[Bugfix] Use latency MOE backend as default for Flashinfer and other misc fixes (#27439)
Signed-off-by: Pavani Majety <pmajety@nvidia.com>
This commit is contained in:
+1
-1
@@ -1385,7 +1385,7 @@ def scaled_fp4_quant(
|
||||
rounded_m = round_up(m, 128)
|
||||
scale_n = n // block_size
|
||||
rounded_n = round_up(scale_n, 4)
|
||||
output_scale = torch.zeros(
|
||||
output_scale = torch.empty(
|
||||
(rounded_m, rounded_n // 4), device=device, dtype=torch.int32
|
||||
)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user