small fix and remove ununsed code about ipex (#12671)

python/llm/src/ipex_llm/transformers/convert.py

@@ -847,18 +847,9 @@ def replace_with_low_bit_linear_for_module(model, qtype, module_name=None,
                         mp_group=mp_group,
                     )
                     device = module.weight.data.device
-                    from ipex_llm.transformers.utils import get_ipex_version
-                    if get_ipex_version() < "2.1.10+xpu":
-                        new_linear._parameters['weight'] = nn.Parameter(module.weight)
-                    else:
-                        # only from 2.1, ipex provides matmul_bias_out
-                        # so we need to transpose weight
-                        new_weight = module.weight.transpose(0, 1).contiguous()
-                        new_linear._parameters['weight'] = nn.Parameter(new_weight)
-                        new_linear.weight_type = 2
+                    new_linear._parameters['weight'] = nn.Parameter(module.weight)
                     if module.bias is not None:
-                        new_linear._parameters['bias'] = nn.Parameter(module.bias.data)\
-                            .to(device)
+                        new_linear._parameters['bias'] = nn.Parameter(module.bias.data).to(device)
                 elif qtype == ggml_tensor_qtype["bf16"]:
                     module.to(torch.bfloat16)
                     new_linear = BF16Linear(

python/llm/src/ipex_llm/transformers/low_bit_linear.py

@@ -51,8 +51,7 @@ from torch import Tensor, device, dtype, nn
 from operator import mul
 from functools import reduce
 from ipex_llm.transformers.xpu_customize_fwd import custom_fwd, custom_bwd
-from ipex_llm.transformers.utils import get_autocast_dtype, get_xpu_device_name, \
-    get_ipex_version
+from ipex_llm.transformers.utils import get_autocast_dtype, get_xpu_device_name
 from ipex_llm.transformers.convert import is_deepspeed_available, get_use_vllm
 
 T = TypeVar("T", bound="torch.nn.Module")

python/llm/src/ipex_llm/transformers/models/utils.py

@@ -19,7 +19,7 @@ import torch
 import warnings
 from ipex_llm.utils.common import invalidInputError
 from ipex_llm.ggml.quantize import ggml_tensor_qtype
-from ipex_llm.transformers.utils import get_ipex_version, get_xpu_device_name
+from ipex_llm.transformers.utils import get_xpu_device_name
 from ipex_llm.transformers.low_bit_linear import SYM_INT4, SYM_INT8, FP8E5, IQ2_XXS, FP4, FP8E4,\
     FP6, ASYM_INT4
 

python/llm/src/ipex_llm/transformers/utils.py

@@ -154,20 +154,6 @@ def get_autocast_dtype(x):
                           f"Device {x.device} is not supported.")
 
 
-_ipex_version = None
-
-
-def get_ipex_version():
-
-    global _ipex_version
-    if _ipex_version is not None:
-        return _ipex_version
-
-    import intel_extension_for_pytorch as ipex
-    _ipex_version = ipex.__version__
-    return _ipex_version
-
-
 def get_xpu_device_name(device: torch.device):
     if device.type != "xpu":
         return device.type