Update app.py

app.py

@@ -14,6 +14,7 @@ from transformers import AutoProcessor, Qwen2VLForConditionalGeneration
 from olmocr.data.renderpdf import render_pdf_to_base64png
 from olmocr.prompts import build_finetuning_prompt
 from olmocr.prompts.anchor import get_anchor_text
+from olmocr.pipeline import PDFToTextOCR  # ✅ Import the OCR pipeline
 
 # Set Hugging Face and Torch cache to a guaranteed-writable location
 cache_dir = "/tmp/huggingface_cache"
@@ -29,6 +30,9 @@ model = Qwen2VLForConditionalGeneration.from_pretrained(
 ).eval().to(device)
 processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-7B-Instruct")
 
+# Initialize olmocr OCR pipeline
+ocr_pipeline = PDFToTextOCR()
+
 def ocr_page(pdf_path, page_num):
     image_b64 = render_pdf_to_base64png(pdf_path, page_num + 1, target_longest_image_dim=1024)
     anchor_text = get_anchor_text(pdf_path, page_num + 1, pdf_engine="pdfreport", target_length=4000)
@@ -89,22 +93,19 @@ def create_epub_from_text(text, output_path, title, author, language, cover_imag
     epub.write_epub(output_path, book)
 
 def convert_pdf_to_epub(pdf_file, title, author, language):
-    # Save the uploaded file to a temporary path
-    tmp_pdf_path = pdf_file.name  # Use the actual temp file path from Gradio
+    tmp_pdf_path = pdf_file.name
 
-    # Now it's safe to read it
+    # Read the first page for cover
     reader = PdfReader(tmp_pdf_path)
-
-    # Extract the first page for the cover (if needed)
-    first_page = reader.pages[0]
     cover_path = "/tmp/cover.jpg"
     images = convert_from_path(tmp_pdf_path, first_page=1, last_page=1)
     images[0].save(cover_path, "JPEG")
 
-    # Run OCR and get text from olmocr
-    ocr_text = olmocr.process(tmp_pdf_path)
+    # Run OCR using olmocr pipeline
+    ocr_result = ocr_pipeline(tmp_pdf_path)
+    ocr_text = "\n\n".join([page.text for page in ocr_result.pages])
 
-    # Use metadata
+    # Create EPUB
     epub_path = "/tmp/output.epub"
     create_epub_from_text(
         text=ocr_text,