Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -8,11 +8,11 @@ import time
|
|
| 8 |
import zipfile
|
| 9 |
from pathlib import Path
|
| 10 |
|
| 11 |
-
os.system('pip uninstall -y magic-pdf')
|
| 12 |
os.system('pip uninstall -y mineru')
|
| 13 |
-
os.system('pip install git+https://github.com/
|
|
|
|
| 14 |
|
| 15 |
-
os.system('mineru-models-download -s huggingface -m
|
| 16 |
|
| 17 |
with open('/home/user/mineru.json', 'r') as file:
|
| 18 |
config = json.load(file)
|
|
@@ -35,24 +35,27 @@ from gradio_pdf import PDF
|
|
| 35 |
import gradio as gr
|
| 36 |
from loguru import logger
|
| 37 |
|
| 38 |
-
from mineru.cli.common import prepare_env,
|
| 39 |
from mineru.utils.hash_utils import str_sha256
|
| 40 |
|
| 41 |
os.environ['MINERU_MODEL_SOURCE'] = 'local'
|
| 42 |
|
| 43 |
|
| 44 |
-
def parse_pdf(doc_path, output_dir, end_page_id, is_ocr, formula_enable, table_enable, language):
|
| 45 |
os.makedirs(output_dir, exist_ok=True)
|
| 46 |
|
| 47 |
try:
|
| 48 |
-
file_name = f'{
|
| 49 |
pdf_data = read_fn(doc_path)
|
| 50 |
if is_ocr:
|
| 51 |
parse_method = 'ocr'
|
| 52 |
else:
|
| 53 |
parse_method = 'auto'
|
|
|
|
|
|
|
|
|
|
| 54 |
local_image_dir, local_md_dir = prepare_env(output_dir, file_name, parse_method)
|
| 55 |
-
|
| 56 |
output_dir=output_dir,
|
| 57 |
pdf_file_names=[file_name],
|
| 58 |
pdf_bytes_list=[pdf_data],
|
|
@@ -61,6 +64,8 @@ def parse_pdf(doc_path, output_dir, end_page_id, is_ocr, formula_enable, table_e
|
|
| 61 |
end_page_id=end_page_id,
|
| 62 |
p_formula_enable=formula_enable,
|
| 63 |
p_table_enable=table_enable,
|
|
|
|
|
|
|
| 64 |
)
|
| 65 |
return local_md_dir, file_name
|
| 66 |
except Exception as e:
|
|
@@ -112,12 +117,10 @@ def replace_image_with_base64(markdown_text, image_dir_path):
|
|
| 112 |
return re.sub(pattern, replace, markdown_text)
|
| 113 |
|
| 114 |
|
| 115 |
-
def to_markdown(file_path, end_pages, is_ocr, formula_enable, table_enable, language):
|
| 116 |
file_path = to_pdf(file_path)
|
| 117 |
-
if end_pages > 20:
|
| 118 |
-
end_pages = 20
|
| 119 |
# 获取识别的md文件以及压缩包文件路径
|
| 120 |
-
local_md_dir, file_name = parse_pdf(file_path, './output', end_pages - 1, is_ocr, formula_enable, table_enable, language)
|
| 121 |
archive_zip_path = os.path.join('./output', str_sha256(local_md_dir) + '.zip')
|
| 122 |
zip_archive_success = compress_directory_to_zip(local_md_dir, archive_zip_path)
|
| 123 |
if zip_archive_success == 0:
|
|
@@ -134,8 +137,12 @@ def to_markdown(file_path, end_pages, is_ocr, formula_enable, table_enable, lang
|
|
| 134 |
return md_content, txt_content, archive_zip_path, new_pdf_path
|
| 135 |
|
| 136 |
|
| 137 |
-
latex_delimiters = [
|
| 138 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 139 |
|
| 140 |
|
| 141 |
with open("header.html", "r") as file:
|
|
@@ -192,19 +199,39 @@ def to_pdf(file_path):
|
|
| 192 |
return tmp_file_path
|
| 193 |
|
| 194 |
|
| 195 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 196 |
