Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -52,7 +52,7 @@ preset_prompts = [
|
|
52 |
|
53 |
|
54 |
def send_pdf_to_parse(file_path, server_ip, port, route="/upload", api_key=None):
|
55 |
-
url = f"
|
56 |
headers = {}
|
57 |
if api_key:
|
58 |
headers["Authorization"] = f"Bearer {api_key}"
|
@@ -65,12 +65,12 @@ def send_pdf_to_parse(file_path, server_ip, port, route="/upload", api_key=None)
|
|
65 |
|
66 |
|
67 |
|
68 |
-
async def send_pdf_async_aiohttp(file_path, server_ip,
|
69 |
"""使用aiohttp异步发送PDF"""
|
70 |
url = f"http://{server_ip}:{port}{route}"
|
71 |
headers = {}
|
72 |
-
if
|
73 |
-
headers["Authorization"] = f"Bearer {
|
74 |
|
75 |
try:
|
76 |
async with aiohttp.ClientSession() as session:
|
@@ -100,38 +100,41 @@ IP = os.environ.get("IP")
|
|
100 |
|
101 |
PORT = os.environ.get("PORT")
|
102 |
|
103 |
-
|
104 |
|
105 |
client = AsyncOpenAI(
|
106 |
api_key=openai_api_key,
|
107 |
-
base_url=openai_api_base,
|
108 |
)
|
109 |
|
110 |
|
111 |
async def request(messages):
|
112 |
-
|
113 |
chat_completion_from_base64 = await client.chat.completions.create(
|
114 |
messages=messages,
|
|
|
|
|
|
|
115 |
model="Qwen2_5VL",
|
116 |
max_completion_tokens=4096,
|
117 |
stream=True,
|
118 |
temperature=0.0,
|
119 |
top_p=0.95
|
120 |
)
|
121 |
-
|
122 |
page = ""
|
123 |
async for chunk in chat_completion_from_base64:
|
124 |
if chunk.choices[0].delta.content:
|
125 |
content = chunk.choices[0].delta.content
|
126 |
-
|
127 |
choice = chunk.choices[0]
|
128 |
if choice.finish_reason is not None:
|
129 |
print(f"end reason = {choice.finish_reason}")
|
130 |
break
|
131 |
page += content
|
132 |
-
|
133 |
yield content
|
134 |
-
|
135 |
|
136 |
def images_to_pdf(img_paths, pdf_path):
|
137 |
|
@@ -165,7 +168,7 @@ def encode_image(image_path):
|
|
165 |
return base64.b64encode(image_file.read()).decode("utf-8")
|
166 |
|
167 |
def build_message(image_path, prompt):
|
168 |
-
|
169 |
content = [
|
170 |
{
|
171 |
"type": "image_url",
|
@@ -175,14 +178,14 @@ def build_message(image_path, prompt):
|
|
175 |
},
|
176 |
{"type": "text", 'text': prompt}
|
177 |
]
|
178 |
-
|
179 |
-
|
180 |
messages = [
|
181 |
{"role": "system", "content": "You are a helpful assistant."},
|
182 |
{'role': 'user', 'content': content}
|
183 |
-
|
184 |
]
|
185 |
-
|
186 |
return messages
|
187 |
|
188 |
|
@@ -211,14 +214,14 @@ async def doc_parser(doc_path, prompt):
|
|
211 |
for idx, page in enumerate(pages, start=1):
|
212 |
img_path = tmpdir / f"page_{idx}.png"
|
213 |
page.save(img_path, "PNG")
|
214 |
-
|
215 |
messages = build_message(img_path, prompt)
|
216 |
queries.append(messages)
|
217 |
-
|
218 |
else:
|
219 |
messages = build_message(doc_path, prompt)
|
220 |
queries.append(messages)
|
221 |
-
|
222 |
all_pages = []
|
223 |
all_pages_raw = []
|
224 |
for query in queries:
|
@@ -231,19 +234,19 @@ async def doc_parser(doc_path, prompt):
|
|
231 |
print(all_pages)
|
232 |
yield "\n---\n".join(all_pages), "\n\n".join(all_pages_raw)
|
233 |
|
234 |
-
|
235 |
def compress_directory_to_zip(directory_path, output_zip_path):
|
236 |
try:
|
237 |
with zipfile.ZipFile(output_zip_path, 'w', zipfile.ZIP_DEFLATED) as zipf:
|
238 |
|
239 |
-
|
240 |
for root, dirs, files in os.walk(directory_path):
|
241 |
for file in files:
|
242 |
-
|
243 |
file_path = os.path.join(root, file)
|
244 |
|
245 |
arcname = os.path.relpath(file_path, directory_path)
|
246 |
-
|
247 |
zipf.write(file_path, arcname)
|
248 |
return 0
|
249 |
except Exception as e:
|
@@ -263,7 +266,7 @@ def check_prompt(prompt):
|
|
263 |
return prompt
|
264 |
|
265 |
def to_file(image_path):
|
266 |
-
|
267 |
if image_path.endswith("Academic_Papers.png"):
|
268 |
image_path = image_path.replace("Academic_Papers.png", "Academic_Papers.pdf")
|
269 |
|
@@ -293,7 +296,7 @@ async def process_file(file_path):
|
|
293 |
images_to_pdf(file_path, tmp_file_path)
|
294 |
else:
|
295 |
tmp_file_path = file_path
|
296 |
-
asyncio.create_task(send_pdf_async_aiohttp(tmp_file_path,
|
297 |
|
298 |
return str(tmp_file_path)
|
299 |
|
@@ -302,7 +305,7 @@ if __name__ == '__main__':
|
|
302 |
with gr.Blocks() as demo:
|
303 |
with gr.Row():
|
304 |
with gr.Column(variant='panel', scale=5):
|
305 |
-
|
306 |
file = gr.File(label='Please upload a PDF or image', file_types=['.pdf', '.png', '.jpeg', '.jpg'], type="filepath")
|
307 |
prompts = gr.Dropdown(
|
308 |
choices=preset_prompts,
|
@@ -320,7 +323,7 @@ if __name__ == '__main__':
|
|
320 |
pdf_show = PDF(label='Preview', interactive=False, visible=True, height=800)
|
321 |
|
322 |
|
323 |
-
|
324 |
example_root = os.path.join(os.path.dirname(__file__), 'examples')
|
325 |
images = [
|
326 |
os.path.join(example_root, f)
|
@@ -334,9 +337,9 @@ if __name__ == '__main__':
|
|
334 |
file_path = [
|
335 |
os.path.join(example_root, f)
|
336 |
for f in ["Financial_Reports.png", "Books.png", "Magazines.png", "Academic_Papers.png"]
|
337 |
-
|
338 |
]
|
339 |
-
|
340 |
with gr.Row():
|
341 |
for i, label in enumerate(["Financial Reports(IMG)", "Books(IMG)", "Magazines(IMG)", "Academic Papers(PDF)"]):
|
342 |
with gr.Column(scale=1, min_width=120):
|
@@ -348,11 +351,11 @@ if __name__ == '__main__':
|
|
348 |
show_download_button=False
|
349 |
)
|
350 |
gr.Button(label).click(fn=to_file, inputs=gr.State(file_path[i]), outputs=file)
|
351 |
-
|
352 |
-
|
353 |
download_btn = gr.Button("⬇️ Generate download link", size="sm")
|
354 |
output_file = gr.File(label='Parse result', interactive=False, elem_id="down-file-box",visible=False)
|
355 |
-
|
356 |
gr.HTML("""
|
357 |
<style>
|
358 |
#down-file-box {
|
@@ -367,9 +370,9 @@ if __name__ == '__main__':
|
|
367 |
line_breaks=True)
|
368 |
with gr.Tab('Markdown text'):
|
369 |
md_text = gr.TextArea(lines=45, show_copy_button=True)
|
370 |
-
|
371 |
-
|
372 |
-
|
373 |
file.change(fn=process_file, inputs=file, outputs=pdf_show)
|
374 |
|
375 |
