Update processing.py

processing.py

@@ -34,7 +34,7 @@ class LazyPipeline:
             tokenizer = AutoTokenizer.from_pretrained(model_name, use_auth_token=hf_token)
             model = AutoModelForCausalLM.from_pretrained(
                 model_name,
-                torch_dtype=torch.float16,
+                torch_dtype=torch.bfloat16,
                 device_map="auto",
                 use_auth_token=hf_token
             )