[Refactor] Delete dead code in models (#51838)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
@@ -92,7 +92,6 @@ class MixtralMoE(nn.Module):
|
||||
self.hidden_size = hidden_size
|
||||
|
||||
self.ep_group = get_ep_group().device_group
|
||||
self.ep_rank = get_ep_group().rank_in_group
|
||||
self.ep_size = self.ep_group.size()
|
||||
|
||||
# Expert Parallelism Load balancing settings.
|
||||
@@ -105,11 +104,6 @@ class MixtralMoE(nn.Module):
|
||||
self.n_redundant_experts = parallel_config.eplb_config.num_redundant_experts
|
||||
self.n_physical_experts = self.n_logical_experts + self.n_redundant_experts
|
||||
self.n_local_physical_experts = self.n_physical_experts // self.ep_size
|
||||
self.physical_expert_start = self.ep_rank * self.n_local_physical_experts
|
||||
self.physical_expert_end = (
|
||||
self.physical_expert_start + self.n_local_physical_experts
|
||||
)
|
||||
|
||||
# Gate always runs at half / full precision for now.
|
||||
|
||||
self.gate = ReplicatedLinear(
|
||||
|
||||
Reference in New Issue
Block a user