[Bugfix][MoE] Filter packed expert weights during EP loading (#49558)

Signed-off-by: aoshen02 <aoshen02@users.noreply.github.com>
Signed-off-by: aoshen02 <aoshen@inferact.ai>
Co-authored-by: aoshen02 <aoshen02@users.noreply.github.com>
This commit is contained in:
aoshen02
2026-08-04 16:32:03 +00:00
committed by GitHub
co-authored by aoshen02
parent 166f4e2dc3
commit 7153fd70f6
2 changed files with 16 additions and 1 deletions
@@ -185,6 +185,21 @@ class TestShouldSkipWeight:
"model.layers.0.mlp.experts.200.gate_proj.weight", self.local_ids
)
def test_local_packed_expert_not_skipped(self):
assert not should_skip_weight(
"model.layers.0.mlp.experts.10.gate_proj.weight_packed", self.local_ids
)
def test_remote_packed_expert_skipped(self):
assert should_skip_weight(
"model.layers.0.mlp.experts.200.gate_proj.weight_packed", self.local_ids
)
def test_remote_expert_scale_not_skipped(self):
assert not should_skip_weight(
"model.layers.0.mlp.experts.200.gate_proj.weight_scale", self.local_ids
)
def test_boundary_expert(self):
# Expert 47 is local (last one), 48 is not
assert not should_skip_weight(