with gr.Blocks() as demo:
|
| 197 |
gr.HTML(header)
|
| 198 |
with gr.Row():
|
| 199 |
with gr.Column(variant='panel', scale=5):
|
| 200 |
with gr.Row():
|
| 201 |
file = gr.File(label='Please upload a PDF or image', file_types=['.pdf', '.png', '.jpeg', '.jpg'])
|
| 202 |
-
with gr.Row(equal_height=True):
|
| 203 |
-
with gr.Column(scale=4):
|
| 204 |
-
max_pages = gr.Slider(1, 20, 10, step=1, label='Max convert pages')
|
| 205 |
-
with gr.Column(scale=1):
|
| 206 |
-
language = gr.Dropdown(all_lang, label='Language', value='ch')
|
| 207 |
with gr.Row():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 208 |
is_ocr = gr.Checkbox(label='Force enable OCR', value=False)
|
| 209 |
formula_enable = gr.Checkbox(label='Enable formula recognition', value=True)
|
| 210 |
table_enable = gr.Checkbox(label='Enable table recognition(test)', value=True)
|
|
@@ -212,13 +239,15 @@ if __name__ == '__main__':
|
|
| 212 |
change_bu = gr.Button('Convert')
|
| 213 |
clear_bu = gr.ClearButton(value='Clear')
|
| 214 |
pdf_show = PDF(label='PDF preview', interactive=False, visible=True, height=800)
|
| 215 |
-
|
| 216 |
example_root = os.path.join(os.path.dirname(__file__), 'examples')
|
| 217 |
-
|
| 218 |
-
|
| 219 |
-
|
| 220 |
-
|
| 221 |
-
|
|
|
|
|
|
|
| 222 |
|
| 223 |
with gr.Column(variant='panel', scale=5):
|
| 224 |
output_file = gr.File(label='convert result', interactive=False)
|
|
@@ -229,9 +258,35 @@ if __name__ == '__main__':
|
|
| 229 |
line_breaks=True)
|
| 230 |
with gr.Tab('Markdown text'):
|
| 231 |
md_text = gr.TextArea(lines=45, show_copy_button=True)
|
| 232 |
-
|
| 233 |
-
|
| 234 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 235 |
clear_bu.add([file, md, pdf_show, md_text, output_file, is_ocr])
|
| 236 |
|
| 237 |
-
demo.launch(ssr_mode=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
import zipfile
|
| 9 |
from pathlib import Path
|
| 10 |
|
|
|
|
| 11 |
os.system('pip uninstall -y mineru')
|
| 12 |
+
os.system('pip install git+https://github.com/myhloli/Magic-PDF.git@dev')
|
| 13 |
+
os.system('pip install sglang[all]==0.4.8')
|
| 14 |
|
| 15 |
+
os.system('mineru-models-download -s huggingface -m all')
|
| 16 |
|
| 17 |
with open('/home/user/mineru.json', 'r') as file:
|
| 18 |
config = json.load(file)
|
|
|
|
| 35 |
import gradio as gr
|
| 36 |
from loguru import logger
|
| 37 |
|
| 38 |
+
from mineru.cli.common import prepare_env, read_fn, aio_do_parse
|
| 39 |
from mineru.utils.hash_utils import str_sha256
|
| 40 |
|
| 41 |
os.environ['MINERU_MODEL_SOURCE'] = 'local'
|
| 42 |
|
| 43 |
|
| 44 |
+
async def parse_pdf(doc_path, output_dir, end_page_id, is_ocr, formula_enable, table_enable, language, backend, url):
|
| 45 |
os.makedirs(output_dir, exist_ok=True)
|
| 46 |
|
| 47 |
try:
|
| 48 |
+
file_name = f'{safe_stem(Path(doc_path).stem)}_{time.strftime("%y%m%d_%H%M%S")}'
|
| 49 |
pdf_data = read_fn(doc_path)
|
| 50 |
