Update app.py

app.py

@@ -1,7 +1,19 @@
-# app.py
-import os, json
+import os
+import json
 import gradio as gr
 
+# Ensure default config for magic-pdf
+CONFIG_PATH = os.path.expanduser("~/magic-pdf.json")
+if not os.path.exists(CONFIG_PATH):
+    default_cfg = {
+        "device": "cpu",      # force CPU inference
+        "layout_model": "layout/mobilenetv3",
+        "formula_enable": True,
+        "table_enable": True
+    }
+    with open(CONFIG_PATH, "w", encoding="utf-8") as cfg:
+        json.dump(default_cfg, cfg, ensure_ascii=False, indent=2)
+
 # MinerU API imports
 from magic_pdf.data.read_api import read_local_pdfs
 from magic_pdf.data.data_reader_writer import FileBasedDataWriter
@@ -10,9 +22,8 @@ from magic_pdf.config.enums import SupportedPdfParseMethod
 
 def convert_with_mineru(pdf_path, out_format):
     # 1) Read file into MinerU dataset(s)
-    datasets = read_local_pdfs(pdf_path)  # returns list[PymuDocDataset] :contentReference[oaicite:3]{index=3}
+    datasets = read_local_pdfs(pdf_path)
 
-    # Prepare writers
     tmp_dir = "output"
     img_dir = os.path.join(tmp_dir, "images")
     os.makedirs(img_dir, exist_ok=True)
@@ -22,22 +33,21 @@ def convert_with_mineru(pdf_path, out_format):
     all_pages = []
 
     for ds in datasets:
-        # 2) Classify & infer
-        if ds.classify() == SupportedPdfParseMethod.OCR:
-            infer = ds.apply(doc_analyze, ocr=True)
-            pipe = infer.pipe_ocr_mode(img_writer)
-        else:
-            infer = ds.apply(doc_analyze, ocr=False)
-            pipe = infer.pipe_txt_mode(img_writer)
+        # 2) Classify & infer, with OCR fallback
+        method = ds.classify()
+        infer = ds.apply(doc_analyze, ocr=(method == SupportedPdfParseMethod.OCR))
+        pipe = (infer.pipe_ocr_mode(img_writer)
+                if method == SupportedPdfParseMethod.OCR
+                else infer.pipe_txt_mode(img_writer))
 
-        # 3) Dump per‐document Markdown + collect
+        # 3) Dump per‐document Markdown
         basename = os.path.splitext(os.path.basename(pdf_path))[0]
         md_fname = f"{basename}.md"
         pipe.dump_md(md_writer, md_fname, os.path.basename(img_dir))
         with open(os.path.join(tmp_dir, md_fname), "r", encoding="utf-8") as f:
             page_md = f.read()
 
-        # 4) Collect structured JSON (middle JSON)
+        # 4) Dump structured JSON
         json_fname = f"{basename}_content_list.json"
         pipe.dump_content_list(md_writer, json_fname, os.path.basename(img_dir))
         with open(os.path.join(tmp_dir, json_fname), "r", encoding="utf-8") as f:
@@ -50,7 +60,6 @@ def convert_with_mineru(pdf_path, out_format):
 
     # 5) Return desired format
     if out_format == "markdown":
-        # Concatenate all documents
         return "\n\n---\n\n".join(p["markdown"] for p in all_pages)
     else:
         return json.dumps(all_pages, ensure_ascii=False, indent=2)
@@ -60,13 +69,14 @@ demo = gr.Interface(
     fn=convert_with_mineru,
     inputs=[
         gr.File(label="Upload PDF", file_types=[".pdf"]),
-        gr.Radio(["markdown","json"], value="markdown", label="Output format")
+        gr.Radio(["markdown", "json"], value="markdown", label="Output format")
     ],
     outputs=gr.Code(label="Result"),
     title="MinerU-Powered PDF → Markdown/JSON",
     description=(
         "Leverage the advanced MinerU engine to extract text, images, tables, "
-        "and formulas from your PDF into clean Markdown or structured JSON."
+        "and formulas from your PDF into clean Markdown or structured JSON. "
+        "A default CPU-only config is auto-generated if none is found."
     )
 )