[Misc] Removed force_fp8_e4m3fnuz from FP8LinearOp (#23725)

Signed-off-by: Julien Lin <jullin@nvidia.com>
Signed-off-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
This commit is contained in:
nvjullin
2025-09-04 09:25:40 -04:00
committed by GitHub
co-authored by Luka Govedič
parent c9f7081f9c
commit 37241077d5
5 changed files with 45 additions and 30 deletions
@@ -355,12 +355,10 @@ class Fp8LinearOp:
def __init__(self,
act_quant_static: bool,
act_quant_group_shape: GroupShape = GroupShape.PER_TENSOR,
pad_output: Optional[bool] = None,
force_fp8_e4m3fnuz: bool = False):
pad_output: Optional[bool] = None):
if current_platform.is_rocm():
self.preferred_backend = "rocm"
elif current_platform.is_cuda(
) and not force_fp8_e4m3fnuz and cutlass_fp8_supported():
elif current_platform.is_cuda() and cutlass_fp8_supported():
if has_flashinfer() and current_platform.has_device_capability(
100):
self.preferred_backend = "flashinfer"