mineru 2.6.7__py3-none-any.whl → 2.7.0__py3-none-any.whl

mineru/cli/common.py

@@ -10,12 +10,15 @@ import pypdfium2 as pdfium
 
 from mineru.data.data_reader_writer import FileBasedDataWriter
 from mineru.utils.draw_bbox import draw_layout_bbox, draw_span_bbox, draw_line_sort_bbox
+from mineru.utils.engine_utils import get_vlm_engine
 from mineru.utils.enum_class import MakeMode
 from mineru.utils.guess_suffix_or_lang import guess_suffix_by_bytes
 from mineru.utils.pdf_image_tools import images_bytes_to_pdf_bytes
 from mineru.backend.vlm.vlm_middle_json_mkcontent import union_make as vlm_union_make
 from mineru.backend.vlm.vlm_analyze import doc_analyze as vlm_doc_analyze
 from mineru.backend.vlm.vlm_analyze import aio_doc_analyze as aio_vlm_doc_analyze
+from mineru.backend.hybrid.hybrid_analyze import doc_analyze as hybrid_doc_analyze
+from mineru.backend.hybrid.hybrid_analyze import aio_doc_analyze as aio_hybrid_doc_analyze
 from mineru.utils.pdf_page_id import get_end_page_id
 
 if os.getenv("MINERU_LMDEPLOY_DEVICE", "") == "maca":
@@ -304,6 +307,111 @@ def _process_vlm(
         )
 
 
+def _process_hybrid(
+        output_dir,
+        pdf_file_names,
+        pdf_bytes_list,
+        h_lang_list,
+        parse_method,
+        inline_formula_enable,
+        backend,
+        f_draw_layout_bbox,
+        f_draw_span_bbox,
+        f_dump_md,
+        f_dump_middle_json,
+        f_dump_model_output,
+        f_dump_orig_pdf,
+        f_dump_content_list,
+        f_make_md_mode,
+        server_url=None,
+        **kwargs,
+):
+    """同步处理hybrid后端逻辑"""
+    if not backend.endswith("client"):
+        server_url = None
+
+    for idx, (pdf_bytes, lang) in enumerate(zip(pdf_bytes_list, h_lang_list)):
+        pdf_file_name = pdf_file_names[idx]
+        local_image_dir, local_md_dir = prepare_env(output_dir, pdf_file_name, f"hybrid_{parse_method}")
+        image_writer, md_writer = FileBasedDataWriter(local_image_dir), FileBasedDataWriter(local_md_dir)
+
+        middle_json, infer_result, _vlm_ocr_enable = hybrid_doc_analyze(
+            pdf_bytes,
+            image_writer=image_writer,
+            backend=backend,
+            parse_method=parse_method,
+            language=lang,
+            inline_formula_enable=inline_formula_enable,
+            server_url=server_url,
+            **kwargs,
+        )
+
+        pdf_info = middle_json["pdf_info"]
+
+        # f_draw_span_bbox = not _vlm_ocr_enable
+        f_draw_span_bbox = False
+
+        _process_output(
+            pdf_info, pdf_bytes, pdf_file_name, local_md_dir, local_image_dir,
+            md_writer, f_draw_layout_bbox, f_draw_span_bbox, f_dump_orig_pdf,
+            f_dump_md, f_dump_content_list, f_dump_middle_json, f_dump_model_output,
+            f_make_md_mode, middle_json, infer_result, is_pipeline=False
+        )
+
+
+async def _async_process_hybrid(
+        output_dir,
+        pdf_file_names,
+        pdf_bytes_list,
+        h_lang_list,
+        parse_method,
+        inline_formula_enable,
+        backend,
+        f_draw_layout_bbox,
+        f_draw_span_bbox,
+        f_dump_md,
+        f_dump_middle_json,
+        f_dump_model_output,
+        f_dump_orig_pdf,
+        f_dump_content_list,
+        f_make_md_mode,
+        server_url=None,
+        **kwargs,
+):
+    """异步处理hybrid后端逻辑"""
+
+    if not backend.endswith("client"):
+        server_url = None
+
+    for idx, (pdf_bytes, lang) in enumerate(zip(pdf_bytes_list, h_lang_list)):
+        pdf_file_name = pdf_file_names[idx]
+        local_image_dir, local_md_dir = prepare_env(output_dir, pdf_file_name, f"hybrid_{parse_method}")
+        image_writer, md_writer = FileBasedDataWriter(local_image_dir), FileBasedDataWriter(local_md_dir)
+
+        middle_json, infer_result, _vlm_ocr_enable = await aio_hybrid_doc_analyze(
+            pdf_bytes,
+            image_writer=image_writer,
+            backend=backend,
+            parse_method=parse_method,
+            language=lang,
+            inline_formula_enable=inline_formula_enable,
+            server_url=server_url,
+            **kwargs,
+        )
+
+        pdf_info = middle_json["pdf_info"]
+
+        # f_draw_span_bbox = not _vlm_ocr_enable
+        f_draw_span_bbox = False
+
+        _process_output(
+            pdf_info, pdf_bytes, pdf_file_name, local_md_dir, local_image_dir,
+            md_writer, f_draw_layout_bbox, f_draw_span_bbox, f_dump_orig_pdf,
+            f_dump_md, f_dump_content_list, f_dump_middle_json, f_dump_model_output,
+            f_make_md_mode, middle_json, infer_result, is_pipeline=False
+        )
+
+
 def do_parse(
         output_dir,
         pdf_file_names: list[str],
@@ -340,18 +448,40 @@ def do_parse(
         if backend.startswith("vlm-"):
             backend = backend[4:]
 
