Update app.py

app.py

@@ -11,9 +11,7 @@ from threading import Thread
 DESCRIPTION = '''
 <div>
 <h1 style="text-align: center;">Meta Llama3 8B</h1>
-<p>This Space demonstrates the instruction-tuned model <a href="https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct"><b>Meta Llama3 8b Chat</b></a>. Meta Llama3 is the new open LLM and comes in two sizes: 8b and 70b. Feel free to play with it, or duplicate to run privately!</p>
-<p>🔎 For more details about the Llama3 release and how to use the model with <code>transformers</code>, take a look <a href="https://huggingface.co/blog/llama3">at our blog post</a>.</p>
-<p>🦕 Looking for an even more powerful model? Check out the <a href="https://huggingface.co/chat/"><b>Hugging Chat</b></a> integration for Meta Llama 3 70b</p>
+<p>日本語向け Llama 3 のデモだよ。 <a href="https://huggingface.co/alfredplpl/Llama-3-8B-Instruct-Ja"><b>日本語向け Llama3 8b Chat</b></a>.</p>
 </div>
 '''
 
@@ -71,6 +69,7 @@ def chat_llama3_8b(message: str,
         str: The generated response.
     """
     conversation = []
+     conversation.append({"role": "system", "content": "あなたは日本語で回答するAIアシスタントです。"})
     for user, assistant in history:
         conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
     conversation.append({"role": "user", "content": message})
@@ -85,6 +84,8 @@ def chat_llama3_8b(message: str,
         max_new_tokens=max_new_tokens,
         do_sample=True,
         temperature=temperature,
+        top_p=0.95,
+        repetition_penalty=1.1,
         eos_token_id=terminators,
     )
     # This will enforce greedy generation (do_sample=False) when the temperature is passed 0, avoiding the crash.