back to f32

app.py

@@ -42,7 +42,7 @@ def load_model():
         tokenizer = AutoTokenizer.from_pretrained(model_name)
         model = AutoModelForCausalLM.from_pretrained(
             model_name,
-            torch_dtype=torch.float16,
+            torch_dtype=torch.float32,
             device_map="auto",
             trust_remote_code=True
         )