Update app.py

app.py

@@ -4,7 +4,6 @@ from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor, Bits
 from qwen_vl_utils import process_vision_info
 from threading import Thread
 import re
-import random
 import spaces
 
 # Константы
@@ -20,26 +19,38 @@ PROMPT = (
 QUESTION_TEMPLATE_THINKING = "{Question} First output the thinking process in <think> </think> tags and then output the final answer with only one score in <answer> </answer> tags."
 QUESTION_TEMPLATE_NO_THINKING = "{Question} Please only output the final answer with only one score in <answer> </answer> tags."
 
-# Конфигурация 8-bit квантизации
-quantization_config = BitsAndBytesConfig(
-    load_in_8bit=True,
-    llm_int8_threshold=6.0,
-    llm_int8_has_fp16_weight=False,
-)
+# Глобальные переменные для модели
+model = None
+processor = None
 
-print("Loading model...")
-model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
-    MODEL_PATH,
-    quantization_config=quantization_config,
-    device_map="auto",
-    trust_remote_code=True,
-    torch_dtype=torch.float16,
-)
-model.eval()
 
-processor = AutoProcessor.from_pretrained(MODEL_PATH, trust_remote_code=True)
-processor.tokenizer.padding_side = "left"
-print("Model loaded successfully!")
+def load_model():
+    """Загрузка модели с 8-bit квантизацией"""
+    global model, processor
+    
+    if model is not None:
+        return
+    
+    print("Loading model...")
+    
+    quantization_config = BitsAndBytesConfig(
+        load_in_8bit=True,
+        llm_int8_threshold=6.0,
+    )
+    
+    model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+        MODEL_PATH,
+        quantization_config=quantization_config,
+        device_map="auto",
+        trust_remote_code=True,
+        torch_dtype=torch.float16,
+    )
+    model.eval()
+    
+    processor = AutoProcessor.from_pretrained(MODEL_PATH, trust_remote_code=True)
+    processor.tokenizer.padding_side = "left"
+    
+    print("Model loaded successfully!")
 
 
 def extract_score(text):
@@ -53,7 +64,7 @@ def extract_score(text):
         score_match = re.search(r'\d+(\.\d+)?', model_answer)
         if score_match:
             score = float(score_match.group())
-            return min(max(score, 1.0), 5.0)  # Ограничение от 1 до 5
+            return min(max(score, 1.0), 5.0)
     except Exception as e:
         print(f"Error extracting score: {e}")
     return None
@@ -67,9 +78,14 @@ def extract_thinking(text):
     return None
 
 
-@spaces.GPU(duration=120)
+@spaces.GPU(duration=180)
 def score_image_streaming(image, use_thinking=True):
     """Оценка качества изображения со стримингом"""
+    global model, processor
+    
+    # Загрузка модели при первом вызове
+    load_model()
+    
     if image is None:
         yield "❌ Please upload an image first.", "", ""
         return
@@ -134,7 +150,7 @@ def score_image_streaming(image, use_thinking=True):
     # Стриминг вывода
     generated_text = ""
     current_thinking = ""
-    current_score = ""
+    current_score = "*Analyzing...*"
     
     for new_text in streamer:
         generated_text += new_text
@@ -149,10 +165,7 @@ def score_image_streaming(image, use_thinking=True):
         if score is not None:
             current_score = f"⭐ **Quality Score: {score:.2f} / 5.00**"
         
-        # Форматирование вывода
-        display_text = generated_text
-        
-        yield display_text, current_thinking, current_score
+        yield generated_text, current_thinking, current_score
     
     thread.join()
     
@@ -174,23 +187,6 @@ def create_interface():
     with gr.Blocks(
         title="VisualQuality-R1: Image Quality Assessment",
         theme=gr.themes.Soft(),
-        css="""
-        .score-box {
-            background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
-            border-radius: 10px;
-            padding: 20px;
-            color: white;
-            text-align: center;
-            font-size: 1.2em;
-        }
-        .thinking-box {
-            background-color: #f0f4f8;
-            border-left: 4px solid #667eea;
-            padding: 15px;
-            border-radius: 5px;
-            font-style: italic;
-        }
-        """
     ) as demo:
         
         gr.Markdown("""
@@ -260,16 +256,6 @@ def create_interface():
                     interactive=False
                 )
         
-        # Примеры
-        gr.Markdown("### 📸 Example Images")
-        gr.Examples(
-            examples=[
-                ["https://upload.wikimedia.org/wikipedia/commons/thumb/4/47/PNG_transparency_demonstration_1.png/300px-PNG_transparency_demonstration_1.png"],
-            ],
-            inputs=[image_input],
-            label="Click to try"
-        )
-        
         # Обработка события
         submit_btn.click(
             fn=score_image_streaming,