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

[Hardware][intel GPU] bump up ipex version to 2.3 (#8365)

Browse Source 
Co-authored-by: Yan Ma <yan.ma@intel.com>
This commit is contained in:
Kunshang Ji
2024-09-14 07:54:34 +08:00
committed by GitHub
parent 9ba0817ff1
commit 851725202a
6 changed files with 60 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
vllm/model_executor/layers/activation.py
View File

@@ -114,9 +114,7 @@ class NewGELU(CustomOp):
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
from vllm._ipex_ops import ipex_ops as ops
out = torch.empty_like(x)
ops.gelu_new(out, x)
return out
return ops.gelu_new(x)
class FastGELU(CustomOp):
@@ -136,9 +134,7 @@ class FastGELU(CustomOp):
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
from vllm._ipex_ops import ipex_ops as ops
out = torch.empty_like(x)
ops.gelu_fast(out, x)
return out
return ops.gelu_fast(x)
class QuickGELU(CustomOp):
@@ -155,6 +151,13 @@ class QuickGELU(CustomOp):
ops.gelu_quick(out, x)
return out
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
from vllm._ipex_ops import ipex_ops as ops
out = torch.empty_like(x)
ops.gelu_quick(out, x)
return out
# TODO implement forward_xpu for QuickGELU
# def forward_xpu(self, x: torch.Tensor) -> torch.Tensor: