[Sync] Update mxfp4.py (20250827) (#9724)
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com> Co-authored-by: Shiyang Chen <shiyang@x.ai>
This commit is contained in:
@@ -66,10 +66,15 @@ _is_hip = is_hip()
|
|||||||
|
|
||||||
if _is_hip:
|
if _is_hip:
|
||||||
# import aiter
|
# import aiter
|
||||||
from aiter import ActivationType, QuantType, dtypes
|
try:
|
||||||
from aiter.fused_moe import fused_moe
|
from aiter import ActivationType, QuantType, dtypes
|
||||||
from aiter.ops.triton.quant import dynamic_mxfp4_quant
|
from aiter.fused_moe import fused_moe
|
||||||
from aiter.utility.fp4_utils import e8m0_shuffle
|
from aiter.ops.triton.quant import dynamic_mxfp4_quant
|
||||||
|
from aiter.utility.fp4_utils import e8m0_shuffle
|
||||||
|
except ImportError as err:
|
||||||
|
ActivationType = QuantType = dtypes = fused_moe = dynamic_mxfp4_quant = (
|
||||||
|
e8m0_shuffle
|
||||||
|
) = err
|
||||||
|
|
||||||
|
|
||||||
def _swizzle_mxfp4(quant_tensor, scale, num_warps):
|
def _swizzle_mxfp4(quant_tensor, scale, num_warps):
|
||||||
|
|||||||
Reference in New Issue
Block a user