[Kernel] Consolidate Marlin thread-tile padding across all dense Marlin paths (#45295)

Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin
2026-06-12 13:46:21 -07:00
committed by GitHub
parent 78739c1946
commit 9eaacb23ec
7 changed files with 770 additions and 66 deletions
@@ -468,6 +468,7 @@ class ModelOptFp8LinearMethod(LinearMethodBase):
layer.logical_widths = output_partition_sizes
layer.input_size_per_partition = input_size_per_partition
layer.output_size_per_partition = output_size_per_partition
layer.orig_dtype = params_dtype
weight_dtype = (
torch.float8_e4m3fn
if self.quant_config.is_checkpoint_fp8_serialized