if is_ocr:
|
| 51 |
parse_method = 'ocr'
|
| 52 |
else:
|
| 53 |
parse_method = 'auto'
|
| 54 |
+
|
| 55 |
+
if backend.startswith("vlm"):
|
| 56 |
+
parse_method = "vlm"
|
| 57 |
local_image_dir, local_md_dir = prepare_env(output_dir, file_name, parse_method)
|
| 58 |
+
await aio_do_parse(
|
| 59 |
output_dir=output_dir,
|
| 60 |
pdf_file_names=[file_name],
|
| 61 |
pdf_bytes_list=[pdf_data],
|
|
|
|
| 64 |
end_page_id=end_page_id,
|
| 65 |
p_formula_enable=formula_enable,
|
| 66 |
p_table_enable=table_enable,
|
| 67 |
+
backend=backend,
|
| 68 |
+
server_url=url,
|
| 69 |
)
|
| 70 |
return local_md_dir, file_name
|
| 71 |
except Exception as e:
|
|
|
|
| 117 |
return re.sub(pattern, replace, markdown_text)
|
| 118 |
|
| 119 |
|
| 120 |
+
async def to_markdown(file_path, end_pages=10, is_ocr=False, formula_enable=True, table_enable=True, language="ch", backend="pipeline", url=None):
|
| 121 |
file_path = to_pdf(file_path)
|
|
|
|
|
|
|
| 122 |
# 获取识别的md文件以及压缩包文件路径
|
| 123 |
+
local_md_dir, file_name = await parse_pdf(file_path, './output', end_pages - 1, is_ocr, formula_enable, table_enable, language, backend, url)
|
| 124 |
archive_zip_path = os.path.join('./output', str_sha256(local_md_dir) + '.zip')
|
| 125 |
zip_archive_success = compress_directory_to_zip(local_md_dir, archive_zip_path)
|
| 126 |
if zip_archive_success == 0:
|
|
|
|
| 137 |
return md_content, txt_content, archive_zip_path, new_pdf_path
|
| 138 |
|
| 139 |
|
| 140 |
+
latex_delimiters = [
|
| 141 |
+
{'left': '$$', 'right': '$$', 'display': True},
|
| 142 |
+
{'left': '$', 'right': '$', 'display': False},
|
| 143 |
+
{'left': '\\(', 'right': '\\)', 'display': False},
|
| 144 |
+
{'left': '\\[', 'right': '\\]', 'display': True},
|
| 145 |
+
]
|
| 146 |
|
| 147 |
|
| 148 |
with open("header.html", "r") as file:
|
|
|
|
| 199 |
return tmp_file_path
|
| 200 |
|
| 201 |
|
| 202 |
+
def main():
|
| 203 |
+
example_enable = True
|
| 204 |
+
|
| 205 |
+
try:
|
| 206 |
+
print("预初始化SgLang引擎...")
|
| 207 |
+
from mineru.backend.vlm.vlm_analyze import ModelSingleton
|
| 208 |
+
modelsingleton = ModelSingleton()
|
| 209 |
+
predictor = modelsingleton.get_model(
|
| 210 |
+
"sglang-engine",
|
| 211 |
+
None,
|
| 212 |
+
'http://localhost:30000',
|
| 213 |
+
mem_fraction_static=0.5,
|
| 214 |
+
enable_torch_compile=True,
|
| 215 |
+
)
|
| 216 |
+
print("SgLang引擎初始化完成")
|
| 217 |
+
except Exception as e:
|
| 218 |
+
logger.exception(e)
|
| 219 |
+
|
| 220 |
with gr.Blocks() as demo:
|
| 221 |
gr.HTML(header)
|
| 222 |
with gr.Row():
|
| 223 |
with gr.Column(variant='panel', scale=5):
|
| 224 |
with gr.Row():
|
| 225 |
file = gr.File(label='Please upload a PDF or image', file_types=['.pdf', '.png', '.jpeg', '.jpg'])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 226 |
with gr.Row():
|
| 227 |
+
max_pages = gr.Slider(1, 20, 10, step=1, label='Max convert pages')
|
| 228 |
+
with gr.Row():
|
| 229 |
+
