Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -101,7 +101,42 @@ def enforce_s_u(text: str) -> str:
|
|
| 101 |
return "u"
|
| 102 |
return "s"
|
| 103 |
|
| 104 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 105 |
text = build_prompt(message)
|
| 106 |
inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
| 107 |
do_sample = bool(temperature and temperature > 0.0)
|
|
@@ -114,6 +149,7 @@ def classify_text_stream(message, max_tokens, temperature, top_p):
|
|
| 114 |
eos_token_id=tokenizer.eos_token_id,
|
| 115 |
pad_token_id=tokenizer.eos_token_id
|
| 116 |
)
|
|
|
|
| 117 |
try:
|
| 118 |
streamer = TextIteratorStreamer(tokenizer, skip_special_tokens=True, skip_prompt=True)
|
| 119 |
except TypeError:
|
|
@@ -127,6 +163,9 @@ def classify_text_stream(message, max_tokens, temperature, top_p):
|
|
| 127 |
partial_text = ""
|
| 128 |
token_count = 0
|
| 129 |
start_time = None
|
|
|
|
|
|
|
|
|
|
| 130 |
with torch.inference_mode():
|
| 131 |
thread.start()
|
| 132 |
try:
|
|
@@ -135,7 +174,7 @@ def classify_text_stream(message, max_tokens, temperature, top_p):
|
|
| 135 |
start_time = time.time()
|
| 136 |
partial_text += chunk
|
| 137 |
token_count += 1
|
| 138 |
-
|
| 139 |
finally:
|
| 140 |
thread.join()
|
| 141 |
|
|
@@ -143,48 +182,155 @@ def classify_text_stream(message, max_tokens, temperature, top_p):
|
|
| 143 |
end_time = time.time() if start_time else time.time()
|
| 144 |
duration = max(1e-6, end_time - start_time)
|
| 145 |
tps = token_count / duration if duration > 0 else 0.0
|
| 146 |
-
yield f"{final_label}\n\n⚡ Speed: {tps:.2f} tokens/s"
|
| 147 |
-
|
| 148 |
-
with gr.Blocks() as demo:
|
| 149 |
-
gr.Markdown("# Multilingual Content Moderation Classifier")
|
| 150 |
-
gr.Markdown("Enter any text to classify it as safe ('s') or unsafe ('u').")
|
| 151 |
|
| 152 |
-
|
| 153 |
-
|
| 154 |
-
|
| 155 |
-
|
| 156 |
-
|
| 157 |
-
|
| 158 |
-
|
| 159 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 160 |
|
| 161 |
-
with gr.
|
| 162 |
-
|
| 163 |
-
with gr.Accordion("Advanced Settings", open=False):
|
| 164 |
max_tokens_slider = gr.Slider(
|
| 165 |
-
minimum=1, maximum=
|
|
|
|
|
|
|
| 166 |
)
|
| 167 |
temp_slider = gr.Slider(
|
| 168 |
-
minimum=0.0, maximum=1.0, value=0.
|
|
|
|
|
|
|
| 169 |
)
|
| 170 |
top_p_slider = gr.Slider(
|
| 171 |
-
minimum=0.1, maximum=1.0, value=0.95, step=0.05,
|
|
|
|
|
|
|
| 172 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 173 |
|
| 174 |
-
|
| 175 |
-
examples=[
|
| 176 |
-
["Hello, how are you today?"],
|
| 177 |
-
["I will find you and hurt you."],
|
| 178 |
-
["C'est une belle journée pour apprendre le codage."],
|
| 179 |
-
["I want to die."],
|
| 180 |
-
],
|
| 181 |
-
inputs=text_input
|
| 182 |
-
)
|
| 183 |
-
|
| 184 |
-
submit_button.click(
|
| 185 |
fn=classify_text_stream,
|
| 186 |
inputs=[text_input, max_tokens_slider, temp_slider, top_p_slider],
|
| 187 |
-
outputs=
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 188 |
)
|
| 189 |
|
| 190 |
if __name__ == "__main__":
|
|
@@ -193,4 +339,10 @@ if __name__ == "__main__":
|
|
| 193 |
**tokenizer(["Hi"], return_tensors="pt").to(model.device),
|
| 194 |
max_new_tokens=1, do_sample=False, use_cache=True
|
| 195 |
)
|
| 196 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 101 |
return "u"
|
| 102 |
return "s"
|
| 103 |
|
| 104 |
+
def format_classification_result(classification, tokens_per_second, processing_time):
|
| 105 |
+
if classification == "s":
|
| 106 |
+
status_emoji = "✅"
|
| 107 |
+
status_text = "SAFE"
|
| 108 |
+
status_color = "#22c55e"
|
| 109 |
+
description = "Content appears to be safe and appropriate."
|
| 110 |
+
else:
|
| 111 |
+
status_emoji = "🚫"
|
| 112 |
+
status_text = "UNSAFE"
|
| 113 |
+
status_color = "#ef4444"
|
| 114 |
+
description = "Content may contain inappropriate or harmful material."
|
| 115 |
+
|
| 116 |
+
result_html = f"""
|
| 117 |
+
<div style="text-align: center; padding: 20px; border-radius: 12px;
|
| 118 |
+
background: linear-gradient(135deg, #f8fafc 0%, #e2e8f0 100%);
|
| 119 |
+
border: 2px solid {status_color}; margin: 10px 0;">
|
| 120 |
+
<div style="font-size: 48px; margin-bottom: 10px;">{status_emoji}</div>
|
| 121 |
+
<div style="font-size: 24px; font-weight: bold; color: {status_color}; margin-bottom: 8px;">
|
| 122 |
+
{status_text}
|
| 123 |
+
</div>
|
| 124 |
+
<div style="font-size: 16px; color: #64748b; margin-bottom: 15px;">
|
| 125 |
+
{description}
|
| 126 |
+
</div>
|
| 127 |
+
<div style="display: flex; justify-content: center; gap: 20px; font-size: 14px; color: #475569;">
|
| 128 |
+
<span>⚡ {tokens_per_second:.1f} tok/s</span>
|
| 129 |
+
<span>⏱️ {processing_time:.2f}s</span>
|
| 130 |
+
</div>
|
| 131 |
+
</div>
|
| 132 |
+
"""
|
| 133 |
+
return result_html
|
| 134 |
+
|
| 135 |
+
def classify_text_stream(message, max_tokens, temperature, top_p, progress=gr.Progress()):
|
| 136 |
+
if not message.strip():
|
| 137 |
+
return format_classification_result("s", 0, 0)
|
| 138 |
+
|
| 139 |
+
progress(0, desc="Preparing classification...")
|
| 140 |
text = build_prompt(message)
|
| 141 |
inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
| 142 |
do_sample = bool(temperature and temperature > 0.0)
|
|
|
|
| 149 |
eos_token_id=tokenizer.eos_token_id,
|
| 150 |
pad_token_id=tokenizer.eos_token_id
|
| 151 |
)
|
| 152 |
+
|
| 153 |
try:
|
| 154 |
streamer = TextIteratorStreamer(tokenizer, skip_special_tokens=True, skip_prompt=True)
|
| 155 |
except TypeError:
|
|
|
|
| 163 |
partial_text = ""
|
| 164 |
token_count = 0
|
| 165 |
start_time = None
|
| 166 |
+
|
| 167 |
+
progress(0.3, desc="Processing content...")
|
| 168 |
+
|
| 169 |
with torch.inference_mode():
|
| 170 |
thread.start()
|
| 171 |
try:
|
|
|
|
| 174 |
start_time = time.time()
|
| 175 |
partial_text += chunk
|
| 176 |
token_count += 1
|
| 177 |
+
progress(0.3 + (token_count / max_tokens) * 0.6, desc="Analyzing...")
|
| 178 |
finally:
|
| 179 |
thread.join()
|
| 180 |
|
|
|
|
| 182 |
end_time = time.time() if start_time else time.time()
|
| 183 |
duration = max(1e-6, end_time - start_time)
|
| 184 |
tps = token_count / duration if duration > 0 else 0.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 185 |
|
| 186 |
+
progress(1.0, desc="Complete!")
|
| 187 |
+
|
| 188 |
+
return format_classification_result(final_label, tps, duration)
|
| 189 |
+
|
| 190 |
+
custom_css = """
|
| 191 |
+
.main-container {
|
| 192 |
+
max-width: 1200px !important;
|
| 193 |
+
margin: 0 auto !important;
|
| 194 |
+
}
|
| 195 |
+
|
| 196 |
+
.header-section {
|
| 197 |
+
background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
|
| 198 |
+
padding: 2rem;
|
| 199 |
+
border-radius: 16px;
|
| 200 |
+
margin-bottom: 2rem;
|
| 201 |
+
color: white;
|
| 202 |
+
text-align: center;
|
| 203 |
+
}
|
| 204 |
+
|
| 205 |
+
.classification-panel {
|
| 206 |
+
background: white;
|
| 207 |
+
border-radius: 16px;
|
| 208 |
+
padding: 2rem;
|
| 209 |
+
box-shadow: 0 4px 20px rgba(0, 0, 0, 0.1);
|
| 210 |
+
border: 1px solid #e2e8f0;
|
| 211 |
+
}
|
| 212 |
+
|
| 213 |
+
.example-card {
|
| 214 |
+
transition: transform 0.2s ease;
|
| 215 |
+
}
|
| 216 |
+
|
| 217 |
+
.example-card:hover {
|
| 218 |
+
transform: translateY(-2px);
|
| 219 |
+
}
|
| 220 |
+
|
| 221 |
+
.gradio-container {
|
| 222 |
+
font-family: 'Inter', -apple-system, BlinkMacSystemFont, sans-serif;
|
| 223 |
+
}
|
| 224 |
+
|
| 225 |
+
.input-section {
|
| 226 |
+
background: #f8fafc;
|
| 227 |
+
border-radius: 12px;
|
| 228 |
+
padding: 1.5rem;
|
| 229 |
+
border: 1px solid #e2e8f0;
|
| 230 |
+
}
|
| 231 |
+
"""
|
| 232 |
+
|
| 233 |
+
with gr.Blocks(css=custom_css, title="AI Content Moderator", theme=gr.themes.Soft()) as demo:
|
| 234 |
+
with gr.Column(elem_classes="main-container"):
|
| 235 |
+
gr.HTML("""
|
| 236 |
+
<div class="header-section">
|
| 237 |
+
<h1 style="font-size: 2.5rem; margin-bottom: 0.5rem; font-weight: 700;">
|
| 238 |
+
🛡️ AI Content Moderator
|
| 239 |
+
</h1>
|
| 240 |
+
<p style="font-size: 1.2rem; opacity: 0.9; margin: 0;">
|
| 241 |
+
Advanced multilingual content classification powered by AI
|
| 242 |
+
</p>
|
| 243 |
+
</div>
|
| 244 |
+
""")
|
| 245 |
+
|
| 246 |
+
with gr.Row():
|
| 247 |
+
with gr.Column(scale=3):
|
| 248 |
+
with gr.Group(elem_classes="input-section"):
|
| 249 |
+
gr.Markdown("### 📝 Content Analysis")
|
| 250 |
+
text_input = gr.Textbox(
|
| 251 |
+
label="Text to Analyze",
|
| 252 |
+
placeholder="Enter any text in any language for content moderation analysis...",
|
| 253 |
+
lines=6,
|
| 254 |
+
max_lines=10,
|
| 255 |
+
show_label=False
|
| 256 |
+
)
|
| 257 |
+
|
| 258 |
+
with gr.Row():
|
| 259 |
+
classify_btn = gr.Button(
|
| 260 |
+
"🔍 Analyze Content",
|
| 261 |
+
variant="primary",
|
| 262 |
+
size="lg",
|
| 263 |
+
scale=2
|
| 264 |
+
)
|
| 265 |
+
clear_btn = gr.Button(
|
| 266 |
+
"🗑️ Clear",
|
| 267 |
+
variant="secondary",
|
| 268 |
+
size="lg",
|
| 269 |
+
scale=1
|
| 270 |
+
)
|
| 271 |
+
|
| 272 |
+
with gr.Column(scale=2):
|
| 273 |
+
with gr.Group(elem_classes="classification-panel"):
|
| 274 |
+
gr.Markdown("### 📊 Classification Result")
|
| 275 |
+
result_display = gr.HTML(
|
| 276 |
+
value=format_classification_result("s", 0, 0),
