Set pad_token to eos_token and exclude user query from response

app.py

@@ -6,6 +6,7 @@ from transformers import AutoTokenizer, AutoModelForCausalLM
 import threading
 import torch
 import os
+import re
 
 # Define the API URL to use the internal server
 API_URL = "http://localhost:5000/chat"
@@ -43,7 +44,7 @@ def chat():
     inputs = tokenizer(user_input, return_tensors='pt', padding=True, truncation=True)
     input_ids = inputs['input_ids']
     attention_mask = inputs['attention_mask']
-    outputs = model.generate(input_ids, attention_mask=attention_mask, max_new_tokens=100,pad_token_id=tokenizer.eos_token_id)
+    outputs = model.generate(input_ids, attention_mask=attention_mask, max_new_tokens=1000, pad_token_id=tokenizer.eos_token_id)
     response_text = tokenizer.decode(outputs[0], skip_special_tokens=True).replace(user_input, '').strip()
     
     return jsonify({"response": response_text})
@@ -72,6 +73,9 @@ def messages_to_history(messages: Messages) -> History:
         history.append((q['content'], r['content']))
     return history
 
+def is_hebrew(text: str) -> bool:
+    return bool(re.search(r'[\u0590-\u05FF]', text))
+
 def model_chat(query: str, history: History) -> Tuple[str, History]:
     if not query.strip():
         return '', history
@@ -163,7 +167,10 @@ with gr.Blocks(css='''
         print(f"Query: {query}")  # Debug print statement
         response, history = model_chat(query, history)
         print(f"Response: {response}")  # Debug print statement
-        return history, gr.update(value="", interactive=True), history  # Ensure correct return format
+        if is_hebrew(response):
+            return history, gr.update(value="", interactive=True, lines=2, rtl=True), history
+        else:
+            return history, gr.update(value="", interactive=True, lines=2, rtl=False), history
 
     demo_state = gr.State([])