Update app.py
Browse files
app.py
CHANGED
|
@@ -6,611 +6,413 @@ import spaces
|
|
| 6 |
import warnings
|
| 7 |
warnings.filterwarnings("ignore")
|
| 8 |
|
| 9 |
-
from
|
| 10 |
-
from
|
| 11 |
|
| 12 |
# =========================================================
|
| 13 |
-
#
|
| 14 |
# =========================================================
|
| 15 |
|
| 16 |
-
|
| 17 |
-
"""
|
| 18 |
-
تجاوز مشكلة lm_head في tied weights
|
| 19 |
-
"""
|
| 20 |
-
return
|
| 21 |
-
|
| 22 |
-
def _patched_init_weights(self, module):
|
| 23 |
-
"""
|
| 24 |
-
تجاوز مشكلة initializer_range في Qwen3OmniMoeTalkerConfig
|
| 25 |
-
"""
|
| 26 |
-
# نحاول الحصول على initializer_range، وإذا لم يكن موجود نستخدم قيمة افتراضية
|
| 27 |
-
try:
|
| 28 |
-
std = self.config.initializer_range
|
| 29 |
-
except AttributeError:
|
| 30 |
-
# قيمة افتراضية آمنة
|
| 31 |
-
std = 0.02
|
| 32 |
-
|
| 33 |
-
# تطبيق التهيئة الأساسية
|
| 34 |
-
if isinstance(module, torch.nn.Linear):
|
| 35 |
-
module.weight.data.normal_(mean=0.0, std=std)
|
| 36 |
-
if module.bias is not None:
|
| 37 |
-
module.bias.data.zero_()
|
| 38 |
-
elif isinstance(module, torch.nn.Embedding):
|
| 39 |
-
module.weight.data.normal_(mean=0.0, std=std)
|
| 40 |
-
if module.padding_idx is not None:
|
| 41 |
-
module.weight.data[module.padding_idx].zero_()
|
| 42 |
-
|
| 43 |
-
def _patched_initialize_weights(self):
|
| 44 |
-
"""
|
| 45 |
-
تجاوز كامل لدالة initialize_weights في حالة استمرار المشاكل
|
| 46 |
-
"""
|
| 47 |
-
# نحاول التهيئة العادية، وإذا فشلت نتجاهلها
|
| 48 |
-
try:
|
| 49 |
-
# محاولة استخدام الدالة الأصلية إذا كانت موجودة
|
| 50 |
-
if hasattr(self, '_original_initialize_weights'):
|
| 51 |
-
self._original_initialize_weights()
|
| 52 |
-
else:
|
| 53 |
-
# تهيئة بسيطة آمنة
|
| 54 |
-
for module in self.modules():
|
| 55 |
-
if isinstance(module, (torch.nn.Linear, torch.nn.Embedding)):
|
| 56 |
-
if hasattr(module, 'weight'):
|
| 57 |
-
module.weight.data.normal_(mean=0.0, std=0.02)
|
| 58 |
-
if hasattr(module, 'bias') and module.bias is not None:
|
| 59 |
-
module.bias.data.zero_()
|
| 60 |
-
except Exception as e:
|
| 61 |
-
print(f"Warning: Could not initialize weights properly: {e}")
|
| 62 |
-
# نستمر بدون تهيئة - النموذج المحمل مسبقاً يجب أن يعمل
|
| 63 |
-
|
| 64 |
-
# تطبيق الـ patches قبل أي استدعاء لـ from_pretrained
|
| 65 |
-
def apply_patches():
|
| 66 |
-
"""تطبيق جميع الـ patches اللازمة"""
|
| 67 |
-
|
| 68 |
-
# Patch 1: تجاوز مشكلة lm_head
|
| 69 |
-
if hasattr(Qwen3OmniMoeForConditionalGeneration, "mark_tied_weights_as_initialized"):
|
| 70 |
-
Qwen3OmniMoeForConditionalGeneration.mark_tied_weights_as_initialized = (
|
| 71 |
-
_patched_mark_tied_weights_as_initialized
|
| 72 |
-
)
|
| 73 |
-
|
| 74 |
-
# Patch 2: تجاوز مشكلة initializer_range
|
| 75 |
-
if hasattr(Qwen3OmniMoeForConditionalGeneration, "_init_weights"):
|
| 76 |
-
Qwen3OmniMoeForConditionalGeneration._init_weights = _patched_init_weights
|
| 77 |
-
|
| 78 |
-
# Patch 3: تجاوز initialize_weights بالكامل إذا لزم
|
| 79 |
-
if hasattr(Qwen3OmniMoeForConditionalGeneration, "initialize_weights"):
|
| 80 |
-
# حفظ الدالة الأصلية
|
| 81 |
-
Qwen3OmniMoeForConditionalGeneration._original_initialize_weights = (
|
| 82 |
-
Qwen3OmniMoeForConditionalGeneration.initialize_weights
|
| 83 |
-
)
|
| 84 |
-
Qwen3OmniMoeForConditionalGeneration.initialize_weights = _patched_initialize_weights
|
| 85 |
-
|
| 86 |
-
# تطبيق الـ patches
|
| 87 |
-
apply_patches()
|
| 88 |
-
|
| 89 |
-
# =========================================================
|
| 90 |
-
# إعدادات عامة
|
| 91 |
-
# =========================================================
|
| 92 |
|
| 93 |
-
|
| 94 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 95 |
|
| 96 |
-
VOICE_CHOICES = ["Ethan", "Chelsie", "Aiden"]
|
| 97 |
-
DEFAULT_VOICE = "Ethan"
|
| 98 |
-
|
| 99 |
-
# سنحمّل النموذج كسولياً (عند أول استدعاء فقط)
|
| 100 |
model = None
|
| 101 |
-
|
| 102 |
|
| 103 |
|
| 104 |
def load_model():
|
| 105 |
"""
|
| 106 |
-
تحميل
|
| 107 |
-
- نستخدم attn_implementation="eager" لتفادي الحاجة لـ flash-attn.
|
| 108 |
-
- نضيف low_cpu_mem_usage=True لتحسين الأداء
|
| 109 |
-
- نضيف ignore_mismatched_sizes=True لتجاوز مشاكل الأحجام
|
| 110 |
"""
|
| 111 |
-
global model,
|
| 112 |
-
|
| 113 |
-
if model is not None and
|
| 114 |
return
|
| 115 |
-
|
| 116 |
-
print(f"[ZeroGPU] Loading
|
| 117 |
-
|
| 118 |
-
# اختيار
|
| 119 |
if torch.cuda.is_available():
|
| 120 |
-
torch_dtype = torch.bfloat16
|
| 121 |
device = "cuda"
|
|
|
|
| 122 |
else:
|
|
|
|
| 123 |
torch_dtype = torch.float32
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
# محاولة بديلة مع تعطيل _init_weights
|
| 148 |
-
try:
|
| 149 |
-
# تعطيل _init_weights مؤقتاً
|
| 150 |
-
original_init_weights = None
|
| 151 |
-
if hasattr(Qwen3OmniMoeForConditionalGeneration, "_init_weights"):
|
| 152 |
-
original_init_weights = Qwen3OmniMoeForConditionalGeneration._init_weights
|
| 153 |
-
Qwen3OmniMoeForConditionalGeneration._init_weights = lambda self, module: None
|
| 154 |
-
|
| 155 |
-
local_model = Qwen3OmniMoeForConditionalGeneration.from_pretrained(
|
| 156 |
-
MODEL_PATH,
|
| 157 |
-
torch_dtype=torch_dtype,
|
| 158 |
-
attn_implementation="eager",
|
| 159 |
-
low_cpu_mem_usage=True,
|
| 160 |
-
)
|
| 161 |
-
|
| 162 |
-
# استعادة _init_weights
|
| 163 |
-
if original_init_weights:
|
| 164 |
-
Qwen3OmniMoeForConditionalGeneration._init_weights = original_init_weights
|
| 165 |
-
|
| 166 |
-
local_model = local_model.to(device)
|
| 167 |
-
local_model.eval()
|
| 168 |
-
|
| 169 |
-
except Exception as e2:
|
| 170 |
-
raise RuntimeError(f"Failed to load model: {e2}")
|
| 171 |
-
|
| 172 |
-
# تحميل المعالج
|
| 173 |
-
try:
|
| 174 |
-
local_processor = Qwen3OmniMoeProcessor.from_pretrained(
|
| 175 |
-
MODEL_PATH,
|
| 176 |
-
trust_remote_code=True
|
| 177 |
-
)
|
| 178 |
-
except Exception as e:
|
| 179 |
-
print(f"Error loading processor: {e}")
|
| 180 |
-
raise
|
| 181 |
-
|
| 182 |
-
model = local_model
|
| 183 |
-
processor = local_processor
|
| 184 |
-
print(f"[ZeroGPU] Model loaded successfully on {device} with dtype {torch_dtype}.")
|
| 185 |
-
|
| 186 |
-
|
| 187 |
-
def build_messages_from_history(
|
| 188 |
-
history,
|
| 189 |
-
system_prompt,
|
| 190 |
-
user_text,
|
| 191 |
-
image,
|
| 192 |
-
audio_path,
|
| 193 |
-
video_path,
|
| 194 |
-
):
|
| 195 |
-
"""
|
| 196 |
-
تحويل تاريخ الدردشة + المدخل الحالي إلى conversation بالـ format
|
| 197 |
-
المطلوب من Qwen3-Omni.
|
| 198 |
-
history: list of [user_text, assistant_text]
|
| 199 |
-
"""
|
| 200 |
-
messages = []
|
| 201 |
-
|
| 202 |
-
if system_prompt:
|
| 203 |
-
messages.append(
|
| 204 |
-
{
|
| 205 |
-
"role": "system",
|
| 206 |
-
"content": [{"type": "text", "text": system_prompt}],
|
| 207 |
-
}
|
| 208 |
-
)
|
| 209 |
-
|
| 210 |
-
# تاريخ المحادثة
|
| 211 |
-
for user_msg, assistant_msg in history:
|
| 212 |
-
if user_msg:
|
| 213 |
-
messages.append(
|
| 214 |
-
{
|
| 215 |
-
"role": "user",
|
| 216 |
-
"content": [{"type": "text", "text": user_msg}],
|
| 217 |
-
}
|
| 218 |
-
)
|
| 219 |
-
if assistant_msg:
|
| 220 |
-
messages.append(
|
| 221 |
-
{
|
| 222 |
-
"role": "assistant",
|
| 223 |
-
"content": [{"type": "text", "text": assistant_msg}],
|
| 224 |
-
}
|
| 225 |
-
)
|
| 226 |
-
|
| 227 |
-
# محتوى رسالة المستخدم الحالية
|
| 228 |
-
user_content = []
|
| 229 |
-
|
| 230 |
-
if image is not None:
|
| 231 |
-
user_content.append({"type": "image", "image": image})
|
| 232 |
-
|
| 233 |
-
if audio_path is not None and audio_path != "":
|
| 234 |
-
user_content.append({"type": "audio", "audio": audio_path})
|
| 235 |
-
|
| 236 |
-
if video_path is not None and video_path != "":
|
| 237 |
-
user_content.append({"type": "video", "video": video_path})
|
| 238 |
-
|
| 239 |
-
if user_text and user_text.strip():
|
| 240 |
-
user_content.append({"type": "text", "text": user_text.strip()})
|
| 241 |
|
| 242 |
-
|
| 243 |
-
|
| 244 |
-
|
| 245 |
-
|
| 246 |
-
"content": user_content,
|
| 247 |
-
}
|
| 248 |
-
)
|
| 249 |
|
| 250 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 251 |
|
| 252 |
|
| 253 |
# =========================================================
|
| 254 |
-
# دالة الاستدلال (
|
| 255 |
# =========================================================
|
| 256 |
|
| 257 |
@spaces.GPU(duration=120)
|
| 258 |
-
def
|
| 259 |
-
|
| 260 |
-
|
| 261 |
-
|
| 262 |
-
|
| 263 |
-
|
| 264 |
-
system_prompt,
|
| 265 |
-
return_audio,
|
| 266 |
-
speaker,
|
| 267 |
temperature,
|
| 268 |
top_p,
|
| 269 |
-
|
|
|
|
|
|
|
| 270 |
):
|
| 271 |
"""
|
| 272 |
-
|
| 273 |
-
|
| 274 |
-
- مخرج نصي دائماً، ومخرج صوتي اختياري.
|
| 275 |
"""
|
| 276 |
-
|
| 277 |
-
|
| 278 |
-
|
| 279 |
-
|
| 280 |
-
|
| 281 |
-
|
| 282 |
-
|
| 283 |
-
|
| 284 |
-
|
| 285 |
-
|
| 286 |
-
|
| 287 |
-
system_prompt=system_prompt,
|
| 288 |
-
user_text=user_text,
|
| 289 |
-
image=image,
|
| 290 |
-
audio_path=audio_path,
|
| 291 |
-
video_path=video_path,
|
| 292 |
-
)
|
| 293 |
-
|
| 294 |
-
# بناء نص المحادثة باستخدام chat_template
|
| 295 |
-
text_prompt = processor.apply_chat_template(
|
| 296 |
-
messages,
|
| 297 |
-
add_generation_prompt=True,
|
| 298 |
-
tokenize=False,
|
| 299 |
-
)
|
| 300 |
-
|
| 301 |
-
# تجهيز الوسائط المتعددة
|
| 302 |
-
audios, images, videos = process_mm_info(
|
| 303 |
-
messages,
|
| 304 |
-
use_audio_in_video=USE_AUDIO_IN_VIDEO,
|
| 305 |
-
)
|
| 306 |
-
|
| 307 |
-
# تحويل إلى تينسورات
|
| 308 |
-
inputs = processor(
|
| 309 |
-
text=text_prompt,
|
| 310 |
-
audio=audios,
|
| 311 |
-
images=images,
|
| 312 |
-
videos=videos,
|
| 313 |
-
return_tensors="pt",
|
| 314 |
-
padding=True,
|
| 315 |
-
use_audio_in_video=USE_AUDIO_IN_VIDEO,
|
| 316 |
-
)
|
| 317 |
-
|
| 318 |
-
# نقل إلى جهاز النموذج ونفس dtype
|
| 319 |
-
first_param = next(model.parameters())
|
| 320 |
-
device = first_param.device
|
| 321 |
|
| 322 |
-
|
| 323 |
-
|
| 324 |
-
|
| 325 |
-
inputs[key] = inputs[key].to(device)
|
| 326 |
-
|
| 327 |
-
# إعدادات التوليد
|
| 328 |
-
gen_kwargs = dict(
|
| 329 |
-
temperature=float(temperature) if temperature > 0 else 1e-7,
|
| 330 |
-
top_p=float(top_p),
|
| 331 |
-
max_new_tokens=int(max_tokens),
|
| 332 |
-
do_sample=temperature > 0,
|
| 333 |
-
use_audio_in_video=USE_AUDIO_IN_VIDEO,
|
| 334 |
-
)
|
| 335 |
|
| 336 |
-
|
| 337 |
-
if
|
| 338 |
-
|
| 339 |
-
|
| 340 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 341 |
with torch.no_grad():
|
| 342 |
-
if
|
| 343 |
-
|
| 344 |
-
|
| 345 |
-
|
| 346 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 347 |
else:
|
| 348 |
-
|
| 349 |
-
|
| 350 |
-
|
| 351 |
-
|
| 352 |
-
|
| 353 |
-
|
| 354 |
-
|
| 355 |
-
|
| 356 |
-
|
| 357 |
-
|
| 358 |
-
|
| 359 |
-
|
| 360 |
-
|
| 361 |
-
if hasattr(text_ids, 'sequences'):
|
| 362 |
-
generated_ids = text_ids.sequences[:, input_len:]
|
| 363 |
-
else:
|
| 364 |
-
generated_ids = text_ids[:, input_len:]
|
| 365 |
|
| 366 |
-
|
| 367 |
-
|
| 368 |
-
|
| 369 |
-
clean_up_tokenization_spaces=False,
|
| 370 |
-
)[0]
|
| 371 |
-
|
| 372 |
-
# تحديث تاريخ الدردشة
|
| 373 |
-
user_display = (
|
| 374 |
-
user_text if (user_text and user_text.strip()) else "[Multimodal message]"
|
| 375 |
-
)
|
| 376 |
-
history = history + [[user_display, generated_text]]
|
| 377 |
-
|
| 378 |
-
# تجهيز الصوت الناتج إن وجد
|
| 379 |
-
gr_audio = None
|
| 380 |
-
if audio_out is not None:
|
| 381 |
try:
|
| 382 |
-
|
| 383 |
-
|
| 384 |
-
|
|
|
|
|
|
|
| 385 |
except Exception as e:
|
| 386 |
-
print(f"
|
| 387 |
-
|
| 388 |
-
|
| 389 |
-
|
| 390 |
-
return history, gr_audio, "", None, None, None
|
| 391 |
|
| 392 |
except Exception as e:
|
| 393 |
-
print(f"Error during inference: {e}")
|
| 394 |
import traceback
|
| 395 |
traceback.print_exc()
|
| 396 |
-
|
| 397 |
-
# في حالة الخطأ، نضيف رسالة خطأ للمحادثة
|
| 398 |
-
user_display = (
|
| 399 |
-
user_text if (user_text and user_text.strip()) else "[Multimodal message]"
|
| 400 |
-
)
|
| 401 |
-
error_message = f"عذراً، حدث خطأ أثناء معالجة الرسالة: {str(e)}"
|
| 402 |
-
history = history + [[user_display, error_message]]
|
| 403 |
-
return history, None, "", None, None, None
|
| 404 |
|
| 405 |
|
| 406 |
# =========================================================
|
| 407 |
-
#
|
| 408 |
# =========================================================
|
| 409 |
|
| 410 |
-
def clear_chat():
|
| 411 |
-
"""إعادة تعيين المحادثة ومخرج الصوت."""
