33B inference too slowly

When I load the 33B model by the method shown in below, it's too slow to generate a token. And per token is about 2.9s

```python
tokenizer = AutoTokenizer.from_pretrained(checkpoint)

tokenizer.pad_token = tokenizer.eos_token

model = AutoModelForCausalLM.from_pretrained(checkpoint, trust_remote_code=True, torch_dtype=torch.float16, low_cpu_mem_usage=True, device_map="balanced")
```

have any solution about this?