Refa: treat MinerU as an OCR model (#11849)

### What problem does this PR solve?

 Treat MinerU as an OCR model.

### Type of change

- [x] New Feature (non-breaking change which adds functionality)
- [x] Refactoring

rag/app/naive.py

@@ -34,7 +34,6 @@ from rag.utils.file_utils import extract_embed_file, extract_links_from_pdf, ext
 from deepdoc.parser import DocxParser, ExcelParser, HtmlParser, JsonParser, MarkdownElementExtractor, MarkdownParser, PdfParser, TxtParser
 from deepdoc.parser.figure_parser import VisionFigureParser,vision_figure_parser_docx_wrapper,vision_figure_parser_pdf_wrapper
 from deepdoc.parser.pdf_parser import PlainParser, VisionParser
-from deepdoc.parser.mineru_parser import MinerUParser
 from deepdoc.parser.docling_parser import DoclingParser
 from deepdoc.parser.tcadp_parser import TCADPParser
 from rag.nlp import concat_img, find_codec, naive_merge, naive_merge_with_images, naive_merge_docx, rag_tokenizer, tokenize_chunks, tokenize_chunks_with_images, tokenize_table, attach_media_context
@@ -58,27 +57,42 @@ def by_deepdoc(filename, binary=None, from_page=0, to_page=100000, lang="Chinese
 
 
 def by_mineru(filename, binary=None, from_page=0, to_page=100000, lang="Chinese", callback=None, pdf_cls = None ,**kwargs):
-    mineru_executable = os.environ.get("MINERU_EXECUTABLE", "mineru")
-    mineru_api = os.environ.get("MINERU_APISERVER", "http://host.docker.internal:9987")
-    pdf_parser = MinerUParser(mineru_path=mineru_executable, mineru_api=mineru_api)
     parse_method = kwargs.get("parse_method", "raw")
+    mineru_llm_name = kwargs.get("mineru_llm_name")
+    tenant_id = kwargs.get("tenant_id")
 
-    if not pdf_parser.check_installation():
+    pdf_parser = None
+    if tenant_id:
+        if not mineru_llm_name:
+            try:
+                from api.db.services.tenant_llm_service import TenantLLMService
+
+                env_name = TenantLLMService.ensure_mineru_from_env(tenant_id)
+                candidates = TenantLLMService.query(tenant_id=tenant_id, llm_factory="MinerU", model_type=LLMType.OCR.value)
+                if candidates:
+                    mineru_llm_name = candidates[0].llm_name
+                elif env_name:
+                    mineru_llm_name = env_name
+            except Exception as e:  # best-effort fallback
+                logging.warning(f"fallback to env mineru: {e}")
+
+        if mineru_llm_name:
+            try:
+                ocr_model = LLMBundle(tenant_id, LLMType.OCR, llm_name=mineru_llm_name, lang=lang)
+                pdf_parser = ocr_model.mdl
+                sections, tables = pdf_parser.parse_pdf(
+                    filepath=filename,
+                    binary=binary,
+                    callback=callback,
+                    parse_method=parse_method,
+                )
+                return sections, tables, pdf_parser
+            except Exception as e:
+                logging.error(f"Failed to parse pdf via LLMBundle MinerU ({mineru_llm_name}): {e}")
+
+    if callback:
         callback(-1, "MinerU not found.")
-        return None, None, pdf_parser
-
-    sections, tables = pdf_parser.parse_pdf(
-        filepath=filename,
-        binary=binary,
-        callback=callback,
-        output_dir=os.environ.get("MINERU_OUTPUT_DIR", ""),
-        backend=os.environ.get("MINERU_BACKEND", "pipeline"),
-        server_url=os.environ.get("MINERU_SERVER_URL", ""),
-        delete_output=bool(int(os.environ.get("MINERU_DELETE_OUTPUT", 1))),
-        parse_method=parse_method
-    )
-    return sections, tables, pdf_parser
-
+    return None, None, None
 
 def by_docling(filename, binary=None, from_page=0, to_page=100000, lang="Chinese", callback=None, pdf_cls = None ,**kwargs):
     pdf_parser = DoclingParser()
@@ -692,7 +706,15 @@ def chunk(filename, binary=None, from_page=0, to_page=100000, lang="Chinese", ca
         return res
 
     elif re.search(r"\.pdf$", filename, re.IGNORECASE):
-        layout_recognizer = parser_config.get("layout_recognize", "DeepDOC")
+        layout_recognizer_raw = parser_config.get("layout_recognize", "DeepDOC")
+        parser_model_name = None
+        layout_recognizer = layout_recognizer_raw
+        if isinstance(layout_recognizer_raw, str):
+            lowered = layout_recognizer_raw.lower()
+            if lowered.startswith("mineru@"):
+                parser_model_name = layout_recognizer_raw.split("@", 1)[1]
+                layout_recognizer = "MinerU"
+
         if parser_config.get("analyze_hyperlink", False) and is_root:
             urls = extract_links_from_pdf(binary)
 
@@ -711,6 +733,7 @@ def chunk(filename, binary=None, from_page=0, to_page=100000, lang="Chinese", ca
             lang = lang,
             callback = callback,
             layout_recognizer = layout_recognizer,
+            mineru_llm_name = parser_model_name,
             **kwargs
         )