Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -26,6 +26,9 @@ from transformers import (
|
|
26 |
|
27 |
# --- Constants and Model Setup ---
|
28 |
MAX_INPUT_TOKEN_LENGTH = 4096
|
|
|
|
|
|
|
29 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
30 |
|
31 |
# --- Prompts for Different Tasks ---
|
@@ -78,57 +81,64 @@ model_i = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
|
78 |
MODEL_ID_I, trust_remote_code=True, torch_dtype=torch.float16
|
79 |
).to(device).eval()
|
80 |
|
81 |
-
|
|
|
82 |
"""
|
83 |
-
|
84 |
-
|
85 |
"""
|
86 |
-
return f'''
|
87 |
-
<div style="display: flex; align-items: center;">
|
88 |
-
<span style="margin-right: 10px; font-size: 14px;">{label}</span>
|
89 |
-
<div style="width: 110px; height: 5px; background-color: #AFEEEE; border-radius: 2px; overflow: hidden;">
|
90 |
-
<div style="width: 100%; height: 100%; background-color: #00FFFF; animation: loading 1.5s linear infinite;"></div>
|
91 |
-
</div>
|
92 |
-
</div>
|
93 |
-
<style>
|
94 |
-
@keyframes loading {{
|
95 |
-
0% {{ transform: translateX(-100%); }}
|
96 |
-
100% {{ transform: translateX(100%); }}
|
97 |
-
}}
|
98 |
-
</style>
|
99 |
-
'''
|
100 |
-
|
101 |
-
# --- Utility Functions ---
|
102 |
-
def layoutjson2md(layout_data: List[Dict]) -> str:
|
103 |
-
"""Converts the structured JSON from Layout Analysis into formatted Markdown."""
|
104 |
markdown_lines = []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
105 |
try:
|
106 |
-
#
|
107 |
-
|
|
|
|
|
108 |
for item in sorted_items:
|
109 |
-
category = item.get('category', '')
|
110 |
text = item.get('text', '')
|
111 |
-
if not text:
|
|
|
112 |
|
113 |
-
if category == 'Title':
|
114 |
-
|
|
|
|
|
115 |
elif category == 'Table':
|
116 |
-
# Handle structured table JSON
|
117 |
if isinstance(text, dict) and 'header' in text and 'rows' in text:
|
118 |
header = '| ' + ' | '.join(map(str, text['header'])) + ' |'
|
119 |
separator = '| ' + ' | '.join(['---'] * len(text['header'])) + ' |'
|
120 |
rows = ['| ' + ' | '.join(map(str, row)) + ' |' for row in text['rows']]
|
121 |
markdown_lines.extend([header, separator] + rows)
|
122 |
markdown_lines.append("\n")
|
123 |
-
else:
|
124 |
markdown_lines.append(f"{text}\n")
|
125 |
else:
|
126 |
-
|
|
|
127 |
except Exception as e:
|
128 |
print(f"Error converting to markdown: {e}")
|
129 |
-
|
|
|
|
|
130 |
return "\n".join(markdown_lines)
|
131 |
|
|
|
132 |
# --- Core Application Logic ---
|
133 |
@spaces.GPU
|
134 |
def process_document_stream(model_name: str, task_choice: str, image: Image.Image, max_new_tokens: int):
|
@@ -158,11 +168,10 @@ def process_document_stream(model_name: str, task_choice: str, image: Image.Imag
|
|
158 |
inputs = processor(text=[prompt_full], images=[image], return_tensors="pt", padding=True, truncation=True, max_length=MAX_INPUT_TOKEN_LENGTH).to(device)
|
159 |
streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
|
160 |
generation_kwargs = {**inputs, "streamer": streamer, "max_new_tokens": max_new_tokens}
|
161 |
-
|
162 |
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
163 |
thread.start()
|
164 |
|
165 |
-
|
166 |
# 4. Stream raw output to the UI in real-time
|
167 |
buffer = ""
|
168 |
for new_text in streamer:
|
@@ -179,24 +188,32 @@ def process_document_stream(model_name: str, task_choice: str, image: Image.Imag
|
|
179 |
try:
|
180 |
json_match = re.search(r'```json\s*([\s\S]+?)\s*```', buffer)
|
181 |
if not json_match:
|
182 |
-
|
183 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
184 |
json_str = json_match.group(1)
|
185 |
layout_data = json.loads(json_str)
|
186 |
markdown_content = layoutjson2md(layout_data)
|
187 |
-
|
188 |
yield buffer, markdown_content, layout_data
|
189 |
except Exception as e:
|
190 |
-
error_md = f"❌ **Error:** Failed to parse Layout JSON.\n\n**Details:**\n`{str(e)}
|
191 |
error_json = {"error": "ProcessingError", "details": str(e), "raw_output": buffer}
|
192 |
yield buffer, error_md, error_json
|
193 |
|
|
|
194 |
# --- Gradio UI Definition ---
|
195 |
def create_gradio_interface():
|
196 |
"""Builds and returns the Gradio web interface."""
|
197 |
css = """
|
198 |
.main-container { max-width: 1400px; margin: 0 auto; }
|
199 |
-
.process-button { border: none !important; color: white !important; font-weight: bold !important; background-color: blue !important;}
|
200 |
.process-button:hover { background-color: darkblue !important; transform: translateY(-2px) !important; box-shadow: 0 4px 8px rgba(0,0,0,0.2) !important; }
|
201 |
"""
|
202 |
with gr.Blocks(theme="bethecloud/storj_theme", css=css) as demo:
|
@@ -208,15 +225,15 @@ def create_gradio_interface():
|
|
208 |
</p>
|
209 |
</div>
|
210 |
""")
|
211 |
-
|
212 |
with gr.Row():
|
213 |
# Left Column (Inputs)
|
214 |
with gr.Column(scale=1):
|
215 |
model_choice = gr.Dropdown(
|
216 |
-
choices=["Camel-Doc-OCR-080125",
|
217 |
-
"MonkeyOCR-Recognition",
|
218 |
"olmOCR-7B-0725",
|
219 |
-
"Nanonets-OCR-s",
|
220 |
"Megalodon-OCR-Sync-0713"
|
221 |
],
|
222 |
label="Select Model", value="Nanonets-OCR-s"
|
@@ -228,7 +245,7 @@ def create_gradio_interface():
|
|
228 |
image_input = gr.Image(label="Upload Image", type="pil", sources=['upload'])
|
229 |
with gr.Accordion("Advanced Settings", open=False):
|
230 |
max_new_tokens = gr.Slider(minimum=512, maximum=8192, value=4096, step=256, label="Max New Tokens")
|
231 |
-
|
232 |
process_btn = gr.Button("🚀 Process Document", variant="primary", elem_classes=["process-button"], size="lg")
|
233 |
clear_btn = gr.Button("🗑️ Clear All", variant="secondary")
|
234 |
|
@@ -242,33 +259,32 @@ def create_gradio_interface():
|
|
242 |
examples=["examples/1.png", "examples/2.png", "examples/3.png", "examples/4.png", "examples/5.png"],
|
243 |
inputs=image_input,
|
244 |
label="Examples"
|
245 |
-
)
|
246 |
-
with gr.Tab("📰
|
247 |
-
|
248 |
-
|
249 |
-
|
250 |
with gr.Tab("📋 Layout Analysis Results"):
|
251 |
json_output = gr.JSON(label="Structured Layout Data (JSON)")
|
252 |
-
|
253 |
# Event Handlers
|
254 |
def clear_all_outputs():
|
255 |
return None, "Raw output will appear here.", "Formatted results will appear here.", None
|
256 |
|
257 |
process_btn.click(
|
258 |
fn=process_document_stream,
|
259 |
-
inputs=[model_choice,
|
260 |
-
task_choice,
|
261 |
-
image_input,
|
262 |
max_new_tokens],
|
263 |
-
outputs=[raw_output_stream,
|
264 |
-
markdown_output,
|
265 |
json_output]
|
266 |
)
|
267 |
clear_btn.click(
|
268 |
clear_all_outputs,
|
269 |
-
outputs=[image_input,
|
270 |
-
raw_output_stream,
|
271 |
-
markdown_output,
|
272 |
json_output]
|
273 |
)
|
274 |
return demo
|
|
|
26 |
|
27 |
# --- Constants and Model Setup ---
|
28 |
MAX_INPUT_TOKEN_LENGTH = 4096
|
29 |
+
# Note: The following line correctly falls back to CPU if CUDA is not available.
|
30 |
+
# The "RuntimeError: CUDA driver initialization failed" is an environment issue,
|
31 |
+
# meaning the code is being run where a GPU is expected but not found/configured.
|
32 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
33 |
|
34 |
# --- Prompts for Different Tasks ---
|
|
|
81 |
MODEL_ID_I, trust_remote_code=True, torch_dtype=torch.float16
|
82 |
).to(device).eval()
|
83 |
|
84 |
+
# --- Utility Functions ---
|
85 |
+
def layoutjson2md(layout_data: Any) -> str:
|
86 |
"""
|
87 |
+
FIXED: Converts the structured JSON from Layout Analysis into formatted Markdown.
|
88 |
+
This version is robust against malformed JSON from the model.
|
89 |
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
90 |
markdown_lines = []
|
91 |
+
|
92 |
+
# If the model wraps the list in a dictionary, find and extract the list.
|
93 |
+
if isinstance(layout_data, dict):
|
94 |
+
found_list = None
|
95 |
+
for value in layout_data.values():
|
96 |
+
if isinstance(value, list):
|
97 |
+
found_list = value
|
98 |
+
break
|
99 |
+
if found_list is not None:
|
100 |
+
layout_data = found_list
|
101 |
+
else:
|
102 |
+
return "### Error: Could not find a list of layout items in the JSON object."
|
103 |
+
|
104 |
+
if not isinstance(layout_data, list):
|
105 |
+
return f"### Error: Expected a list of layout items, but received type {type(layout_data).__name__}."
|
106 |
+
|
107 |
try:
|
108 |
+
# Filter out any non-dictionary items and sort by reading order.
|
109 |
+
valid_items = [item for item in layout_data if isinstance(item, dict)]
|
110 |
+
sorted_items = sorted(valid_items, key=lambda x: (x.get('bbox', [0, 0, 0, 0])[1], x.get('bbox', [0, 0, 0, 0])[0]))
|
111 |
+
|
112 |
for item in sorted_items:
|
113 |
+
category = item.get('category', 'Text') # Default to 'Text' if no category
|
114 |
text = item.get('text', '')
|
115 |
+
if not text:
|
116 |
+
continue
|
117 |
|
118 |
+
if category == 'Title':
|
119 |
+
markdown_lines.append(f"# {text}\n")
|
120 |
+
elif category == 'Section-header':
|
121 |
+
markdown_lines.append(f"## {text}\n")
|
122 |
elif category == 'Table':
|
|
|
123 |
if isinstance(text, dict) and 'header' in text and 'rows' in text:
|
124 |
header = '| ' + ' | '.join(map(str, text['header'])) + ' |'
|
125 |
separator = '| ' + ' | '.join(['---'] * len(text['header'])) + ' |'
|
126 |
rows = ['| ' + ' | '.join(map(str, row)) + ' |' for row in text['rows']]
|
127 |
markdown_lines.extend([header, separator] + rows)
|
128 |
markdown_lines.append("\n")
|
129 |
+
else: # Fallback for simple text or malformed tables
|
130 |
markdown_lines.append(f"{text}\n")
|
131 |
else:
|
132 |
+
markdown_lines.append(f"{text}\n")
|
133 |
+
|
134 |
except Exception as e:
|
135 |
print(f"Error converting to markdown: {e}")
|
136 |
+
traceback.print_exc()
|
137 |
+
return "### Error: An unexpected error occurred while converting JSON to Markdown."
|
138 |
+
|
139 |
return "\n".join(markdown_lines)
|
140 |
|
141 |
+
|
142 |
# --- Core Application Logic ---
|
143 |
@spaces.GPU
|
144 |
def process_document_stream(model_name: str, task_choice: str, image: Image.Image, max_new_tokens: int):
|
|
|
168 |
inputs = processor(text=[prompt_full], images=[image], return_tensors="pt", padding=True, truncation=True, max_length=MAX_INPUT_TOKEN_LENGTH).to(device)
|
169 |
streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
|
170 |
generation_kwargs = {**inputs, "streamer": streamer, "max_new_tokens": max_new_tokens}
|
171 |
+
|
172 |
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
173 |
thread.start()
|
174 |
|
|
|
175 |
# 4. Stream raw output to the UI in real-time
|
176 |
buffer = ""
|
177 |
for new_text in streamer:
|
|
|
188 |
try:
|
189 |
json_match = re.search(r'```json\s*([\s\S]+?)\s*```', buffer)
|
190 |
if not json_match:
|
191 |
+
# If no JSON block is found, try to parse the whole buffer as a fallback.
|
192 |
+
try:
|
193 |
+
layout_data = json.loads(buffer)
|
194 |
+
markdown_content = layoutjson2md(layout_data)
|
195 |
+
yield buffer, markdown_content, layout_data
|
196 |
+
return
|
197 |
+
except json.JSONDecodeError:
|
198 |
+
raise ValueError("JSON object not found in the model's output.")
|
199 |
+
|
200 |
json_str = json_match.group(1)
|
201 |
layout_data = json.loads(json_str)
|
202 |
markdown_content = layoutjson2md(layout_data)
|
203 |
+
|
204 |
yield buffer, markdown_content, layout_data
|
205 |
except Exception as e:
|
206 |
+
error_md = f"❌ **Error:** Failed to parse Layout JSON.\n\n**Details:**\n`{str(e)}`\n\n**Raw Output:**\n```\n{buffer}\n```"
|
207 |
error_json = {"error": "ProcessingError", "details": str(e), "raw_output": buffer}
|
208 |
yield buffer, error_md, error_json
|
209 |
|
210 |
+
|
211 |
# --- Gradio UI Definition ---
|
212 |
def create_gradio_interface():
|
213 |
"""Builds and returns the Gradio web interface."""
|
214 |
css = """
|
215 |
.main-container { max-width: 1400px; margin: 0 auto; }
|
216 |
+
.process-button { border: none !important; color: white !important; font-weight: bold !important; background-color: blue !important;}
|
217 |
.process-button:hover { background-color: darkblue !important; transform: translateY(-2px) !important; box-shadow: 0 4px 8px rgba(0,0,0,0.2) !important; }
|
218 |
"""
|
219 |
with gr.Blocks(theme="bethecloud/storj_theme", css=css) as demo:
|
|
|
225 |
</p>
|
226 |
</div>
|
227 |
""")
|
228 |
+
|
229 |
with gr.Row():
|
230 |
# Left Column (Inputs)
|
231 |
with gr.Column(scale=1):
|
232 |
model_choice = gr.Dropdown(
|
233 |
+
choices=["Camel-Doc-OCR-080125",
|
234 |
+
"MonkeyOCR-Recognition",
|
235 |
"olmOCR-7B-0725",
|
236 |
+
"Nanonets-OCR-s",
|
237 |
"Megalodon-OCR-Sync-0713"
|
238 |
],
|
239 |
label="Select Model", value="Nanonets-OCR-s"
|
|
|
245 |
image_input = gr.Image(label="Upload Image", type="pil", sources=['upload'])
|
246 |
with gr.Accordion("Advanced Settings", open=False):
|
247 |
max_new_tokens = gr.Slider(minimum=512, maximum=8192, value=4096, step=256, label="Max New Tokens")
|
248 |
+
|
249 |
process_btn = gr.Button("🚀 Process Document", variant="primary", elem_classes=["process-button"], size="lg")
|
250 |
clear_btn = gr.Button("🗑️ Clear All", variant="secondary")
|
251 |
|
|
|
259 |
examples=["examples/1.png", "examples/2.png", "examples/3.png", "examples/4.png", "examples/5.png"],
|
260 |
inputs=image_input,
|
261 |
label="Examples"
|
262 |
+
)
|
263 |
+
with gr.Tab("📰 Formatted Result"):
|
264 |
+
markdown_output = gr.Markdown(label="Formatted Markdown")
|
265 |
+
|
|
|
266 |
with gr.Tab("📋 Layout Analysis Results"):
|
267 |
json_output = gr.JSON(label="Structured Layout Data (JSON)")
|
268 |
+
|
269 |
# Event Handlers
|
270 |
def clear_all_outputs():
|
271 |
return None, "Raw output will appear here.", "Formatted results will appear here.", None
|
272 |
|
273 |
process_btn.click(
|
274 |
fn=process_document_stream,
|
275 |
+
inputs=[model_choice,
|
276 |
+
task_choice,
|
277 |
+
image_input,
|
278 |
max_new_tokens],
|
279 |
+
outputs=[raw_output_stream,
|
280 |
+
markdown_output,
|
281 |
json_output]
|
282 |
)
|
283 |
clear_btn.click(
|
284 |
clear_all_outputs,
|
285 |
+
outputs=[image_input,
|
286 |
+
raw_output_stream,
|
287 |
+
markdown_output,
|
288 |
json_output]
|
289 |
)
|
290 |
return demo
|