Jawi OCR Qwen v2

--- library_name: transformers metrics: - CER: 0.0302592098 language: - ms license: apache-2.0 base_model: Qwen/Qwen2.5-VL-3B-Instruct tags: - vision - ocr - jawi - image-to-text pipeline_tag: image-to-text --- # Jawi-OCR-Qwen-v2 This model is a fine-tuned version of Qwen2.5-VL-3B-Instruct for optical character recognition of Jawi script. Jawi is an Arabic-based writing system traditionally used for writing Malay and other languages in Southeast Asia. ## Usage ``` from transformers import AutoModelForImageTextToText, AutoProcessor import torch from PIL import Image from datasets import load_dataset import pandas as pd prompt = "Transcribe the Jawi script in this image into Jawi text" model = AutoModelForImageTextToText.from_pretrained("culturalheritagenus/Jawi-OCR-Qwen-v2") model = model.to("cuda" if torch.cuda.is_available() else "cpu") processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-3B-Instruct") def process_image(image): try: messages = [ { "role": "user", "content": [ {"type": "image", "image": image}, {"type": "text", "text": prompt} ] }, ] inputs = processor.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) generated_ids = model.generate( **inputs, max_new_tokens=128, do_sample=False, temperature=1.0, pad_token_id=processor.tokenizer.pad_token_id, eos_token_id=processor.tokenizer.eos_token_id ) generated_ids_trimmed = [ out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs["input_ids"], generated_ids) ] generated_text = processor.tokenizer.batch_decode(generated_ids_trimmed, skip_special_tokens=True)[0] return generated_text except Exception as e: print(f"Error during generation: {e}") return "Error: Could not process image" ``` Running inference at scale, assuming an HF-style dataset `ds`. ``` for i, row in enumerate(ds): img: Image.Image = row["Image"] # PIL image from HF dataset identifier = row["Identifier"] text = row["Text"] buffered = BytesIO() img.save(buffered, format="PNG") img_b64 = base64.b64encode(buffered.getvalue()).decode("utf-8") jawi = process_image(img_b64) ```

Source context: 39 downloads · 1 likes · Pipeline image-to-text · Library transformers · Repo culturalheritagenus/Jawi-OCR-Qwen-v2