|
| 412 |
-
return [], None
|
| 413 |
-
|
| 414 |
-
|
| 415 |
def create_interface():
|
| 416 |
-
"""إنشاء واجهة Gradio
|
| 417 |
-
|
|
|
|
| 418 |
gr.Markdown(
|
| 419 |
"""
|
| 420 |
-
|
| 421 |
-
|
| 422 |
-
|
| 423 |
-
|
| 424 |
-
|
| 425 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 426 |
"""
|
| 427 |
)
|
| 428 |
-
|
| 429 |
with gr.Row():
|
| 430 |
-
# العمود الأيسر: المحادثة
|
| 431 |
with gr.Column(scale=3):
|
| 432 |
-
|
| 433 |
-
|
| 434 |
-
|
| 435 |
-
|
|
|
|
|
|
|
| 436 |
)
|
| 437 |
-
|
| 438 |
-
|
| 439 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 440 |
type="numpy",
|
| 441 |
-
|
| 442 |
-
visible=True,
|
| 443 |
)
|
| 444 |
-
|
| 445 |
-
with gr.Row():
|
| 446 |
-
user_text = gr.Textbox(
|
| 447 |
-
label="رسالتك",
|
| 448 |
-
placeholder="اكتب رسالتك هنا (يمكنك أيضاً إرفاق صورة/صوت/فيديو من الأسفل)...",
|
| 449 |
-
lines=3,
|
| 450 |
-
show_label=False,
|
| 451 |
-
elem_id="message",
|
| 452 |
-
)
|
| 453 |
-
|
| 454 |
-
with gr.Row():
|
| 455 |
-
image_input = gr.Image(
|
| 456 |
-
label="📷 صورة (اختياري)",
|
| 457 |
-
type="pil",
|
| 458 |
-
sources=["upload", "webcam"],
|
| 459 |
-
height=150,
|
| 460 |
-
)
|
| 461 |
-
audio_input = gr.Audio(
|
| 462 |
-
label="🎙️ صوت (اختياري)",
|
| 463 |
-
type="filepath",
|
| 464 |
-
sources=["microphone", "upload"],
|
| 465 |
-
)
|
| 466 |
-
video_input = gr.Video(
|
| 467 |
-
label="🎬 فيديو (اختياري)",
|
| 468 |
-
height=150,
|
| 469 |
-
)
|
| 470 |
-
|
| 471 |
-
with gr.Row():
|
| 472 |
-
send_btn = gr.Button("إرسال", variant="primary", scale=2)
|
| 473 |
-
clear_btn = gr.Button("مسح المحادثة", variant="secondary")
|
| 474 |
-
|
| 475 |
-
# العمود الأيمن: الإعدادات
|
| 476 |
with gr.Column(scale=1):
|
| 477 |
-
gr.Markdown("### ⚙️
|
| 478 |
-
|
| 479 |
-
|
| 480 |
-
label="
|
| 481 |
-
|
| 482 |
-
|
| 483 |
-
|
|
|
|
| 484 |
)
|
| 485 |
-
|
| 486 |
-
|
| 487 |
-
label="
|
| 488 |
-
|
|
|
|
|
|
|
|
|
|
| 489 |
)
|
| 490 |
-
|
| 491 |
-
|
| 492 |
-
label="
|
| 493 |
-
|
| 494 |
-
|
|
|
|
|
|
|
| 495 |
)
|
| 496 |
-
|
| 497 |
-
|
| 498 |
-
|
| 499 |
-
|
| 500 |
-
|
| 501 |
-
|
| 502 |
-
|
| 503 |
-
step=0.05,
|
| 504 |
-
)
|
| 505 |
-
|
| 506 |
-
top_p = gr.Slider(
|
| 507 |
-
label="Top-p (حجم العينة)",
|
| 508 |
-
minimum=0.1,
|
| 509 |
-
maximum=1.0,
|
| 510 |
-
value=0.95,
|
| 511 |
-
step=0.05,
|
| 512 |
-
)
|
| 513 |
-
|
| 514 |
-
max_tokens = gr.Slider(
|
| 515 |
-
label="Max new tokens (طول الرد الأقصى)",
|
| 516 |
-
minimum=16,
|
| 517 |
-
maximum=1024,
|
| 518 |
-
value=384,
|
| 519 |
-
step=16,
|
| 520 |
-
)
|
| 521 |
-
|
| 522 |
-
gr.Markdown(
|
| 523 |
-
"""
|
| 524 |
-
**📝 ملاحظات:**
|
| 525 |
-
- يمكنك إرسال نص فقط، أو نص مع صورة/صوت/فيديو في رسالة واحدة
|
| 526 |
-
- Enter للإرسال، Shift+Enter لسطر جديد
|
| 527 |
-
- تشغيل النموذج على ZeroGPU قد يستغرق عدة ثوانٍ حسب طول الرسالة
|
| 528 |
-
- النموذج يدعم اللغات المتعددة بما فيها العربية والإنجليزية
|
| 529 |
-
"""
|
| 530 |
)
|
| 531 |
-
|
| 532 |
-
|
| 533 |
-
|
| 534 |
-
|
| 535 |
-
|
| 536 |
-
|
| 537 |
-
|
| 538 |
-
|
| 539 |
-
|
| 540 |
-
|
| 541 |
-
|
| 542 |
-
|
| 543 |
-
|
| 544 |
-
|
| 545 |
-
|
| 546 |
-
|
| 547 |
-
|
| 548 |
-
|
| 549 |
-
|
| 550 |
-
|
| 551 |
-
history_state,
|
| 552 |
-
audio_output,
|
| 553 |
-
user_text,
|
| 554 |
-
image_input,
|
| 555 |
-
audio_input,
|
| 556 |
-
video_input,
|
| 557 |
-
]
|
| 558 |
-
|
| 559 |
-
# إرسال بالزر
|
| 560 |
-
send_btn.click(
|
| 561 |
-
fn=qwen3_omni_inference,
|
| 562 |
-
inputs=send_inputs,
|
| 563 |
-
outputs=send_outputs,
|
| 564 |
-
queue=True,
|
| 565 |
-
).then(
|
| 566 |
-
lambda h: h,
|
| 567 |
-
inputs=history_state,
|
| 568 |
-
outputs=chatbot,
|
| 569 |
)
|
| 570 |
-
|
| 571 |
-
#
|
| 572 |
-
|
| 573 |
-
fn=
|
| 574 |
-
|
| 575 |
-
|
| 576 |
-
|
| 577 |
-
|
| 578 |
-
|
| 579 |
-
inputs=history_state,
|
| 580 |
-
outputs=chatbot,
|
| 581 |
)
|
| 582 |
-
|
| 583 |
-
#
|
| 584 |
-
|
| 585 |
-
fn=
|
| 586 |
-
inputs=
|
| 587 |
-
|
| 588 |
-
|
| 589 |
-
|
| 590 |
-
|
| 591 |
-
|
| 592 |
-
|
| 593 |
-
|
| 594 |
-
|
| 595 |
-
|
|
|
|
|
|
|
|
|
|
| 596 |
)
|
| 597 |
-
|
| 598 |
-
#
|
| 599 |
-
|
| 600 |
-
|
| 601 |
-
|
| 602 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 603 |
)
|
| 604 |
-
|
| 605 |
return demo
|
| 606 |
|
| 607 |
|
| 608 |
-
#
|
| 609 |
-
|
|
|
|
| 610 |
|
| 611 |
if __name__ == "__main__":
|
| 612 |
-
|
| 613 |
demo.launch(
|
| 614 |
-
ssr_mode=False,
|
| 615 |
-
show_error=True,
|
|
|
|
| 616 |
)
|
|
|
|
| 6 |
import warnings
|
| 7 |
warnings.filterwarnings("ignore")
|
| 8 |
|
| 9 |
+
from PIL import Image
|
| 10 |
+
from transformers import AutoModel, AutoTokenizer
|
| 11 |
|
| 12 |
# =========================================================
|
| 13 |
+
# إعدادات النموذج
|
| 14 |
# =========================================================
|
| 15 |
|
| 16 |
+
MODEL_PATH = "openbmb/MiniCPM-o-2_6"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
+
# النموذج يدعم:
|
| 19 |
+
# - Vision (الصور)
|
| 20 |
+
# - Audio (الصوت)
|
| 21 |
+
# - TTS (تحويل النص إلى كلام)
|
| 22 |
+
# - ASR (التعرف على الكلام)
|
| 23 |
+
# - Video (الفيديو)
|
| 24 |
+
# - Voice Cloning (استنساخ الصوت)
|
| 25 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
model = None
|
| 27 |
+
tokenizer = None
|
| 28 |
|
| 29 |
|
| 30 |
def load_model():
|
| 31 |
"""
|
| 32 |
+
تحميل MiniCPM-o-2_6 مع دعم جميع الوسائط
|
|
|
|
|
|
|
|
|
|
| 33 |
"""
|
| 34 |
+
global model, tokenizer
|
| 35 |
+
|
| 36 |
+
if model is not None and tokenizer is not None:
|
| 37 |
return
|
| 38 |
+
|
| 39 |
+
print(f"[ZeroGPU] Loading MiniCPM-o-2_6...")
