fix bug when loading 4bit checkpoint quantized in INC

```
SRAM_SLICER_SHARED_MME_INPUT_EXPANSION_ENABLED=false ENABLE_EXPERIMENTAL_FLAGS=1 \
python run_generation.py \
-o acc_load_uint4_model.txt \
--model_name_or_path meta-llama/Llama-2-7b-hf \
--use_hpu_graphs \
--use_kv_cache \
--trim_logits \
--batch_size 1 \
--bf16 \
--attn_softmax_bf16 \
--quantized_inc_model_path <local_model_path_from_inc> 
```
upper command will got error due to dtype mismatch.
```
  File "/home/xinhe3/optimum-habana-fork/examples/text-generation/run_generation.py", line 656, in generate                                              
    output_tokens = model.generate(                                                                                                                      
  File "/home/xinhe3/qnpu/venv-118/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context                               
    return func(*args, **kwargs)                                                                                                                         
  File "/home/xinhe3/optimum-habana-fork/optimum/habana/transformers/generation/utils.py", line 1292, in generate                                        
    result = self._sample(                                                                                                                               
  File "/home/xinhe3/optimum-habana-fork/optimum/habana/transformers/generation/utils.py", line 2259, in _sample                                         
    outputs = self(                                                                                                                                      
  File "/home/xinhe3/qnpu/venv-118/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1556, in _wrapped_call_impl                            
    return self._call_impl(*args, **kwargs)                                                                                                              
  File "/home/xinhe3/qnpu/venv-118/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1565, in _call_impl                                    
    return forward_call(*args, **kwargs)                                                                                                                 
  File "/home/xinhe3/qnpu/venv-118/lib/python3.10/site-packages/habana_frameworks/torch/hpu/graphs.py", line 724, in forward                             
    return wrapped_hpugraph_forward(                                                                                                                     
  File "/home/xinhe3/qnpu/venv-118/lib/python3.10/site-packages/habana_frameworks/torch/hpu/graphs.py", line 597, in wrapped_hpugraph_forward            
    outputs = orig_fwd(*args, **kwargs)                                                                                                                  
  File "/home/xinhe3/optimum-habana-fork/optimum/habana/transformers/models/llama/modeling_llama.py", line 1345, in forward                              
    outputs = self.model(
  File "/home/xinhe3/qnpu/venv-118/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1556, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/xinhe3/qnpu/venv-118/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1606, in _call_impl
    result = forward_call(*args, **kwargs)
  File "/home/xinhe3/optimum-habana-fork/optimum/habana/transformers/models/llama/modeling_llama.py", line 1236, in forward
    layer_outputs = decoder_layer(
  File "/home/xinhe3/qnpu/venv-118/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1556, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/xinhe3/qnpu/venv-118/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1606, in _call_impl
    result = forward_call(*args, **kwargs)
  File "/home/xinhe3/optimum-habana-fork/optimum/habana/transformers/models/llama/modeling_llama.py", line 925, in forward
    hidden_states, self_attn_weights, present_key_value = self.pre_attn(
  File "/home/xinhe3/optimum-habana-fork/optimum/habana/transformers/models/llama/modeling_llama.py", line 982, in pre_attn
    hidden_states, attn_weights, present_key_value = self.self_attn.pre_attn_forward(
  File "/home/xinhe3/optimum-habana-fork/optimum/habana/transformers/models/llama/modeling_llama.py", line 647, in pre_attn_forward
    key_states = self.k_cache.update(past_key_value[0], key_states, 2, token_idx, self.inp_seq_len)
  File "/home/xinhe3/optimum-habana-fork/optimum/habana/transformers/models/llama/modeling_llama.py", line 424, in update
    prev.index_copy_(dim, idx - 1, cur)
RuntimeError: cpu fallback is not supported during hpu graph capturing
```

Author: xin3he

examples/text-generation/utils.py

@@ -269,6 +269,9 @@ def setup_model(args, model_dtype, model_kwargs, logger):
             original_model=org_model,
             **model_kwargs,
         )
+        # TODO: This will be removed in v1.19 Synapse release
+        # the loaded model should have the same dtype as original_model
+        model = model.to(model_kwargs["torch_dtype"])
     else:
         if args.assistant_model is not None:
             assistant_model = AutoModelForCausalLM.from_pretrained(