.

app.py

@@ -75,8 +75,8 @@ class BottleneckT5Autoencoder:
     def __init__(self, model_path: str, device='cuda'):
         self.device = device
         self.tokenizer = AutoTokenizer.from_pretrained(model_path, model_max_length=512, torch_dtype=torch.bfloat16)
-        self.model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True, low_cpu_mem_usage=True)
-        self.model.eval().to('cuda')
+        self.model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True, low_cpu_mem_usage=True).to('cuda')
+        self.model.eval()
 
 
     def embed(self, text: str) -> torch.FloatTensor: