7

app.py

@@ -25,7 +25,7 @@ def _load_model_tokenizer(checkpoint_path):
     torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
     
     model = AutoModelForCausalLM.from_pretrained(
-        checkpoint_path, torch_dtype=torch_dtype, device_map=device_map, resume_download=True
+        checkpoint_path, torch_dtype=torch_dtype, resume_download=True
     ).eval()
     
     model.generation_config.max_new_tokens = DEFAULT_MAX_NEW_TOKENS