[vLLM IR] rework gemma_rms_norm (#39014)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com> Signed-off-by: Jiangyun Zhu <riverclouds.zhu@qq.com> Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
This commit is contained in:
co-authored by
Luka Govedič
parent
da4c0e4db9
commit
8060bb0333
@@ -11,8 +11,11 @@ current_platform.import_kernels()
|
||||
CUDA_ALIKE = current_platform.is_cuda_alike()
|
||||
"""Most kernels in this file are supported on all CUDA-alike platforms."""
|
||||
|
||||
rms_no_var_size = lambda x, weight, epsilon, variance_size=None: variance_size is None
|
||||
"""vLLM kernel does not support variance_size parameter."""
|
||||
rms_no_var_size = (
|
||||
lambda x, weight, epsilon, variance_size=None: variance_size is None
|
||||
and (weight is None or weight.dtype == x.dtype)
|
||||
)
|
||||
"""vLLM kernel requires no variance_size override and matching input/weight dtype."""
|
||||
|
||||
|
||||
@ir.ops.rms_norm.register_impl(
|
||||
|
||||
Reference in New Issue
Block a user