Fix missing IS_HIP_EXTENSION

Signed-off-by: wenjh <wenjh@sugon.com>

transformer_engine/pytorch/cpp_extensions/gemm.py

@@ -40,6 +40,7 @@ from transformer_engine.pytorch.triton.per_token_group_quant import (per_token_q
                                                                      tensorwise_dequantize_float,
                                                                      tensorwise_dequantize_float_add)
 from transformer_engine.pytorch.utils import get_device_compute_capability
+from torch.utils.cpp_extension import IS_HIP_EXTENSION
 
 from transformer_engine.pytorch.quantization import int8_simulation_fp8, int8_simulation_fp8_tensorwise
 int8_simulation_fp8_tensorwise_batched = bool(int(os.getenv("NVTE_INT8_SIM_FP8_TENSORWISE_BATCHED", "0")))