|
| 40 |
+
|
| 41 |
+
# اختيار الجهاز ونوع البيانات
|
| 42 |
if torch.cuda.is_available():
|
|
|
|
| 43 |
device = "cuda"
|
| 44 |
+
torch_dtype = torch.bfloat16
|
| 45 |
else:
|
| 46 |
+
device = "cpu"
|
| 47 |
torch_dtype = torch.float32
|
| 48 |
+
|
| 49 |
+
# تحميل النموذج مع جميع القدرات
|
| 50 |
+
model = AutoModel.from_pretrained(
|
| 51 |
+
MODEL_PATH,
|
| 52 |
+
trust_remote_code=True,
|
| 53 |
+
attn_implementation='sdpa', # sdpa أو flash_attention_2
|
| 54 |
+
torch_dtype=torch_dtype,
|
| 55 |
+
init_vision=True, # تفعيل الرؤية
|
| 56 |
+
init_audio=True, # تفعيل الصوت
|
| 57 |
+
init_tts=True # تفعيل TTS
|
| 58 |
+
)
|
| 59 |
+
|
| 60 |
+
model = model.eval().to(device)
|
| 61 |
+
|
| 62 |
+
# تحميل tokenizer
|
| 63 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
| 64 |
+
MODEL_PATH,
|
| 65 |
+
trust_remote_code=True
|
| 66 |
+
)
|
| 67 |
+
|
| 68 |
+
print(f"[ZeroGPU] Model loaded successfully on {device}")
|
| 69 |
+
```<!--citation:1-->
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 70 |
|
| 71 |
+
```python
|
| 72 |
+
# =========================================================
|
| 73 |
+
# دالة معالجة الصور
|
| 74 |
+
# =========================================================
|
|
|
|
|
|
|
|
|
|
| 75 |
|
| 76 |
+
def process_image(image_path_or_pil):
|
| 77 |
+
"""معالجة الصورة للنموذج"""
|
| 78 |
+
if isinstance(image_path_or_pil, str):
|
| 79 |
+
image = Image.open(image_path_or_pil).convert('RGB')
|
| 80 |
+
else:
|
| 81 |
+
image = image_path_or_pil.convert('RGB')
|
| 82 |
+
return image
|
| 83 |
|
| 84 |
|
| 85 |
# =========================================================
|
| 86 |
+
# دالة الاستدلال الرئيسية (مع دعم ZeroGPU)
|
| 87 |
# =========================================================
|
| 88 |
|
| 89 |
@spaces.GPU(duration=120)
|
| 90 |
+
def minicpm_o_inference(
|
| 91 |
+
text_input,
|
| 92 |
+
image_input,
|
| 93 |
+
audio_input,
|
| 94 |
+
video_input,
|
| 95 |
+
mode,
|
|
|
|
|
|
|
|
|
|
| 96 |
temperature,
|
| 97 |
top_p,
|
| 98 |
+
max_new_tokens,
|
| 99 |
+
enable_tts,
|
| 100 |
+
tts_style
|
| 101 |
):
|
| 102 |
"""
|
| 103 |
+
دالة الاستدلال الرئيسية لـ MiniCPM-o-2_6
|
| 104 |
+
تدعم: نص، صورة، صوت، فيديو
|
|
|
|
| 105 |
"""
|
| 106 |
+
|
| 107 |
+
load_model()
|
| 108 |
+
global model, tokenizer
|
| 109 |
+
|
| 110 |
+
# بناء الرسائل حسب نوع المدخل
|
| 111 |
+
messages = []
|
| 112 |
+
|
| 113 |
+
# إضافة المحتوى حسب نوع المدخل
|
| 114 |
+
if mode == "Text Only":
|
| 115 |
+
if not text_input:
|
| 116 |
+
return "Please provide text input.", None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 117 |
|
| 118 |
+
messages = [
|
| 119 |
+
{"role": "user", "content": text_input}
|
| 120 |
+
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 121 |
|
| 122 |
+
elif mode == "Image + Text":
|
| 123 |
+
if not image_input:
|
| 124 |
+
return "Please provide an image.", None
|
| 125 |
+
|
| 126 |
+
image = process_image(image_input)
|
| 127 |
+
|
| 128 |
+
# صياغة السؤال
|
| 129 |
+
question = text_input if text_input else "What is shown in this image?"
|
| 130 |
+
|
| 131 |
+
messages = [
|
| 132 |
+
{
|
| 133 |
+
"role": "user",
|
| 134 |
+
"content": [
|
| 135 |
+
Image.open(image_input) if isinstance(image_input, str) else image_input,
|
| 136 |
+
question
|
| 137 |
+
]
|
| 138 |
+
}
|
| 139 |
+
]
|
| 140 |
+
|
| 141 |
+
elif mode == "Audio + Text":
|
| 142 |
+
if not audio_input:
|
| 143 |
+
return "Please provide audio input.", None
|
| 144 |
+
|
| 145 |
+
# معالجة الصوت
|
| 146 |
+
question = text_input if text_input else "What is the content of this audio?"
|
| 147 |
+
|
| 148 |
+
# النموذج يدعم الصوت مباشرة
|
| 149 |
+
messages = [
|
| 150 |
+
{
|
| 151 |
+
"role": "user",
|
| 152 |
+
"content": [
|
| 153 |
+
{"type": "audio", "audio": audio_input},
|
| 154 |
+
{"type": "text", "text": question}
|
| 155 |
+
]
|
| 156 |
+
}
|
| 157 |
+
]
|
| 158 |
+
|
| 159 |
+
elif mode == "Video + Text":
|
| 160 |
+
if not video_input:
|
| 161 |
+
return "Please provide a video.", None
|
| 162 |
+
|
| 163 |
+
question = text_input if text_input else "What happens in this video?"
|
| 164 |
+
|
| 165 |
+
# معالجة الفيديو
|
| 166 |
+
messages = [
|
| 167 |
+
{
|
| 168 |
+
"role": "user",
|
| 169 |
+
"content": [
|
| 170 |
+
{"type": "video", "video": video_input},
|
| 171 |
+
{"type": "text", "text": question}
|
| 172 |
+
]
|
| 173 |
+
}
|
| 174 |
+
]
|
| 175 |
+
|
| 176 |
+
# إعدادات التوليد
|
| 177 |
+
generation_config = {
|
| 178 |
+
"max_new_tokens": max_new_tokens,
|
| 179 |
+
"temperature": temperature,
|
| 180 |
+
"top_p": top_p,
|
| 181 |
+
"do_sample": temperature > 0,
|
| 182 |
+
}
|
| 183 |
+
|
| 184 |
+
try:
|
| 185 |
+
# التوليد
|
| 186 |
with torch.no_grad():
|
| 187 |
+
if mode == "Image + Text" and image_input:
|
| 188 |
+
# معالجة خاصة للصور
|
| 189 |
+
image = process_image(image_input)
|
| 190 |
+
question = text_input if text_input else "What is shown in this image?"
|
| 191 |
+
|
| 192 |
+
# استخدام chat للصور
|
| 193 |
+
response = model.chat(
|
| 194 |
+
image=image,
|
| 195 |
+
msgs=[{"role": "user", "content": question}],
|
| 196 |
+
tokenizer=tokenizer,
|
| 197 |
+
**generation_config
|
| 198 |
+
)
|
| 199 |
+
|
| 200 |
else:
|
| 201 |
+
# للنص والأنواع الأخرى
|
| 202 |
+
inputs = tokenizer(messages, return_tensors="pt")
|
| 203 |
+
inputs = inputs.to(model.device)
|
| 204 |
+
|
| 205 |
+
outputs = model.generate(
|
| 206 |
+
**inputs,
|
| 207 |
+
**generation_config
|
| 208 |
+
)
|
| 209 |
+
|
| 210 |
+
response = tokenizer.decode(
|
| 211 |
+
outputs[0][inputs['input_ids'].shape[1]:],
|
| 212 |
+
skip_special_tokens=True
|
| 213 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 214 |
|
| 215 |
+
# إذا كان TTS مفعل، نولد صوت
|
| 216 |
+
audio_output = None
|
| 217 |
+
if enable_tts and isinstance(response, str):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 218 |
try:
|
| 219 |
+
# استخدام TTS المدمج في النموذج
|
| 220 |
+
audio_output = model.generate_speech(
|
| 221 |
+
text=response,
|
| 222 |
+
style=tts_style
|
| 223 |
+
)
|
| 224 |
except Exception as e:
|
| 225 |
+
print(f"TTS generation failed: {e}")
|
| 226 |
+
audio_output = None
|
| 227 |
+
|
| 228 |
+
return response, audio_output
|
|
|
|
| 229 |
|
| 230 |
except Exception as e:
|
|
|
|
| 231 |
import traceback
|
| 232 |
traceback.print_exc()
|
| 233 |
+
return f"Error: {str(e)}", None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 234 |
|
| 235 |
|
| 236 |
# =========================================================
|
| 237 |
+
# واجهة Gradio
|
| 238 |
# =========================================================
|
| 239 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 240 |
def create_interface():
|
| 241 |
+
"""إنشاء واجهة Gradio لـ MiniCPM-o-2_6"""
|
| 242 |
+
|
| 243 |
+
with gr.Blocks(title="MiniCPM-o-2_6 - Multimodal AI") as demo:
|
| 244 |
gr.Markdown(
|
| 245 |
"""
|
| 246 |
+
# 🤖 MiniCPM-o-2_6 - Multimodal AI Assistant
|
| 247 |
+
|
| 248 |
+
**القدرات:**
|
| 249 |
+
- 🖼️ فهم الصور (OCR، وصف، تحليل)
|
| 250 |
+
- 🎙️ معالجة الصوت (ASR، فهم المحتوى)
|
| 251 |
+
- 🎬 تحليل الفيديو
|
| 252 |
+
- 🗣️ تحويل النص إلى كلام (TTS)
