Update app.py

app.py

@@ -4,24 +4,28 @@ import torch
 import gradio as gr
 
 # --- Конфиг ---
-MODEL_ID = "deepseek-ai/DeepSeek-V3-0324"
+MODEL_ID = "deepseek-ai/deepseek-llm-7b"
+
+# --- Загрузка модели (ленивая) ---
+model = None
+tokenizer = None
 
-# --- Загрузка модели ---
 @spaces.GPU
 def load_model():
-    model = AutoModelForCausalLM.from_pretrained(
-        MODEL_ID,
-        device_map="auto",
-        torch_dtype=torch.bfloat16,  # Оптимально для T4/A10G
-        trust_remote_code=True
-    )
-    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+    global model, tokenizer
+    if model is None:
+        model = AutoModelForCausalLM.from_pretrained(
+            MODEL_ID,
+            device_map="auto",
+            torch_dtype=torch.bfloat16,
+            trust_remote_code=True
+        )
+        tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
     return model, tokenizer
 
-model, tokenizer = load_model()
-
 # --- Генерация ---
 def generate(prompt: str) -> str:
+    model, tokenizer = load_model()  # Загружаем только при первом вызове
     inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
     outputs = model.generate(
         **inputs,
@@ -36,4 +40,4 @@ with gr.Blocks() as demo:
     gr.ChatInterface(generate)
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()