Spaces:
Sleeping
Sleeping
| import os | |
| # Ensure Java is installed in Hugging Face Space | |
| if not os.path.exists("/usr/bin/java"): | |
| print("🔄 Installing Java manually...") | |
| os.system("wget https://download.java.net/openjdk/jdk11/ri/openjdk-11+28_linux-x64_bin.tar.gz") | |
| os.system("tar -xvzf openjdk-11+28_linux-x64_bin.tar.gz") | |
| os.system("mv jdk-11 java") | |
| os.environ["JAVA_HOME"] = os.path.abspath("java") | |
| os.environ["PATH"] += os.pathsep + os.path.join(os.path.abspath("java"), "bin") | |
| print("✅ Java is installed!") | |
| import language_tool_python | |
| # Initialize LanguageTool | |
| tool = language_tool_python.LanguageTool('en-US') | |
| import gradio as gr | |
| import ftfy | |
| import language_tool_python | |
| import re | |
| import torch | |
| from sentence_transformers import SentenceTransformer, util | |
| from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer | |
| device = "cuda" if torch.cuda.is_available() else "cpu" # ✅ Use GPU if available | |
| # Load fine-tuned GPT-2 model | |
| model_path = "cpv2280/gpt2-tinystories-generator" # Update if needed | |
| model = AutoModelForCausalLM.from_pretrained(model_path) | |
| tokenizer = AutoTokenizer.from_pretrained(model_path) | |
| # Create a text-generation pipeline | |
| story_generator = pipeline("text-generation", model=model, tokenizer=tokenizer) | |
| # Load NLP tools | |
| tool = language_tool_python.LanguageTool('en-UK') | |
| sentence_model = SentenceTransformer('all-MiniLM-L6-v2') | |
| def refine_story(text): | |
| """Refines the generated story by fixing encoding, grammar, and redundancy.""" | |
| text = ftfy.fix_text(text) # Fix encoding | |
| matches = tool.check(text) # Check grammar | |
| text = language_tool_python.utils.correct(text, matches) # Apply fixes | |
| # Remove redundant words/phrases | |
| text = re.sub(r'(\b\w+\b) \1', r'\1', text) # Remove duplicate words | |
| text = re.sub(r'(\b\w+ and \w+\b)(,? \1)+', r'\1', text) # Remove phrase repetitions | |
| return text | |
| def detect_inconsistencies(text): | |
| """Checks for logical inconsistencies by comparing sentence similarities.""" | |
| sentences = text.split(". ") | |
| inconsistencies = [] | |
| # Compare each sentence with the next one | |
| for i in range(len(sentences) - 1): | |
| similarity_score = util.pytorch_cos_sim(sentence_model.encode(sentences[i]), sentence_model.encode(sentences[i+1])) | |
| if similarity_score.item() < 0.3: # If similarity is low, flag as inconsistent | |
| inconsistencies.append(f"⚠️ **Possible inconsistency detected:**\n➡ {sentences[i]} \n➡ {sentences[i+1]}") | |
| return "\n\n".join(inconsistencies) if inconsistencies else "✅ No major inconsistencies detected." | |
| def story_pipeline(prompt): | |
| """Generates a story, refines it, and checks inconsistencies.""" | |
| # Generate the story | |
| generated = story_generator(prompt, max_length=200, do_sample=True, temperature=1.0, top_p=0.9, top_k=50, truncation=True) | |
| raw_story = generated[0]['generated_text'] | |
| # Refine the generated story | |
| refined_story = refine_story(raw_story) | |
| # Detect logical inconsistencies | |
| inconsistencies = detect_inconsistencies(refined_story) | |
| return raw_story, refined_story, inconsistencies | |
| # ✅ Gradio Interface with Proper Logical Inconsistency Detection | |
| interface = gr.Interface( | |
| fn=story_pipeline, | |
| inputs=gr.Textbox(label="Enter Story Prompt", placeholder="Once upon a time..."), | |
| outputs=[ | |
| gr.Textbox(label="📖 Generated Story", interactive=True), # Interactive textbox | |
| gr.Textbox(label="✅ Refined Story", interactive=True), # Refined output | |
| gr.Textbox(label="⚠️ Logical Inconsistencies", interactive=False), # Shows inconsistencies correctly | |
| ], | |
| title="📖 FableWeaver AI", | |
| description="Generates AI-powered TinyStories using GPT-2 fine-tuned on TinyStories. Automatically refines the story and detects logical inconsistencies." | |
| ) | |
| # Launch Gradio app | |
| interface.launch(share="True") | |