fix

app.py

@@ -5,15 +5,22 @@ from transformers import AutoModelForCausalLM, LlamaTokenizer
 @st.cache_resource
 def load():
     """
-    model = AutoModelForCausalLM.from_pretrained(
-    "stabilityai/japanese-stablelm-instruct-alpha-7b",
-    trust_remote_code=True,
-    )
+    base_model = AutoModelForCausalLM.from_pretrained(
+        "stabilityai/japanese-stablelm-instruct-alpha-7b", 
+        device_map="auto",
+        low_cpu_mem_usage=True,
+        variant="int8",
+        load_in_8bit=True,
+        trust_remote_code=True,
+        )
+    model = PeftModel.from_pretrained(
+        base_model, 
+        "lora_adapter",
+        device_map="auto",
+        )
     """
-    model = None
     tokenizer = LlamaTokenizer.from_pretrained(
-        "novelai/nerdstash-tokenizer-v1", 
-        additional_special_tokens=['▁▁'],
+        "lora_adapter", 
         )
     return model, tokenizer
 
@@ -23,7 +30,7 @@ def get_prompt(user_query, system_prompt, messages="", sep="\n\n### "):
     msgs = [": \n" + user_query, ": "]
     if messages:
         roles.insert(1, "入力")
-        msgs.insert(1, ": \n" + "\n".join(message for _, message in messages))
+        msgs.insert(1, ": \n" + "\n".join(message["content"] for message in messages))
 
     for role, msg in zip(roles, msgs):
         prompt += sep + role + msg