Update app.py

app.py

@@ -1,9 +1,9 @@
 import gradio as gr
 from transformers import pipeline, AutoTokenizer
-from typing import List, Dict, Any, Tuple
+from typing import List, Tuple, Dict, Any
 import torch
 
-# CPU-модели (маленькие, chat-ready)
+# CPU-модели
 MODELS = {
     "Qwen2.5-0.5B": "Qwen/Qwen2.5-0.5B-Instruct",
     "Qwen2.5-1.5B": "Qwen/Qwen2.5-1.5B-Instruct",
@@ -11,7 +11,6 @@ MODELS = {
 }
 
 def load_model(model_key: str):
-    """Lazy load pipeline."""
     model_id = MODELS[model_key]
     print(f"🚀 Загрузка {model_id}...")
     tokenizer = AutoTokenizer.from_pretrained(model_id)
@@ -32,14 +31,12 @@ def load_model(model_key: str):
     print(f"✅ {model_id} загружена!")
     return pipe
 
-# Global cache
 model_cache = {}
 
 def respond(message: str, 
-            history: List[Dict[str, str]], 
+            history: List[Tuple[str, str]], 
             model_key: str, 
-            system_prompt: str) -> Tuple[List[Dict[str, str]], str, Dict[str, Any]]:
-    """Локальный чат с pipeline."""
+            system_prompt: str) -> Tuple[List[Tuple[str, str]], str, Dict[str, Any]]:
     try:
         if model_key not in model_cache:
             model_cache[model_key] = load_model(model_key)
@@ -47,39 +44,33 @@ def respond(message: str,
         
         print(f"🚀 Генерация: {model_key}, Msg='{message[:30]}...'")
         
-        # Chat format (system + history + user)
         messages = []
         if system_prompt.strip():
             messages.append({"role": "system", "content": system_prompt})
-        messages.extend(history)
+        
+        for user_msg, bot_reply in history:
+            messages.append({"role": "user", "content": user_msg})
+            messages.append({"role": "assistant", "content": bot_reply})
+        
         messages.append({"role": "user", "content": message})
         
-        # Apply chat template (для instruct)
         tokenizer = pipe.tokenizer
         prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
         
-        # Generate
         outputs = pipe(prompt, max_new_tokens=512, do_sample=True, temperature=0.7)
         bot_reply = outputs[0]["generated_text"][len(prompt):].strip()
         
         print(f"✅ Ответ: {bot_reply[:50]}...")
         
-        new_history = history + [
-            {"role": "user", "content": message},
-            {"role": "assistant", "content": bot_reply}
-        ]
+        new_history = history + [(message, bot_reply)]
         return new_history, "", gr.update(value="")
     
     except Exception as e:
         error_msg = f"❌ {model_key}: {str(e)}"
         print(f"💥 {error_msg}")
-        new_history = history + [
-            {"role": "user", "content": message},
-            {"role": "assistant", "content": error_msg}
-        ]
+        new_history = history + [(message, error_msg)]
         return new_history, error_msg, gr.update(value="")
 
-# UI — ИСПРАВЛЕНО: убраны theme и type, несовместимые с Gradio 5
 with gr.Blocks(title="🚀 Локальный HF Чат (на слабом CPU!)") as demo:
     gr.Markdown("# Локальный Inference (без API!)\n**Маленькие модели** — 1-3 сек CPU. Большие думают ооочень долго. Нет limits/token. В качестве примера.")
     
@@ -87,8 +78,7 @@ with gr.Blocks(title="🚀 Локальный HF Чат (на слабом CPU!)
         model_dropdown = gr.Dropdown(choices=list(MODELS.keys()), value="Qwen2.5-0.5B", label="🧠 Модель")
         system_prompt = gr.Textbox(label="📝 System", placeholder="Ты весёлый ИИ.", lines=2)
     
-    # В Gradio 5+ gr.Chatbot всегда использует messages-формат, параметр type удалён
-    chatbot = gr.Chatbot(height=500, label="Чат")
+    chatbot = gr.Chatbot(height=500, label="Чат")  # ← без type
     
     with gr.Row():
         msg_input = gr.Textbox(placeholder="Привет! (Enter)", show_label=False, lines=1)
@@ -100,7 +90,6 @@ with gr.Blocks(title="🚀 Локальный HF Чат (на слабом CPU!)
     
     status = gr.Textbox(label="Логи", interactive=False, lines=4)
     
-    # Events
     send_btn.click(fn=respond, inputs=[msg_input, chatbot, model_dropdown, system_prompt], outputs=[chatbot, status, msg_input])
     msg_input.submit(fn=respond, inputs=[msg_input, chatbot, model_dropdown, system_prompt], outputs=[chatbot, status, msg_input])
     
@@ -108,11 +97,11 @@ with gr.Blocks(title="🚀 Локальный HF Чат (на слабом CPU!)
         return [], "", gr.update(value="")
     clear_btn.click(clear, outputs=[chatbot, status, msg_input])
     
-    def retry(history: List[Dict[str, str]]):
-        if len(history) >= 2 and history[-2]["role"] == "user":
-            return history[-2]["content"]
+    def retry(history: List[Tuple[str, str]]):
+        if history:
+            return history[-1][0]
         return ""
     retry_btn.click(retry, inputs=[chatbot], outputs=[msg_input])
 
 if __name__ == "__main__":
-    demo.queue(max_size=10).launch(debug=True)
+    demo.queue(max_size=10).launch(debug=True, ssr_mode=False)