Skip to content

GitLab

Unverified Commit c4d62618 authored by yuttian1's avatar yuttian1 Committed by GitHub
Browse files

Fix AWQ MoE marlin check issue in marlin_utils.py for AMD backend (#30102) 


Signed-off-by: default avataryuttian1 <yuttian@amd.com>
parent 62079d86
Hide whitespace changes
Inline Side-by-side
Showing with 4 additions and 0 deletions
vllm/model_executor/layers/quantization/utils/marlin_utils.py
View file @ c4d62618
...@@ -179,6 +179,8 @@ def check_marlin_supports_shape( ...@@ -179,6 +179,8 @@ def check_marlin_supports_shape(
def check_marlin_supports_layer(layer: LinearBase, group_size: int) -> bool: def check_marlin_supports_layer(layer: LinearBase, group_size: int) -> bool:
if current_platform.is_rocm():
return False
output_size_per_partition = ( output_size_per_partition = (
getattr(layer, "output_size_per_partition", None) or layer.output_size getattr(layer, "output_size_per_partition", None) or layer.output_size
) )
...@@ -195,6 +197,8 @@ def check_marlin_supports_layer(layer: LinearBase, group_size: int) -> bool: ...@@ -195,6 +197,8 @@ def check_marlin_supports_layer(layer: LinearBase, group_size: int) -> bool:
def check_moe_marlin_supports_layer(layer: LinearBase, group_size: int) -> bool: def check_moe_marlin_supports_layer(layer: LinearBase, group_size: int) -> bool:
if current_platform.is_rocm():
return False
hidden_size = layer.hidden_size hidden_size = layer.hidden_size
intermediate_size_per_partition = layer.intermediate_size_per_partition intermediate_size_per_partition = layer.intermediate_size_per_partition
# apply_router_weight_on_input is not supported for moe marlin # apply_router_weight_on_input is not supported for moe marlin
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment