Update app.py

app.py

@@ -34,14 +34,18 @@ def generate_ai(message: str):
     with torch.no_grad():
        outputs = model.generate(
     **inputs,
-    max_new_tokens=25,
+    max_new_tokens=60,
+    min_new_tokens=20,   # 🔥 важно
     do_sample=True,
     temperature=0.7,
     top_p=0.9,
     eos_token_id=tokenizer.eos_token_id
 )
 
-    reply = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    input_length = inputs.input_ids.shape[1]
+generated_tokens = outputs[0][input_length:]
+
+reply = tokenizer.decode(generated_tokens, skip_special_tokens=True).strip()
 
     if "Assistant:" in reply:
         reply = reply.split("Assistant:")[-1].strip()