Update app.py

app.py

@@ -9,12 +9,13 @@ from threading import Thread
 
 # Load model directly
 tokenizer = AutoTokenizer.from_pretrained("Navid-AI/Mulhem-1-Mini", token=os.getenv("HF_TOKEN"))
-model = AutoModelForCausalLM.from_pretrained("Navid-AI/Mulhem-1-Mini", torch_dtype=torch.bfloat16, attn_implementation="flash_attention_2", token=os.getenv("HF_TOKEN"))
+model = AutoModelForCausalLM.from_pretrained("Navid-AI/Mulhem-1-Mini", torch_dtype=torch.bfloat16, device_map="auto", attn_implementation="flash_attention_2", token=os.getenv("HF_TOKEN"))
 streamer = TextIteratorStreamer(tokenizer, skip_prompt=True)
 
 def respond(
     message,
     history: list[tuple[str, str]],
+    enable_reasoning,
     system_message,
     max_tokens,
     temperature,
@@ -29,7 +30,7 @@ def respond(
             messages.append({"role": "assistant", "content": val[1]})
 
     messages.append({"role": "user", "content": message})
-    inputs = tokenizer.apply_chat_template(messages, return_tensors="pt")
+    inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True, enable_reasoning=enable_reasoning)
     generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=max_tokens, temperature=temperature, top_p=top_p)
     thread = Thread(target=model.generate, kwargs=generation_kwargs)
 
@@ -44,8 +45,9 @@ For information on how to customize the ChatInterface, peruse the gradio docs: h
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=8192, value=512, step=1, label="Max new tokens"),
+        gr.Checkbox(label="Enable reasoning", value=False),
+        gr.Textbox(value="أنت مُلهم. ذكاء اصطناعي تم إنشاؤه من شركة نفيد لإلهام وتحفيز المستخدمين على التعلّم، النمو، وتحقيق أهدافهم.", label="System message"),
+        gr.Slider(minimum=1, maximum=8192, value=2048, step=1, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
         gr.Slider(
             minimum=0.1,
@@ -59,4 +61,4 @@ demo = gr.ChatInterface(
 
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()