backend = gr.Dropdown(["pipeline", "vlm-sglang-engine"], label="Backend", value="pipeline")
|
| 230 |
+
with gr.Row(visible=True) as ocr_options:
|
| 231 |
+
language = gr.Dropdown(all_lang, label='Language', value='ch')
|
| 232 |
+
with gr.Row(visible=False) as client_options:
|
| 233 |
+
url = gr.Textbox(label='Server URL', value='http://localhost:30000', placeholder='http://localhost:30000')
|
| 234 |
+
with gr.Row(visible=True) as pipeline_options:
|
| 235 |
is_ocr = gr.Checkbox(label='Force enable OCR', value=False)
|
| 236 |
formula_enable = gr.Checkbox(label='Enable formula recognition', value=True)
|
| 237 |
table_enable = gr.Checkbox(label='Enable table recognition(test)', value=True)
|
|
|
|
| 239 |
change_bu = gr.Button('Convert')
|
| 240 |
clear_bu = gr.ClearButton(value='Clear')
|
| 241 |
pdf_show = PDF(label='PDF preview', interactive=False, visible=True, height=800)
|
| 242 |
+
if example_enable:
|
| 243 |
example_root = os.path.join(os.path.dirname(__file__), 'examples')
|
| 244 |
+
if os.path.exists(example_root):
|
| 245 |
+
with gr.Accordion('Examples:'):
|
| 246 |
+
gr.Examples(
|
| 247 |
+
examples=[os.path.join(example_root, _) for _ in os.listdir(example_root) if
|
| 248 |
+
_.endswith('pdf')],
|
| 249 |
+
inputs=file
|
| 250 |
+
)
|
| 251 |
|
| 252 |
with gr.Column(variant='panel', scale=5):
|
| 253 |
output_file = gr.File(label='convert result', interactive=False)
|
|
|
|
| 258 |
line_breaks=True)
|
| 259 |
with gr.Tab('Markdown text'):
|
| 260 |
md_text = gr.TextArea(lines=45, show_copy_button=True)
|
| 261 |
+
|
| 262 |
+
|
| 263 |
+
# 更新界面函数
|
| 264 |
+
def update_interface(backend_choice):
|
| 265 |
+
if backend_choice in ["vlm-transformers", "vlm-sglang-engine"]:
|
| 266 |
+
return gr.update(visible=False), gr.update(visible=False), gr.update(visible=False)
|
| 267 |
+
elif backend_choice in ["vlm-sglang-client"]: # pipeline
|
| 268 |
+
return gr.update(visible=True), gr.update(visible=False), gr.update(visible=False)
|
| 269 |
+
elif backend_choice in ["pipeline"]:
|
| 270 |
+
return gr.update(visible=False), gr.update(visible=True), gr.update(visible=True)
|
| 271 |
+
else:
|
| 272 |
+
pass
|
| 273 |
+
|
| 274 |
+
|
| 275 |
+
# 添加事件处理
|
| 276 |
+
backend.change(
|
| 277 |
+
fn=update_interface,
|
| 278 |
+
inputs=[backend],
|
| 279 |
+
outputs=[client_options, ocr_options, pipeline_options],
|
| 280 |
+
api_name=False
|
| 281 |
+
)
|
| 282 |
+
|
| 283 |
+
file.change(fn=to_pdf, inputs=file, outputs=pdf_show, api_name=False)
|
| 284 |
+
change_bu.click(fn=to_markdown, inputs=[file, max_pages, is_ocr, formula_enable, table_enable, language, backend, url],
|
| 285 |
+
outputs=[md, md_text, output_file, pdf_show], api_name=False)
|
| 286 |
clear_bu.add([file, md, pdf_show, md_text, output_file, is_ocr])
|
| 287 |
|
| 288 |
+
demo.launch(server_name='localhost', ssr_mode=True)
|
| 289 |
+
|
| 290 |
+
|
| 291 |
+
if __name__ == '__main__':
|
| 292 |
+
main()
|