summarized) return corrected def detect_bias(self, text): """Detects and flags biased content.""" return detect_bias(text) def evaluate(self, text): """Evaluates the content sentiment and readability.""" sentiment = analyze_sentiment(text) return sentiment def run(self, prompt): """Runs the full pipeline.""" generated = self.generate(prompt) refined = self.refine(generated) bias_check = self.detect_bias(refined) sentiment = self.evaluate(refined) return generated, refined, bias_check, sentiment # Initialize Pipeline pipeline = TextGenerationPipeline() # Build Gradio UI with gr.Blocks() as demo: gr.Markdown("# 🚀 GPT-NeoX-20B Educational Content Generator") gr.Markdown("### AI-powered content generation with bias detection and advanced refinement.") prompt_input = gr.Textbox(label="Enter your prompt:") generate_btn = gr.Button("Generate Content") output_text = gr.Textbox(label="Generated Content", interactive=False) refined_text = gr.Textbox(label="Refined Content", interactive=False) bias_check = gr.Textbox(label="Bias Check Result", interactive=False) sentiment_analysis = gr.Textbox(label="Sentiment Analysis", interactive=False) generate_btn.click(pipeline.run, inputs=[prompt_input], outputs=[output_text, refined_text, bias_check, sentiment_analysis]) demo.launch()