removed cache_dir

app.py

@@ -11,10 +11,10 @@ temp_dir = tempfile.TemporaryDirectory()
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
 def get_tokenizer() -> PreTrainedTokenizer:
-    return AutoTokenizer.from_pretrained('thenlper/gte-large', trust_remote_code=True, cache_dir=temp_dir.name)
+    return AutoTokenizer.from_pretrained('thenlper/gte-large', trust_remote_code=True)
 
 def get_model() -> PreTrainedModel:
-    return AutoModel.from_pretrained('thenlper/gte-large', trust_remote_code=True, cache_dir=temp_dir.name).to(device)
+    return AutoModel.from_pretrained('thenlper/gte-large', trust_remote_code=True).to(device)
 
 def average_pooling(last_hidden_states: torch.Tensor, attention_mask: torch.Tensor) -> torch.Tensor:
     last_hidden = last_hidden_states.masked_fill(~attention_mask[..., None].bool(), 0.0)