Update app.py

app.py

@@ -89,7 +89,7 @@ vectordb = FAISS.load_local(CFG.Output_folder + '/faiss_index_ml_papers', embedd
 @spaces.GPU
 def build_model(model_repo = CFG.model_name):
     tokenizer = AutoTokenizer.from_pretrained(model_repo)
-    model = AutoModelForCausalLM.from_pretrained(model_repo, attn_implementation="flash_attention_2")
+    model = AutoModelForCausalLM.from_pretrained(model_repo, attn_implementation="flash_attention_2", torch_dtype=torch.bfloat16)
 
     return tokenizer, model