|
@@ -386,9 +389,9 @@ if __name__ == '__main__':
|
|
386 |
inputs=[file, prompts],
|
387 |
outputs=[md, md_text]
|
388 |
)
|
389 |
-
|
390 |
clear_bu.add([file, md, pdf_show, md_text])
|
391 |
-
|
392 |
download_btn.click(
|
393 |
fn=download_markdown_file,
|
394 |
inputs=md_text,
|
|
|
52 |
|
53 |
|
54 |
def send_pdf_to_parse(file_path, server_ip, port, route="/upload", api_key=None):
|
55 |
+
url = f"{openai_api_base}{route}"
|
56 |
headers = {}
|
57 |
if api_key:
|
58 |
headers["Authorization"] = f"Bearer {api_key}"
|
|
|
65 |
|
66 |
|
67 |
|
68 |
+
async def send_pdf_async_aiohttp(file_path, server_ip, route="/upload", Authorization=None):
|
69 |
"""使用aiohttp异步发送PDF"""
|
70 |
url = f"http://{server_ip}:{port}{route}"
|
71 |
headers = {}
|
72 |
+
if Authorization:
|
73 |
+
headers["Authorization"] = f"Bearer {Authorization}"
|
74 |
|
75 |
try:
|
76 |
async with aiohttp.ClientSession() as session:
|
|
|
100 |
|
101 |
PORT = os.environ.get("PORT")
|
102 |
|
103 |
+
Authorization = os.environ.get("Authorization")
|
104 |
|
105 |
client = AsyncOpenAI(
|
106 |
api_key=openai_api_key,
|
107 |
+
base_url=openai_api_base + "/v1",
|
108 |
)
|
109 |
|
110 |
|
111 |
async def request(messages):
|
112 |
+
|
113 |
chat_completion_from_base64 = await client.chat.completions.create(
|
114 |
messages=messages,
|
115 |
+
extra_headers={
|
116 |
+
"Authorization": f"Bearer {Authorization}"
|
117 |
+
},
|
118 |
model="Qwen2_5VL",
|
119 |
max_completion_tokens=4096,
|
120 |
stream=True,
|
121 |
temperature=0.0,
|
122 |
top_p=0.95
|
123 |
)
|
124 |
+
|
125 |
page = ""
|
126 |
async for chunk in chat_completion_from_base64:
|
127 |
if chunk.choices[0].delta.content:
|
128 |
content = chunk.choices[0].delta.content
|
129 |
+
|
130 |
choice = chunk.choices[0]
|
131 |
if choice.finish_reason is not None:
|
132 |
print(f"end reason = {choice.finish_reason}")
|
133 |
break
|
134 |
page += content
|
135 |
+
|
136 |
yield content
|
137 |
+
|
138 |
|
139 |
def images_to_pdf(img_paths, pdf_path):
|
140 |
|
|
|
168 |
return base64.b64encode(image_file.read()).decode("utf-8")
|
169 |
|
170 |
def build_message(image_path, prompt):
|
171 |
+
|
172 |
content = [
|
173 |
{
|
174 |
"type": "image_url",
|
|
|
178 |
},
|
179 |
{"type": "text", 'text': prompt}
|
180 |
]
|
181 |
+
|
182 |
+
|
183 |
messages = [
|
184 |
{"role": "system", "content": "You are a helpful assistant."},
|
185 |
{'role': 'user', 'content': content}
|
186 |
+
|
187 |
]
|
188 |
+
|
189 |
return messages
|
190 |
|
191 |
|
|
|
214 |
for idx, page in enumerate(pages, start=1):
|
215 |
img_path = tmpdir / f"page_{idx}.png"
|
216 |
page.save(img_path, "PNG")
|
217 |
+
|
218 |
messages = build_message(img_path, prompt)
|
219 |
queries.append(messages)
|
220 |
+
|
221 |
else:
|
222 |
messages = build_message(doc_path, prompt)
|
223 |
queries.append(messages)
|
224 |
+
|
225 |
all_pages = []
|
226 |
all_pages_raw = []
|
227 |
for query in queries:
|
|
|
234 |
print(all_pages)
|
235 |
yield "\n---\n".join(all_pages), "\n\n".join(all_pages_raw)
|
236 |
|
237 |
+
|
238 |
def compress_directory_to_zip(directory_path, output_zip_path):
|
239 |
try:
|
240 |
with zipfile.ZipFile(output_zip_path, 'w', zipfile.ZIP_DEFLATED) as zipf:
|
241 |
|
242 |
+
|
243 |
for root, dirs, files in os.walk(directory_path):
|
244 |
for file in files:
|
245 |
+
|
246 |
file_path = os.path.join(root, file)
|
247 |
|
248 |
arcname = os.path.relpath(file_path, directory_path)
|
249 |
+
|
250 |
zipf.write(file_path, arcname)
|
251 |
return 0
|
252 |
except Exception as e:
|
|
|
266 |
return prompt
|
267 |
|
268 |
def to_file(image_path):
|
269 |
+
|
270 |
if image_path.endswith("Academic_Papers.png"):
|
271 |
image_path = image_path.replace("Academic_Papers.png", "Academic_Papers.pdf")
|
272 |
|
|
|
296 |
images_to_pdf(file_path, tmp_file_path)
|
297 |
else:
|
298 |
tmp_file_path = file_path
|
299 |
+
asyncio.create_task(send_pdf_async_aiohttp(tmp_file_path, server_ip=openai_api_base, Authorization=Authorization))
|
300 |
|
301 |
return str(tmp_file_path)
|
302 |
|
|
|
305 |
with gr.Blocks() as demo:
|
306 |
with gr.Row():
|
307 |
with gr.Column(variant='panel', scale=5):
|
308 |
+
|
309 |
file = gr.File(label='Please upload a PDF or image', file_types=['.pdf', '.png', '.jpeg', '.jpg'], type="filepath")
|
310 |
prompts = gr.Dropdown(
|
311 |
choices=preset_prompts,
|
|
|
323 |
pdf_show = PDF(label='Preview', interactive=False, visible=True, height=800)
|
324 |
|
325 |
|
326 |
+
|
327 |
example_root = os.path.join(os.path.dirname(__file__), 'examples')
|
328 |
images = [
|
329 |
os.path.join(example_root, f)
|
|
|
337 |
file_path = [
|
338 |
os.path.join(example_root, f)
|
339 |
for f in ["Financial_Reports.png", "Books.png", "Magazines.png", "Academic_Papers.png"]
|
340 |
+
|
341 |
]
|
342 |
+
|
343 |
with gr.Row():
|
344 |
for i, label in enumerate(["Financial Reports(IMG)", "Books(IMG)", "Magazines(IMG)", "Academic Papers(PDF)"]):
|
345 |
with gr.Column(scale=1, min_width=120):
|
|
|
351 |
show_download_button=False
|
352 |
)
|
353 |
gr.Button(label).click(fn=to_file, inputs=gr.State(file_path[i]), outputs=file)
|
354 |
+
|
355 |
+
|
356 |
download_btn = gr.Button("⬇️ Generate download link", size="sm")
|
357 |
output_file = gr.File(label='Parse result', interactive=False, elem_id="down-file-box",visible=False)
|
358 |
+
|
359 |
gr.HTML("""
|
360 |
<style>
|
361 |
#down-file-box {
|
|
|
370 |
line_breaks=True)
|
371 |
with gr.Tab('Markdown text'):
|
372 |
md_text = gr.TextArea(lines=45, show_copy_button=True)
|
373 |
+
|
374 |
+
|
375 |
+
|
376 |
file.change(fn=process_file, inputs=file, outputs=pdf_show)
|
377 |
|
378 |
|
|
|
389 |
inputs=[file, prompts],
|
390 |
outputs=[md, md_text]
|
391 |
)
|
392 |
+
|
393 |
clear_bu.add([file, md, pdf_show, md_text])
|
394 |
+
|
395 |
download_btn.click(
|
396 |
fn=download_markdown_file,
|
397 |
inputs=md_text,
|