-        if backend == "vllm-async-engine":
-            raise Exception("vlm-vllm-async-engine backend is not supported in sync mode, please use vlm-vllm-engine backend")
+            if backend == "vllm-async-engine":
+                raise Exception("vlm-vllm-async-engine backend is not supported in sync mode, please use vlm-vllm-engine backend")
 
-        os.environ['MINERU_VLM_FORMULA_ENABLE'] = str(formula_enable)
-        os.environ['MINERU_VLM_TABLE_ENABLE'] = str(table_enable)
+            if backend == "auto-engine":
+                backend = get_vlm_engine(inference_engine='auto', is_async=False)
 
-        _process_vlm(
-            output_dir, pdf_file_names, pdf_bytes_list, backend,
-            f_draw_layout_bbox, f_draw_span_bbox, f_dump_md, f_dump_middle_json,
-            f_dump_model_output, f_dump_orig_pdf, f_dump_content_list, f_make_md_mode,
-            server_url, **kwargs,
-        )
+            os.environ['MINERU_VLM_FORMULA_ENABLE'] = str(formula_enable)
+            os.environ['MINERU_VLM_TABLE_ENABLE'] = str(table_enable)
+
+            _process_vlm(
+                output_dir, pdf_file_names, pdf_bytes_list, backend,
+                f_draw_layout_bbox, f_draw_span_bbox, f_dump_md, f_dump_middle_json,
+                f_dump_model_output, f_dump_orig_pdf, f_dump_content_list, f_make_md_mode,
+                server_url, **kwargs,
+            )
+        elif backend.startswith("hybrid-"):
+            backend = backend[7:]
+
+            if backend == "vllm-async-engine":
+                raise Exception(
+                    "hybrid-vllm-async-engine backend is not supported in sync mode, please use hybrid-vllm-engine backend")
+
+            if backend == "auto-engine":
+                backend = get_vlm_engine(inference_engine='auto', is_async=False)
+
+            os.environ['MINERU_VLM_TABLE_ENABLE'] = str(table_enable)
+            os.environ['MINERU_VLM_FORMULA_ENABLE'] = "true"
+
+            _process_hybrid(
+                output_dir, pdf_file_names, pdf_bytes_list, p_lang_list, parse_method, formula_enable, backend,
+                f_draw_layout_bbox, f_draw_span_bbox, f_dump_md, f_dump_middle_json,
+                f_dump_model_output, f_dump_orig_pdf, f_dump_content_list, f_make_md_mode,
+                server_url, **kwargs,
+            )
 
 
 async def aio_do_parse(
@@ -391,19 +521,39 @@ async def aio_do_parse(
         if backend.startswith("vlm-"):
             backend = backend[4:]
 
-        if backend == "vllm-engine":
-            raise Exception("vlm-vllm-engine backend is not supported in async mode, please use vlm-vllm-async-engine backend")
+            if backend == "vllm-engine":
+                raise Exception("vlm-vllm-engine backend is not supported in async mode, please use vlm-vllm-async-engine backend")
 
-        os.environ['MINERU_VLM_FORMULA_ENABLE'] = str(formula_enable)
-        os.environ['MINERU_VLM_TABLE_ENABLE'] = str(table_enable)
+            if backend == "auto-engine":
+                backend = get_vlm_engine(inference_engine='auto', is_async=True)
 
-        await _async_process_vlm(
-            output_dir, pdf_file_names, pdf_bytes_list, backend,
-            f_draw_layout_bbox, f_draw_span_bbox, f_dump_md, f_dump_middle_json,
-            f_dump_model_output, f_dump_orig_pdf, f_dump_content_list, f_make_md_mode,
-            server_url, **kwargs,
-        )
+            os.environ['MINERU_VLM_FORMULA_ENABLE'] = str(formula_enable)
+            os.environ['MINERU_VLM_TABLE_ENABLE'] = str(table_enable)
 
+            await _async_process_vlm(
+                output_dir, pdf_file_names, pdf_bytes_list, backend,
+                f_draw_layout_bbox, f_draw_span_bbox, f_dump_md, f_dump_middle_json,
+                f_dump_model_output, f_dump_orig_pdf, f_dump_content_list, f_make_md_mode,
+                server_url, **kwargs,
+            )
+        elif backend.startswith("hybrid-"):
+            backend = backend[7:]
+
+            if backend == "vllm-engine":
+                raise Exception("hybrid-vllm-engine backend is not supported in async mode, please use hybrid-vllm-async-engine backend")
+
+            if backend == "auto-engine":
+                backend = get_vlm_engine(inference_engine='auto', is_async=True)
+
+            os.environ['MINERU_VLM_TABLE_ENABLE'] = str(table_enable)
+            os.environ['MINERU_VLM_FORMULA_ENABLE'] = "true"
+
+            await _async_process_hybrid(
+                output_dir, pdf_file_names, pdf_bytes_list, p_lang_list, parse_method, formula_enable, backend,
+                f_draw_layout_bbox, f_draw_span_bbox, f_dump_md, f_dump_middle_json,
+                f_dump_model_output, f_dump_orig_pdf, f_dump_content_list, f_make_md_mode,
+                server_url, **kwargs,
+            )
 
 
 if __name__ == "__main__":

mineru/cli/fast_api.py

@@ -1,3 +1,4 @@
+import sys
 import uuid
 import os
 import re
@@ -14,6 +15,11 @@ from fastapi.responses import JSONResponse, FileResponse
 from starlette.background import BackgroundTask
 from typing import List, Optional
 from loguru import logger
+
+log_level = os.getenv("MINERU_LOG_LEVEL", "INFO").upper()
+logger.remove()  # 移除默认handler
+logger.add(sys.stderr, level=log_level)  # 添加新handler
+
 from base64 import b64encode
 
