Merge pull request #2705 from opendatalab/dev

update docs

README.md

@@ -51,9 +51,14 @@ Easier to use: Just grab MinerU Desktop. No coding, no login, just a simple inte
 </div>
 
 # Changelog
-- 2025/06/15 2.0.2 released
+- 2025/06/17 2.0.5 Released 
+  - Fixed the issue where models were still required to be downloaded in the `sglang-client` mode  
+  - Fixed the issue where the `sglang-client` mode unnecessarily depended on packages like `torch` during runtime.
+  - Fixed the issue where only the first instance would take effect when attempting to launch multiple `sglang-client` instances via multiple URLs within the same process
+- 2025/06/15 2.0.3 released
   - Fixed a configuration file key-value update error that occurred when downloading model type was set to `all`
-  - Fixed an issue where formula and table feature toggle parameters in the command line interface could not be effectively disabled
+  - Fixed the issue where the formula and table feature toggle switches were not working in `command line mode`, causing the features to remain enabled.
+  - Fixed compatibility issues with sglang version 0.4.7 in the `sglang-engine` mode.
   - Updated Dockerfile and installation documentation for deploying the full version of MinerU in sglang environment
 - 2025/06/13 2.0.0 Released
   - MinerU 2.0 represents a comprehensive reconstruction and upgrade from architecture to functionality, delivering a more streamlined design, enhanced performance, and more flexible user experience.
@@ -366,7 +371,7 @@ Easier to use: Just grab MinerU Desktop. No coding, no login, just a simple inte
         <li><a href="#quick-start">Quick Start</a>
             <ul>
             <li><a href="#online-demo">Online Demo</a></li>
-            <li><a href="#quick-cpu-demo">Local Deployment</a></li>
+            <li><a href="#local-deployment">Local Deployment</a></li>
             </ul>
         </li>
       </ul>
@@ -532,6 +537,14 @@ If you need to use **sglang to accelerate VLM model inference**, you can choose
 > [!TIP]
 > The Dockerfile uses `lmsysorg/sglang:v0.4.7-cu124` as the default base image. If necessary, you can modify it to another platform version.
 
+
+#### 1.4 Install client  (for connecting to sglang-server on edge devices that require only CPU and network connectivity)
+
+```bash
+uv pip install -U mineru
+mineru -p <input_path> -o <output_path> -b vlm-sglang-client -u http://<host_ip>:<port>
+```
+
 ---
 
 ### 2. Using MinerU

README_zh-CN.md

@@ -50,9 +50,14 @@
 </div>
 
 # 更新记录
-- 2025/06/15 2.0.2发布
+- 2025/06/17 2.0.5发布
+  - 修复了`sglang-client`模式下依然需要下载模型的问题
+  - 修复了`sglang-client`模式需要依赖`torch`等实际运行不需要的包的问题
+  - 修复了同一进程内尝试通过多个url启动多个`sglang-client`实例时，只有第一个生效的问题
+- 2025/06/15 2.0.3发布
   - 修复了当下载模型类型设置为`all`时，配置文件出现键值更新错误的问题
-  - 修复了命令行接口中公式和表格功能的开关参数实际无法关闭的问题
+  - 修复了命令行模式下公式和表格功能开关不生效导致功能无法关闭的问题
+  - 修复了`sglang-engine`模式下，0.4.7版本sglang的兼容性问题
   - 更新了sglang环境下部署完整版MinerU的Dockerfile和相关安装文档
 - 2025/06/13 2.0.0发布
   - MinerU 2.0 是一次从架构到功能的全面重构与升级，带来了更简洁的设计、更强的性能以及更灵活的使用体验。
@@ -521,6 +526,13 @@ uv pip install -e .[core] -i https://mirrors.aliyun.com/pypi/simple
   
 > [!TIP]
 > Dockerfile默认使用`lmsysorg/sglang:v0.4.7-cu124`作为基础镜像，如有需要，您可以自行修改为其他平台版本。
+  
+#### 1.4 安装client（用于在仅需 CPU 和网络连接的边缘设备上连接 sglang-server）
+
+```bash
+uv pip install -U mineru -i https://mirrors.aliyun.com/pypi/simple
+mineru -p <input_path> -o <output_path> -b vlm-sglang-client -u http://<host_ip>:<port>
+```
 
 ---
 

demo/demo.py

@@ -240,4 +240,4 @@ if __name__ == '__main__':
     """To enable VLM mode, change the backend to 'vlm-xxx'"""
     # parse_doc(doc_path_list, output_dir, backend="vlm-transformers")  # more general.
     # parse_doc(doc_path_list, output_dir, backend="vlm-sglang-engine")  # faster(engine).
-    # parse_doc(doc_path_list, output_dir, backend="vlm-sglang-client", server_url="http://127.0.0.1:30000"）  # faster(client).
+    # parse_doc(doc_path_list, output_dir, backend="vlm-sglang-client", server_url="http://127.0.0.1:30000")  # faster(client).

mineru/backend/pipeline/pipeline_analyze.py

@@ -2,15 +2,12 @@ import os
 import time
 from typing import List, Tuple
 import PIL.Image
-import torch
+from loguru import logger
 
 from .model_init import MineruPipelineModel
 from mineru.utils.config_reader import get_device
 from ...utils.pdf_classify import classify
 from ...utils.pdf_image_tools import load_images_from_pdf
-
-from loguru import logger
-
 from ...utils.model_utils import get_vram, clean_memory
 
 
@@ -76,7 +73,11 @@ def doc_analyze(
         formula_enable=True,
         table_enable=True,
 ):
-    MIN_BATCH_INFERENCE_SIZE = int(os.environ.get('MINERU_MIN_BATCH_INFERENCE_SIZE', 100))
+    """
+    适当调大MIN_BATCH_INFERENCE_SIZE可以提高性能，可能会增加显存使用量，
+    可通过环境变量MINERU_MIN_BATCH_INFERENCE_SIZE设置，默认值为100。
+    """
+    min_batch_inference_size = int(os.environ.get('MINERU_MIN_BATCH_INFERENCE_SIZE', 100))
 
     # 收集所有页面信息
     all_pages_info = []  # 存储(dataset_index, page_index, img, ocr, lang, width, height)
@@ -109,7 +110,7 @@ def doc_analyze(
 
     # 准备批处理
     images_with_extra_info = [(info[2], info[3], info[4]) for info in all_pages_info]
-    batch_size = MIN_BATCH_INFERENCE_SIZE
+    batch_size = min_batch_inference_size
     batch_images = [
         images_with_extra_info[i:i + batch_size]
         for i in range(0, len(images_with_extra_info), batch_size)
@@ -162,7 +163,7 @@ def batch_image_analyze(
         try:
             import torch_npu
             if torch_npu.npu.is_available():
-                torch.npu.set_compile_mode(jit_compile=False)
+                torch_npu.npu.set_compile_mode(jit_compile=False)
         except Exception as e:
             raise RuntimeError(
                 "NPU is selected as device, but torch_npu is not available. "

mineru/backend/vlm/vlm_analyze.py

@@ -8,7 +8,6 @@ from mineru.utils.pdf_image_tools import load_images_from_pdf
 from .base_predictor import BasePredictor
 from .predictor import get_predictor
 from .token_to_middle_json import result_to_middle_json
-from ...utils.enum_class import ModelPath
 from ...utils.models_download_utils import auto_download_and_get_model_root_path
 
 
@@ -27,9 +26,9 @@ class ModelSingleton:
         model_path: str | None,
         server_url: str | None,
     ) -> BasePredictor:
-        key = (backend,)
+        key = (backend, model_path, server_url)
         if key not in self._models:
-            if not model_path:
+            if backend in ['transformers', 'sglang-engine'] and not model_path:
                 model_path = auto_download_and_get_model_root_path("/","vlm")
             self._models[key] = get_predictor(
                 backend=backend,

mineru/cli/client.py

@@ -9,7 +9,6 @@ from mineru.utils.model_utils import get_vram
 from ..version import __version__
 from .common import do_parse, read_fn, pdf_suffixes, image_suffixes
 
-
 @click.command()
 @click.version_option(__version__,
                       '--version',
@@ -139,25 +138,26 @@ from .common import do_parse, read_fn, pdf_suffixes, image_suffixes
 
 def main(input_path, output_dir, method, backend, lang, server_url, start_page_id, end_page_id, formula_enable, table_enable, device_mode, virtual_vram, model_source):
 
-    def get_device_mode() -> str:
-        if device_mode is not None:
-            return device_mode
-        else:
-            return get_device()
-    if os.getenv('MINERU_DEVICE_MODE', None) is None:
-        os.environ['MINERU_DEVICE_MODE'] = get_device_mode()
+    if not backend.endswith('-client'):
+        def get_device_mode() -> str:
+            if device_mode is not None:
+                return device_mode
+            else:
+                return get_device()
+        if os.getenv('MINERU_DEVICE_MODE', None) is None:
+            os.environ['MINERU_DEVICE_MODE'] = get_device_mode()
 
-    def get_virtual_vram_size() -> int:
-        if virtual_vram is not None:
-            return virtual_vram
-        if get_device_mode().startswith("cuda") or get_device_mode().startswith("npu"):
-            return round(get_vram(get_device_mode()))
-        return 1
-    if os.getenv('MINERU_VIRTUAL_VRAM_SIZE', None) is None:
-        os.environ['MINERU_VIRTUAL_VRAM_SIZE']= str(get_virtual_vram_size())
+        def get_virtual_vram_size() -> int:
+            if virtual_vram is not None:
+                return virtual_vram
+            if get_device_mode().startswith("cuda") or get_device_mode().startswith("npu"):
+                return round(get_vram(get_device_mode()))
+            return 1
+        if os.getenv('MINERU_VIRTUAL_VRAM_SIZE', None) is None:
+            os.environ['MINERU_VIRTUAL_VRAM_SIZE']= str(get_virtual_vram_size())
 
-    if os.getenv('MINERU_MODEL_SOURCE', None) is None:
-        os.environ['MINERU_MODEL_SOURCE'] = model_source
+        if os.getenv('MINERU_MODEL_SOURCE', None) is None:
+            os.environ['MINERU_MODEL_SOURCE'] = model_source
 
     os.makedirs(output_dir, exist_ok=True)
 

mineru/cli/common.py

@@ -8,15 +8,12 @@ from pathlib import Path
 import pypdfium2 as pdfium
 from loguru import logger
 
-from mineru.backend.pipeline.pipeline_middle_json_mkcontent import union_make as pipeline_union_make
-from mineru.backend.pipeline.model_json_to_middle_json import result_to_middle_json as pipeline_result_to_middle_json
-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.pipeline.pipeline_analyze import doc_analyze as pipeline_doc_analyze
 from mineru.data.data_reader_writer import FileBasedDataWriter
 from mineru.utils.draw_bbox import draw_layout_bbox, draw_span_bbox
 from mineru.utils.enum_class import MakeMode
 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
 
 pdf_suffixes = [".pdf"]
 image_suffixes = [".png", ".jpeg", ".jpg"]
@@ -99,6 +96,11 @@ def do_parse(
 ):
 
     if backend == "pipeline":
+
+        from mineru.backend.pipeline.pipeline_middle_json_mkcontent import union_make as pipeline_union_make
+        from mineru.backend.pipeline.model_json_to_middle_json import result_to_middle_json as pipeline_result_to_middle_json
+        from mineru.backend.pipeline.pipeline_analyze import doc_analyze as pipeline_doc_analyze
+
         for idx, pdf_bytes in enumerate(pdf_bytes_list):
             new_pdf_bytes = convert_pdf_bytes_to_bytes_by_pypdfium2(pdf_bytes, start_page_id, end_page_id)
             pdf_bytes_list[idx] = new_pdf_bytes
@@ -163,6 +165,7 @@ def do_parse(
 
             logger.info(f"local output dir is {local_md_dir}")
     else:
+
         if backend.startswith("vlm-"):
             backend = backend[4:]
 

mineru/model/mfr/unimernet/Unimernet.py

@@ -21,7 +21,7 @@ class MathDataset(Dataset):
 class UnimernetModel(object):
     def __init__(self, weight_dir, _device_="cpu"):
         from .unimernet_hf import UnimernetModel
-        if _device_.startswith("mps"):
+        if _device_.startswith("mps") or _device_.startswith("npu"):
             self.model = UnimernetModel.from_pretrained(weight_dir, attn_implementation="eager")
         else:
             self.model = UnimernetModel.from_pretrained(weight_dir)

mineru/model/table/rapid_table.py

@@ -1,4 +1,5 @@
 import os
+import html
 import cv2
 import numpy as np
 from loguru import logger
@@ -8,6 +9,11 @@ from mineru.utils.enum_class import ModelPath
 from mineru.utils.models_download_utils import auto_download_and_get_model_root_path
 
 
+def escape_html(input_string):
+    """Escape HTML Entities."""
+    return html.escape(input_string)
+
+
 class RapidTableModel(object):
     def __init__(self, ocr_engine):
         slanet_plus_model_path = os.path.join(auto_download_and_get_model_root_path(ModelPath.slanet_plus), ModelPath.slanet_plus)
@@ -63,7 +69,7 @@ class RapidTableModel(object):
         # Continue with OCR on potentially rotated image
         ocr_result = self.ocr_engine.ocr(bgr_image)[0]
         if ocr_result:
-            ocr_result = [[item[0], item[1][0], item[1][1]] for item in ocr_result if
+            ocr_result = [[item[0], escape_html(item[1][0]), item[1][1]] for item in ocr_result if
                       len(item) == 2 and isinstance(item[1], tuple)]
         else:
             ocr_result = None

mineru/model/vlm_sglang_model/engine.py

@@ -183,8 +183,8 @@ async def _one_request(
     created_time: Optional[float],
 ):
     tokenized_obj = await self._tokenize_one_request(obj)
-    self._send_one_request(obj, tokenized_obj, created_time)
-    async for out in self._wait_one_response(obj, request):
+    state = self._send_one_request(obj, tokenized_obj, created_time)
+    async for out in self._wait_one_response(obj, state, request):
         yield out
 
 
@@ -256,8 +256,8 @@ async def _generate_request(
         is_single = obj.is_single
         if is_single:
             tokenized_obj = await self._tokenize_one_request(obj)
-            self._send_one_request(obj, tokenized_obj, created_time)
-            async for response in self._wait_one_response(obj, request):
+            state = self._send_one_request(obj, tokenized_obj, created_time)
+            async for response in self._wait_one_response(obj, state, request):
                 yield response
         else:
             async for response in _handle_batch_request(self, obj, request, created_time):

mineru/model/vlm_sglang_model/server.py

@@ -6,10 +6,26 @@ from sglang.srt.entrypoints.http_server import app, generate_request, launch_ser
 from sglang.srt.managers.io_struct import GenerateReqInput
 from sglang.srt.server_args import prepare_server_args
 from sglang.srt.utils import kill_process_tree
+from sglang.srt.conversation import Conversation
 
 from mineru.utils.models_download_utils import auto_download_and_get_model_root_path
 from .logit_processor import Mineru2LogitProcessor
 
+# mineru2.0的chat_template与chatml在换行上有微小区别
+def custom_get_prompt(self) -> str:
+    system_prompt = self.system_template.format(system_message=self.system_message)
+    if self.system_message == "":
+        ret = ""
+    else:
+        ret = system_prompt + self.sep
+
+    for role, message in self.messages:
+        if message:
+            ret += role + "\n" + message + self.sep
+        else:
+            ret += role + "\n"
+    return ret
+
 _custom_logit_processor_str = Mineru2LogitProcessor().to_str()
 
 # remote the existing /generate route
@@ -45,6 +61,7 @@ def main():
 
     if server_args.chat_template is None:
         server_args.chat_template = "chatml"
+        Conversation.get_prompt = custom_get_prompt
 
     server_args.enable_custom_logit_processor = True
 

mineru/utils/config_reader.py

@@ -1,10 +1,15 @@
 # Copyright (c) Opendatalab. All rights reserved.
 import json
 import os
-
-import torch
 from loguru import logger
 
+try:
+    import torch
+    import torch_npu
+except ImportError:
+    pass
+
+
 # 定义配置文件名常量
 CONFIG_FILE_NAME = os.getenv('MINERU_TOOLS_CONFIG_JSON', 'mineru.json')
 
@@ -78,7 +83,6 @@ def get_device():
             return "mps"
         else:
             try:
-                import torch_npu
                 if torch_npu.npu.is_available():
                     return "npu"
             except Exception as e:

mineru/utils/enum_class.py

@@ -33,9 +33,11 @@ class CategoryId:
     TableCaption = 6
     TableFootnote = 7
     InterlineEquation_Layout = 8
+    InterlineEquationNumber_Layout = 9
     InlineEquation = 13
     InterlineEquation_YOLO = 14
     OcrText = 15
+    LowScoreText = 16
     ImageFootnote = 101
 
 

mineru/utils/model_utils.py

@@ -1,5 +1,4 @@
 import time
-import torch
 import gc
 from PIL import Image
 from loguru import logger
@@ -7,6 +6,12 @@ import numpy as np
 
 from mineru.utils.boxbase import get_minbox_if_overlap_by_ratio
 
+try:
+    import torch
+    import torch_npu
+except ImportError:
+    pass
+
 
 def crop_img(input_res, input_img, crop_paste_x=0, crop_paste_y=0):
 
@@ -303,7 +308,6 @@ def clean_memory(device='cuda'):
             torch.cuda.empty_cache()
             torch.cuda.ipc_collect()
     elif str(device).startswith("npu"):
-        import torch_npu
         if torch_npu.npu.is_available():
             torch_npu.npu.empty_cache()
     elif str(device).startswith("mps"):
@@ -325,7 +329,6 @@ def get_vram(device):
         total_memory = torch.cuda.get_device_properties(device).total_memory / (1024 ** 3)  # 将字节转换为 GB
         return total_memory
     elif str(device).startswith("npu"):
-        import torch_npu
         if torch_npu.npu.is_available():
             total_memory = torch_npu.npu.get_device_properties(device).total_memory / (1024 ** 3)  # 转为 GB
             return total_memory

mineru/version.py

@@ -1 +1 @@
-__version__ = "2.0.1"
+__version__ = "2.0.4"

pyproject.toml

@@ -43,7 +43,7 @@ vlm = [
     "pydantic",
 ]
 sglang = [
-    "sglang[all]>=0.4.7",
+    "sglang[all]==0.4.7",
 ]
 pipeline = [
     "matplotlib>=3.10,<4",

signatures/version1/cla.json

@@ -311,6 +311,22 @@
       "created_at": "2025-06-13T14:02:16Z",
       "repoId": 765083837,
       "pullRequestNo": 2634
+    },
+    {
+      "name": "hotelll",
+      "id": 45009029,
+      "comment_id": 2978780331,
+      "created_at": "2025-06-17T03:09:54Z",
+      "repoId": 765083837,
+      "pullRequestNo": 2676
+    },
+    {
+      "name": "hsia",
+      "id": 654127,
+      "comment_id": 2979415817,
+      "created_at": "2025-06-17T17:35:10Z",
+      "repoId": 765083837,
+      "pullRequestNo": 2699
     }
   ]
 }