Remove .to(device) calls

app.py

@@ -5,13 +5,13 @@ import gradio as gr
 import spaces
 
 HF_TOKEN = os.environ.get("HF_TOKEN", None)
-if torch.cuda.is_available():
-    device = "cuda:0"
-else:
-    device = "cpu"
+# if torch.cuda.is_available():
+#     device = "cuda:0"
+# else:
+#     device = "cpu"
 
 tokenizer = AutoTokenizer.from_pretrained("PatronusAI/Llama-3-Patronus-Lynx-8B-Instruct")
-model = AutoModelForCausalLM.from_pretrained("PatronusAI/Llama-3-Patronus-Lynx-8B-Instruct", torch_dtype=torch.float16, device_map="auto").to(device)
+model = AutoModelForCausalLM.from_pretrained("PatronusAI/Llama-3-Patronus-Lynx-8B-Instruct", torch_dtype=torch.float16, device_map="auto")
 model.gradient_checkpointing_enable()
 
 # def load_model_and_tokenizer(model_choice):
@@ -79,10 +79,10 @@ HEADER = """
 @spaces.GPU()
 # def model_call(question, document, answer, tokenizer, model):
 def model_call(question, document, answer):
-    device = next(model.parameters()).device
+    # device = next(model.parameters()).device
     NEW_FORMAT = PROMPT.format(question=question, document=document, answer=answer)
     print("ENTIRE NEW_FORMAT", NEW_FORMAT)
-    inputs = tokenizer(NEW_FORMAT, return_tensors="pt").to(device)
+    inputs = tokenizer(NEW_FORMAT, return_tensors="pt")
     print("INPUTS", inputs)
     input_ids = inputs.input_ids
     attention_mask = inputs.attention_mask