biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[XPU][1/N] Deprecate ipex and switch to vllm-xpu-kernels for xpu platform (#33379)

Browse Source 
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
This commit is contained in:
Kunshang Ji
2026-02-03 14:46:10 +08:00
committed by GitHub
parent bf001da4bf
commit e10604480b
18 changed files with 150 additions and 927 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
vllm/model_executor/layers/quantization/inc.py
View File

@@ -38,7 +38,6 @@ class INCConfig(QuantizationConfig):
"awq",
"awq:marlin",
"marlin",
"ipex",
}
def __init__(
@@ -410,31 +409,10 @@ class INCConfig(QuantizationConfig):
return UnquantizedLinearMethod()
else:
return None
from vllm.model_executor.layers.quantization.ipex_quant import (
IPEXAWQLinearMethod,
IPEXConfig,
IPEXGPTQLinearMethod,
raise NotImplementedError(
"INC quantization is not supported during xpu kernel migration."
)
if isinstance(layer, (LinearBase, ParallelLMHead)):
if "awq" in self.packing_format:
config = IPEXConfig(
method="awq", weight_bits=weight_bits, group_size=group_size
)
return IPEXAWQLinearMethod(config)
elif "gptq" in self.packing_format:
config = IPEXConfig(
method="gptq", weight_bits=weight_bits, group_size=group_size
)
return IPEXGPTQLinearMethod(config)
else:
raise ValueError(
f"ipex backend only supports awq "
f"and gptq format,but got {self.packing_format}"
)
else:
return None
def get_quant_method(self, layer: torch.nn.Module, prefix: str):
if prefix and self.extra_config:
for layer_name in self.extra_config: