Update: コンテキスト長を拡張

app.py

@@ -10,22 +10,23 @@ model_name_or_path = "mmnga/ELYZA-japanese-Llama-2-7b-fast-instruct-gguf"
 model_basename = "ELYZA-japanese-Llama-2-7b-fast-instruct-q5_K_M.gguf"
 
 model_path = hf_hub_download(repo_id=model_name_or_path, filename=model_basename, revision="main")
-llama = Llama(model_path)
+llama = Llama(model_path, n_ctx=5120)
 
 def predict(messages):
     # Llamaでの回答を取得（ストリーミングオン）
-    streamer = llama.create_chat_completion(messages, stream=True)
+    streamer = llama.create_chat_completion(messages, stream=True, max_tokens=512)
 
     partial_message = ""
     for msg in streamer:
         message = msg['choices'][0]['delta']
+        print(f"message: {message}")
         if 'content' in message:
             partial_message += message['content']
             yield partial_message
 
 
 def main():
-    st.title("Chat with ChatGPT Clone!")
+    st.title("Chat with Elyza!")
 
     # Session state for retaining messages
     if 'messages' not in st.session_state: