Spaces:

psyche
/

llama3-korean-qa-chatbot

Sleeping

psyche commited on Jun 9

Commit

209d329

•

1 Parent(s): 4b9b79f

Update app.py

Files changed (1) hide show

app.py CHANGED Viewed

@@ -50,7 +50,7 @@ if not torch.cuda.is_available():
 if torch.cuda.is_available():
     model_id = "psyche/llama3-8b-instruct-ko"
-    model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_in_4bit=True, revision="v4.0")
     tokenizer = AutoTokenizer.from_pretrained(model_id)

 if torch.cuda.is_available():
     model_id = "psyche/llama3-8b-instruct-ko"
+    model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_in_4bit=True, revision="v4.1")
     tokenizer = AutoTokenizer.from_pretrained(model_id)