Support Llama 4 for fused_marlin_moe (#20457)
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
@@ -700,6 +700,7 @@ class ModelOptNvFp4FusedMoE(FusedMoEMethodBase):
|
||||
global_scale1=layer.w13_weight_scale_2,
|
||||
global_scale2=layer.w2_weight_scale_2,
|
||||
quant_type_id=scalar_types.float4_e2m1f.id,
|
||||
apply_router_weight_on_input=apply_router_weight_on_input,
|
||||
global_num_experts=global_num_experts,
|
||||
expert_map=expert_map)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user