[Core] Ensure LoRA linear respect the base_layer's tp_size and tp_rank (#25487)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
@@ -18,7 +18,6 @@ class ReplicatedLinearWithLoRA(BaseLinearLayerWithLoRA):
|
||||
def __init__(self, base_layer: ReplicatedLinear) -> None:
|
||||
super().__init__(base_layer, )
|
||||
# To ensure interface compatibility, set to 1 always.
|
||||
self.tp_size = 1
|
||||
self.output_size = self.base_layer.output_size
|
||||
self.n_slices = 1
|
||||
|
||||
|
||||
Reference in New Issue
Block a user