Update app.py

app.py

@@ -73,7 +73,9 @@ try:
     gen_tok.padding_side = "left"                # required for batched generation
     if gen_tok.pad_token is None:
         gen_tok.pad_token = gen_tok.eos_token
-    gen_model = AutoModelForCausalLM.from_pretrained(GEN_MODEL_ID, dtype=torch.float32)
+    # transformers 4.44 expects torch_dtype=, not dtype=. float32 is the CPU
+    # default anyway, so the argument is simply omitted.
+    gen_model = AutoModelForCausalLM.from_pretrained(GEN_MODEL_ID)
     gen_model.eval()
     print("Generation model ready.")
 except Exception as e: