Fix Transformers backend FP8 MoE and remove some boilerplate (#46820)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2026-06-27 00:16:05 +01:00
committed by GitHub
parent 2ff76a5e85
commit d8eb734d94
29 changed files with 80 additions and 156 deletions
+1 -2
View File
@@ -751,7 +751,7 @@ class Param2MoEMixtureOfExperts(MixtureOfExperts):
logical_to_physical_map: torch.Tensor,
logical_replica_count: torch.Tensor,
) -> None:
self.expert_weights.clear()
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())
@@ -832,7 +832,6 @@ class Param2MoEForCausalLM(
self.model.make_empty_intermediate_tensors
)
self.expert_weights: list[torch.Tensor] = []
self.num_moe_layers: int = 0
self.moe_layers: list = []
self.moe_mlp_layers: list = []