|
| 277 |
+
label="Result"
|
| 278 |
+
)
|
| 279 |
|
| 280 |
+
with gr.Accordion("⚙️ Advanced Configuration", open=False):
|
| 281 |
+
with gr.Row():
|
|
|
|
| 282 |
max_tokens_slider = gr.Slider(
|
| 283 |
+
minimum=1, maximum=10, value=3, step=1,
|
| 284 |
+
label="Max Tokens",
|
| 285 |
+
info="Maximum number of tokens to generate"
|
| 286 |
)
|
| 287 |
temp_slider = gr.Slider(
|
| 288 |
+
minimum=0.0, maximum=1.0, value=0.1, step=0.1,
|
| 289 |
+
label="Temperature",
|
| 290 |
+
info="Controls randomness in generation"
|
| 291 |
)
|
| 292 |
top_p_slider = gr.Slider(
|
| 293 |
+
minimum=0.1, maximum=1.0, value=0.95, step=0.05,
|
| 294 |
+
label="Top-p",
|
| 295 |
+
info="Nucleus sampling parameter"
|
| 296 |
)
|
| 297 |
+
|
| 298 |
+
gr.Markdown("### 💡 Try These Examples")
|
| 299 |
+
|
| 300 |
+
example_data = [
|
| 301 |
+
["Hello, how are you today? I hope you're having a wonderful time!"],
|
| 302 |
+
["I hate you and I will find you and hurt you badly."],
|
| 303 |
+
["C'est une belle journée pour apprendre la programmation et l'intelligence artificielle."],
|
| 304 |
+
["I can't take this anymore. I want to end everything and disappear forever."],
|
| 305 |
+
["¡Hola! Me encanta aprender nuevos idiomas y conocer diferentes culturas."],
|
| 306 |
+
["You're absolutely worthless and nobody will ever love someone like you."]
|
| 307 |
+
]
|
| 308 |
+
|
| 309 |
+
examples = gr.Examples(
|
| 310 |
+
examples=example_data,
|
| 311 |
+
inputs=text_input,
|
| 312 |
+
examples_per_page=6
|
| 313 |
+
)
|
| 314 |
+
|
| 315 |
+
gr.Markdown("""
|
| 316 |
+
---
|
| 317 |
+
<div style="text-align: center; padding: 1rem; color: #64748b; font-size: 0.9rem;">
|
| 318 |
+
<p><strong>🌍 Multilingual Support:</strong> English, Spanish, French, German, and many more languages</p>
|
| 319 |
+
<p><strong>🚀 Real-time Analysis:</strong> Fast content classification with detailed feedback</p>
|
| 320 |
+
<p><strong>🔒 Privacy First:</strong> All processing happens locally on your machine</p>
|
| 321 |
+
</div>
|
| 322 |
+
""")
|
| 323 |
|
| 324 |
+
classify_btn.click(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 325 |
fn=classify_text_stream,
|
| 326 |
inputs=[text_input, max_tokens_slider, temp_slider, top_p_slider],
|
| 327 |
+
outputs=result_display,
|
| 328 |
+
show_progress=True
|
| 329 |
+
)
|
| 330 |
+
|
| 331 |
+
clear_btn.click(
|
| 332 |
+
fn=lambda: ("", format_classification_result("s", 0, 0)),
|
| 333 |
+
outputs=[text_input, result_display]
|
| 334 |
)
|
| 335 |
|
| 336 |
if __name__ == "__main__":
|
|
|
|
| 339 |
**tokenizer(["Hi"], return_tensors="pt").to(model.device),
|
| 340 |
max_new_tokens=1, do_sample=False, use_cache=True
|
| 341 |
)
|
| 342 |
+
print("🚀 Starting AI Content Moderator...")
|
| 343 |
+
demo.queue(max_size=64, concurrency_count=4).launch(
|
| 344 |
+
server_name="0.0.0.0",
|
| 345 |
+
server_port=7860,
|
| 346 |
+
share=False,
|
| 347 |
+
show_error=True
|
| 348 |
+
)
|