update

app.py

@@ -22,11 +22,10 @@ MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "32768"))
 if torch.cuda.is_available():
     model_id = "vericava/llm-jp-3-1.8b-instruct-lora-vericava7-llama"
     base_model_id = "llm-jp/llm-jp-3-1.8b-instruct"
-    tokenizer = AutoTokenizer.from_pretrained(base_model_id)
-    model = AutoModelForCausalLM.from_pretrained(model_id)
     my_pipeline=pipeline(
         model=model_id,
         tokenizer=base_model_id,
+        use_safetensors=True,
     )
     my_pipeline.tokenizer.chat_template = "{{bos_token}}{% for message in messages %}{% if message['role'] == 'user' %}{{ '\\n\\n### 前の投稿:\\n' + message['content'] + '' }}{% elif message['role'] == 'system' %}{{ '以下は、SNS上の投稿です。あなたはSNSの投稿生成botとして、次に続く投稿を考えなさい。説明はせず、投稿の内容のみを鉤括弧をつけずに答えよ。' }}{% elif message['role'] == 'assistant' %}{{ '\\n\\n### 次の投稿:\\n' + message['content'] + eos_token }}{% endif %}{% if loop.last and add_generation_prompt %}{{ '\\n\\n### 次の投稿:\\n' }}{% endif %}{% endfor %}"