Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| from llama_cpp import Llama | |
| from paddleocr import PaddleOCR | |
| from PIL import Image | |
| # Load GGUF model | |
| llm = Llama( | |
| model_path="./deepseek-v3-0324.Q4_K_M.gguf", # Make sure this file is in your repo | |
| n_ctx=2048, | |
| n_threads=8, | |
| n_gpu_layers=20 # Set to 0 if you are on CPU-only | |
| ) | |
| # OCR Function | |
| def ocr_inference(img, lang): | |
| ocr = PaddleOCR(use_angle_cls=True, lang=lang, use_gpu=False) | |
| result = ocr.ocr(img, cls=True)[0] | |
| txts = [line[1][0] for line in result] | |
| return " ".join(txts) | |
| # Step 1: Convert text to base form words | |
| def text_inference(text, language): | |
| prompt = ( | |
| f"Given the following {language} text, convert each word into its base form. " | |
| f"Remove all duplicates. Return the base form words as a comma-separated list.\n\n" | |
| f"Text:\n{text}" | |
| ) | |
| response = llm(prompt, max_tokens=256, stop=["</s>"]) | |
| output_text = response["choices"][0]["text"].strip() | |
| words = [w.strip() for w in output_text.split(",") if w.strip()] | |
| return words | |
| # Step 2: Generate flashcards for those words | |
| def make_flashcards(words, language): | |
| prompt = ( | |
| f"For each {language} word in the list, write a flashcard in this format:\n" | |
| f"word - definition - example sentence - translated sentence.\n\n" | |
| f"Words:\n{', '.join(words)}" | |
| ) | |
| response = llm(prompt, max_tokens=512, stop=["</s>"]) | |
| return response["choices"][0]["text"].strip() | |
| # Wrapper logic to handle OCR or text | |
| def flashcard_pipeline(text, image, language): | |
| if image: | |
| text = ocr_inference(image, language) | |
| if not text: | |
| return "", "Please provide either text or an image." | |
| words = text_inference(text, language) | |
| flashcards = make_flashcards(words, language) | |
| return "\n".join(words), flashcards | |
| # Gradio UI | |
| demo = gr.Interface( | |
| fn=flashcard_pipeline, | |
| inputs=[ | |
| gr.Textbox(label="Input Text (leave empty to use image)", lines=4, placeholder="Type or paste sentence here..."), | |
| gr.Image(label="Upload Image for OCR (optional)", type="filepath"), | |
| gr.Dropdown(choices=["korean", "japan", "french", "ch"], label="Language (for OCR and LLM)") | |
| ], | |
| outputs=[ | |
| gr.Textbox(label="Base Form Words"), | |
| gr.Textbox(label="Flashcards"), | |
| ], | |
| title="Language Flashcard Generator (with OCR + DeepSeek GGUF)", | |
| description="Either input text or upload an image. The app will extract words, lemmatize them, and generate flashcards." | |
| ) | |
| if __name__ == "__main__": | |
| demo.launch() | |