 from mineru.cli.common import aio_do_parse, read_fn, pdf_suffixes, image_suffixes
@@ -105,23 +111,38 @@ async def parse_pdf(
         output_dir: str = Form("./output", description="Output local directory"),
         lang_list: List[str] = Form(
             ["ch"],
-            description="""(Adapted only for pipeline backend)Input the languages in the pdf to improve OCR accuracy.
-Options: ch, ch_server, ch_lite, en, korean, japan, chinese_cht, ta, te, ka, th, el, latin, arabic, east_slavic, cyrillic, devanagari.
+            description="""(Adapted only for pipeline and hybrid backend)Input the languages in the pdf to improve OCR accuracy.Options:
+- ch: Chinese, English, Chinese Traditional.
+- ch_lite: Chinese, English, Chinese Traditional, Japanese.
+- ch_server: Chinese, English, Chinese Traditional, Japanese.
+- en: English.
+- korean: Korean, English.
+- japan: Chinese, English, Chinese Traditional, Japanese.
+- chinese_cht: Chinese, English, Chinese Traditional, Japanese.
+- ta: Tamil, English.
+- te: Telugu, English.
+- ka: Kannada.
+- th: Thai, English.
+- el: Greek, English.
+- latin: French, German, Afrikaans, Italian, Spanish, Bosnian, Portuguese, Czech, Welsh, Danish, Estonian, Irish, Croatian, Uzbek, Hungarian, Serbian (Latin), Indonesian, Occitan, Icelandic, Lithuanian, Maori, Malay, Dutch, Norwegian, Polish, Slovak, Slovenian, Albanian, Swedish, Swahili, Tagalog, Turkish, Latin, Azerbaijani, Kurdish, Latvian, Maltese, Pali, Romanian, Vietnamese, Finnish, Basque, Galician, Luxembourgish, Romansh, Catalan, Quechua.
+- arabic: Arabic, Persian, Uyghur, Urdu, Pashto, Kurdish, Sindhi, Balochi, English.
+- east_slavic: Russian, Belarusian, Ukrainian, English.
+- cyrillic: Russian, Belarusian, Ukrainian, Serbian (Cyrillic), Bulgarian, Mongolian, Abkhazian, Adyghe, Kabardian, Avar, Dargin, Ingush, Chechen, Lak, Lezgin, Tabasaran, Kazakh, Kyrgyz, Tajik, Macedonian, Tatar, Chuvash, Bashkir, Malian, Moldovan, Udmurt, Komi, Ossetian, Buryat, Kalmyk, Tuvan, Sakha, Karakalpak, English.
+- devanagari: Hindi, Marathi, Nepali, Bihari, Maithili, Angika, Bhojpuri, Magahi, Santali, Newari, Konkani, Sanskrit, Haryanvi, English.
 """
         ),
         backend: str = Form(
-            "pipeline",
+            "hybrid-auto-engine",
             description="""The backend for parsing:
-- pipeline: More general
-- vlm-transformers: More general, but slower
-- vlm-mlx-engine: Faster than transformers (need apple silicon and macOS 13.5+)
-- vlm-vllm-async-engine: Faster (vllm-engine, need vllm installed)
-- vlm-lmdeploy-engine: Faster (lmdeploy-engine, need lmdeploy installed)
-- vlm-http-client: Faster (client suitable for openai-compatible servers)"""
+- pipeline: More general, supports multiple languages, hallucination-free.
+- vlm-auto-engine: High accuracy via local computing power, supports Chinese and English documents only.
+- vlm-http-client: High accuracy via remote computing power(client suitable for openai-compatible servers), supports Chinese and English documents only.
+- hybrid-auto-engine: Next-generation high accuracy solution via local computing power, supports multiple languages.
+- hybrid-http-client: High accuracy via remote computing power but requires a little local computing power(client suitable for openai-compatible servers), supports multiple languages."""
         ),
         parse_method: str = Form(
             "auto",
-            description="""(Adapted only for pipeline backend)The method for parsing PDF:
+            description="""(Adapted only for pipeline and hybrid backend)The method for parsing PDF:
 - auto: Automatically determine the method based on the file type
 - txt: Use text extraction method
 - ocr: Use OCR method for image-based PDFs
@@ -131,7 +152,7 @@ Options: ch, ch_server, ch_lite, en, korean, japan, chinese_cht, ta, te, ka, th,
         table_enable: bool = Form(True, description="Enable table parsing."),
         server_url: Optional[str] = Form(
             None,
-            description="(Adapted only for vlm-http-client backend)openai compatible server url, e.g., http://127.0.0.1:30000"
+            description="(Adapted only for <vlm/hybrid>-http-client backend)openai compatible server url, e.g., http://127.0.0.1:30000"
         ),
         return_md: bool = Form(True, description="Return markdown content in response"),
         return_middle_json: bool = Form(False, description="Return middle JSON in response"),
@@ -220,10 +241,13 @@ Options: ch, ch_server, ch_lite, en, korean, japan, chinese_cht, ta, te, ka, th,
             with zipfile.ZipFile(zip_path, "w", compression=zipfile.ZIP_DEFLATED) as zf:
                 for pdf_name in pdf_file_names:
                     safe_pdf_name = sanitize_filename(pdf_name)
+
                     if backend.startswith("pipeline"):
                         parse_dir = os.path.join(unique_dir, pdf_name, parse_method)
-                    else:
+                    elif backend.startswith("vlm"):
                         parse_dir = os.path.join(unique_dir, pdf_name, "vlm")
+                    elif backend.startswith("hybrid"):
+                        parse_dir = os.path.join(unique_dir, pdf_name, f"hybrid_{parse_method}")
 
                     if not os.path.exists(parse_dir):
                         continue
@@ -271,8 +295,10 @@ Options: ch, ch_server, ch_lite, en, korean, japan, chinese_cht, ta, te, ka, th,
 
                 if backend.startswith("pipeline"):
                     parse_dir = os.path.join(unique_dir, pdf_name, parse_method)
-                else:
+                elif backend.startswith("vlm"):
                     parse_dir = os.path.join(unique_dir, pdf_name, "vlm")
+                elif backend.startswith("hybrid"):
+                    parse_dir = os.path.join(unique_dir, pdf_name, f"hybrid_{parse_method}")
 
                 if os.path.exists(parse_dir):
                     if return_md: