Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
@@ -705,8 +705,6 @@ class Param2MoEModel(nn.Module):
|
||||
class Param2MoEMixtureOfExperts(MixtureOfExperts):
|
||||
"""Implements the vLLM MixtureOfExperts protocol for Param2MoE."""
|
||||
|
||||
expert_weights: list[torch.Tensor]
|
||||
|
||||
def extract_moe_parameters(self, example_moe: Param2MoEMoEBlock | None) -> None:
|
||||
if example_moe is None:
|
||||
raise RuntimeError(
|
||||
@@ -745,24 +743,6 @@ class Param2MoEMixtureOfExperts(MixtureOfExperts):
|
||||
if hasattr(fused, "update_expert_map"):
|
||||
fused.update_expert_map()
|
||||
|
||||
def set_eplb_state(
|
||||
self,
|
||||
expert_load_view: torch.Tensor,
|
||||
logical_to_physical_map: torch.Tensor,
|
||||
logical_replica_count: torch.Tensor,
|
||||
) -> None:
|
||||
self.expert_weights = []
|
||||
for layer_idx, layer in enumerate(self.moe_layers):
|
||||
if hasattr(layer, "get_expert_weights"):
|
||||
self.expert_weights.append(layer.get_expert_weights())
|
||||
if hasattr(layer, "set_eplb_state"):
|
||||
layer.set_eplb_state(
|
||||
moe_layer_idx=layer_idx,
|
||||
expert_load_view=expert_load_view,
|
||||
logical_to_physical_map=logical_to_physical_map,
|
||||
logical_replica_count=logical_replica_count,
|
||||
)
|
||||
|
||||
|
||||
class Param2MoEForCausalLM(
|
||||
nn.Module, SupportsPP, SupportsLoRA, Param2MoEMixtureOfExperts
|
||||
|
||||
Reference in New Issue
Block a user