Upload app.py with huggingface_hub

app.py

@@ -50,6 +50,7 @@ def chat_stream(message, history, system_message, max_tokens, temperature, top_p
         temperature=temperature,
         top_p=top_p,
         pad_token_id=tokenizer.eos_token_id,
+        eos_token_id=[tokenizer.eos_token_id, tokenizer.convert_tokens_to_ids("<|im_end|>")],
     )
 
     t = Thread(target=model.generate, kwargs=generate_kwargs)
@@ -58,17 +59,21 @@ def chat_stream(message, history, system_message, max_tokens, temperature, top_p
     partial_text = ""
     for new_text in streamer:
         partial_text += new_text
+        if "<|im_end|>" in partial_text:
+            partial_text = partial_text.split("<|im_end|>")[0]
+            yield parse_thought(partial_text)
+            break
         yield parse_thought(partial_text)
 
 with gr.Blocks(title="Lumin Nano 2.1") as demo:
-    gr.Markdown("Lumin Nano 2.1")
+    gr.Markdown("Lumin Nano 2.1 - Spanish Only")
     
     gr.ChatInterface(
         chat_stream,
         additional_inputs=[
-            gr.Textbox(value="Eres Lumin Nano 2.1. Eres un modelo de lenguaje avanzado. Piensa y responde SIEMPRE en español de forma concisa y directa. No uses emojis.", label="System Message"),
-            gr.Slider(1, 1024, 128, label="Max Tokens"),
-            gr.Slider(0.1, 1.0, 0.2, label="Temperature"),
+            gr.Textbox(value="Eres Lumin Nano 2.1. UNICAMENTE puedes pensar y responder en ESPAÑOL. Tienes PROHIBIDO usar el inglés. Sé directo, conciso y nunca uses emojis.", label="System Message"),
+            gr.Slider(1, 1024, 256, label="Max Tokens"),
+            gr.Slider(0.01, 1.0, 0.1, label="Temperature"),
             gr.Slider(0.1, 1.0, 0.9, label="Top-p"),
         ],
     )