A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Bugfix] Fix packed_factor missing attribute error (#23902)
Signed-off-by: Kyuyeun Kim <kyuyeunk@google.com>
Kyuyeun Kim committed
Sep 2, 2025 at 10:56 UTC
9480ae24e38cb73c5b665f5843ebd92c75a2039f
1 file changed
+8
-8
vllm/model_executor/layers/linear.py
+8
-8
@@ -722,8 +722,8 @@ class MergedColumnParallelLinear(ColumnParallelLinear):
722
# If quantized, we need to adjust the offset and size to account
723
# for the packing.
724
if packed_dim == output_dim:
725
- shard_size = shard_size // param.pack_factor
726
- shard_offset = shard_offset // param.pack_factor
725
+ shard_size = shard_size // param.packed_factor
726
+ shard_offset = shard_offset // param.packed_factor
727
# Special case for Marlin.
728
shard_size, shard_offset = adjust_marlin_shard(
729
param, shard_size, shard_offset)
@@ -756,8 +756,8 @@ class MergedColumnParallelLinear(ColumnParallelLinear):
756
# for the packing.
757
packed_dim = getattr(param, "packed_dim", None)
758
if packed_dim == output_dim:
759
- shard_size = shard_size // param.pack_factor
760
- shard_offset = shard_offset // param.pack_factor
759
+ shard_size = shard_size // param.packed_factor
760
+ shard_offset = shard_offset // param.packed_factor
761
# Special case for Marlin.
762
shard_size, shard_offset = adjust_marlin_shard(
763
param, shard_size, shard_offset)
@@ -1107,8 +1107,8 @@ class QKVParallelLinear(ColumnParallelLinear):
1107
# If quantized, we need to adjust the offset and size to account
1108
# for the packing.
1109
if packed_dim == output_dim:
1110
- shard_size = shard_size // param.pack_factor
1111
- shard_offset = shard_offset // param.pack_factor
1110
+ shard_size = shard_size // param.packed_factor
1111
+ shard_offset = shard_offset // param.packed_factor
1112
1113
# Special case for Marlin.
1114
shard_size, shard_offset = adjust_marlin_shard(
@@ -1155,8 +1155,8 @@ class QKVParallelLinear(ColumnParallelLinear):
1155
# for the packing.
1156
packed_dim = getattr(param, "packed_dim", None)
1157
if packed_dim == output_dim:
1158
- shard_size = shard_size // param.pack_factor
1159
- shard_offset = shard_offset // param.pack_factor
1158
+ shard_size = shard_size // param.packed_factor
1159
+ shard_offset = shard_offset // param.packed_factor
1160
1161
# Special case for Marlin.
1162
shard_size, shard_offset = adjust_marlin_shard(