|
| 253 |
+
- 🎭 استنساخ الصوت
|
| 254 |
+
- 💬 محادثة في الوقت الفعلي
|
| 255 |
+
|
| 256 |
+
**الأداء:** يتفوق على GPT-4o و Claude 3.5 في العديد من المهام!
|
| 257 |
"""
|
| 258 |
)
|
| 259 |
+
|
| 260 |
with gr.Row():
|
|
|
|
| 261 |
with gr.Column(scale=3):
|
| 262 |
+
# اختيار نوع المدخل
|
| 263 |
+
mode = gr.Radio(
|
| 264 |
+
choices=["Text Only", "Image + Text", "Audio + Text", "Video + Text"],
|
| 265 |
+
value="Text Only",
|
| 266 |
+
label="Input Mode",
|
| 267 |
+
info="اختر نوع المدخل"
|
| 268 |
)
|
| 269 |
+
|
| 270 |
+
# المدخلات
|
| 271 |
+
text_input = gr.Textbox(
|
| 272 |
+
label="Text Input",
|
| 273 |
+
placeholder="اكتب سؤالك أو النص هنا...",
|
| 274 |
+
lines=3
|
| 275 |
+
)
|
| 276 |
+
|
| 277 |
+
image_input = gr.Image(
|
| 278 |
+
label="Image Input",
|
| 279 |
+
type="pil",
|
| 280 |
+
visible=False
|
| 281 |
+
)
|
| 282 |
+
|
| 283 |
+
audio_input = gr.Audio(
|
| 284 |
+
label="Audio Input",
|
| 285 |
+
type="filepath",
|
| 286 |
+
visible=False
|
| 287 |
+
)
|
| 288 |
+
|
| 289 |
+
video_input = gr.Video(
|
| 290 |
+
label="Video Input",
|
| 291 |
+
visible=False
|
| 292 |
+
)
|
| 293 |
+
|
| 294 |
+
# زر الإرسال
|
| 295 |
+
submit_btn = gr.Button("🚀 Process", variant="primary")
|
| 296 |
+
|
| 297 |
+
# المخرجات
|
| 298 |
+
output_text = gr.Textbox(
|
| 299 |
+
label="Response",
|
| 300 |
+
lines=5,
|
| 301 |
+
interactive=False
|
| 302 |
+
)
|
| 303 |
+
|
| 304 |
+
output_audio = gr.Audio(
|
| 305 |
+
label="Generated Speech (TTS)",
|
| 306 |
type="numpy",
|
| 307 |
+
visible=False
|
|
|
|
| 308 |
)
|
| 309 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 310 |
with gr.Column(scale=1):
|
| 311 |
+
gr.Markdown("### ⚙️ Settings")
|
| 312 |
+
|
| 313 |
+
temperature = gr.Slider(
|
| 314 |
+
label="Temperature",
|
| 315 |
+
minimum=0.0,
|
| 316 |
+
maximum=1.5,
|
| 317 |
+
value=0.7,
|
| 318 |
+
step=0.1
|
| 319 |
)
|
| 320 |
+
|
| 321 |
+
top_p = gr.Slider(
|
| 322 |
+
label="Top-p",
|
| 323 |
+
minimum=0.1,
|
| 324 |
+
maximum=1.0,
|
| 325 |
+
value=0.9,
|
| 326 |
+
step=0.05
|
| 327 |
)
|
| 328 |
+
|
| 329 |
+
max_new_tokens = gr.Slider(
|
| 330 |
+
label="Max Tokens",
|
| 331 |
+
minimum=50,
|
| 332 |
+
maximum=2048,
|
| 333 |
+
value=512,
|
| 334 |
+
step=50
|
| 335 |
)
|
| 336 |
+
|
| 337 |
+
gr.Markdown("### 🗣️ TTS Settings")
|
| 338 |
+
|
| 339 |
+
enable_tts = gr.Checkbox(
|
| 340 |
+
label="Enable TTS",
|
| 341 |
+
value=False,
|
| 342 |
+
info="تحويل الرد إلى كلام"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 343 |
)
|
| 344 |
+
|
| 345 |
+
tts_style = gr.Dropdown(
|
| 346 |
+
choices=["default", "emotional", "calm", "energetic"],
|
| 347 |
+
value="default",
|
| 348 |
+
label="TTS Style",
|
| 349 |
+
visible=False
|
| 350 |
+
)
|
| 351 |
+
|
| 352 |
+
# تحديث visibility حسب الوضع
|
| 353 |
+
def update_inputs(mode_value):
|
| 354 |
+
return {
|
| 355 |
+
image_input: gr.update(visible="Image" in mode_value),
|
| 356 |
+
audio_input: gr.update(visible="Audio" in mode_value),
|
| 357 |
+
video_input: gr.update(visible="Video" in mode_value),
|
| 358 |
+
}
|
| 359 |
+
|
| 360 |
+
mode.change(
|
| 361 |
+
fn=update_inputs,
|
| 362 |
+
inputs=[mode],
|
| 363 |
+
outputs=[image_input, audio_input, video_input]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 364 |
)
|
| 365 |
+
|
| 366 |
+
# تحديث visibility لإعدادات TTS
|
| 367 |
+
enable_tts.change(
|
| 368 |
+
fn=lambda x: {
|
| 369 |
+
tts_style: gr.update(visible=x),
|
| 370 |
+
output_audio: gr.update(visible=x)
|
| 371 |
+
},
|
| 372 |
+
inputs=[enable_tts],
|
| 373 |
+
outputs=[tts_style, output_audio]
|
|
|
|
|
|
|
| 374 |
)
|
| 375 |
+
|
| 376 |
+
# معالجة الإرسال
|
| 377 |
+
submit_btn.click(
|
| 378 |
+
fn=minicpm_o_inference,
|
| 379 |
+
inputs=[
|
| 380 |
+
text_input,
|
| 381 |
+
image_input,
|
| 382 |
+
audio_input,
|
| 383 |
+
video_input,
|
| 384 |
+
mode,
|
| 385 |
+
temperature,
|
| 386 |
+
top_p,
|
| 387 |
+
max_new_tokens,
|
| 388 |
+
enable_tts,
|
| 389 |
+
tts_style
|
| 390 |
+
],
|
| 391 |
+
outputs=[output_text, output_audio]
|
| 392 |
)
|
| 393 |
+
|
| 394 |
+
# أمثلة
|
| 395 |
+
gr.Examples(
|
| 396 |
+
examples=[
|
| 397 |
+
["What is artificial intelligence?", None, None, None, "Text Only"],
|
| 398 |
+
["Describe this image in detail", "examples/sample.jpg", None, None, "Image + Text"],
|
| 399 |
+
["Transcribe this audio", None, "examples/audio.wav", None, "Audio + Text"],
|
| 400 |
+
["What happens in this video?", None, None, "examples/video.mp4", "Video + Text"],
|
| 401 |
+
],
|
| 402 |
+
inputs=[text_input, image_input, audio_input, video_input, mode],
|
| 403 |
)
|
| 404 |
+
|
| 405 |
return demo
|
| 406 |
|
| 407 |
|
| 408 |
+
# =========================================================
|
| 409 |
+
# تشغيل التطبيق
|
| 410 |
+
# =========================================================
|
| 411 |
|
| 412 |
if __name__ == "__main__":
|
| 413 |
+
demo = create_interface()
|
| 414 |
demo.launch(
|
| 415 |
+
ssr_mode=False,
|
| 416 |
+
show_error=True,
|
| 417 |
+
share=False
|
| 418 |
)
|