[feat] FlashInfer CuteDSL MegaMoE integration (#54049)
Signed-off-by: jdebache <jdebache@nvidia.com>
This commit is contained in:
@@ -657,7 +657,7 @@ def test_deepseek_v4_mega_moe_does_not_double_add_fused_shared_expert(
|
||||
|
||||
moe = DeepseekV4MoE.__new__(DeepseekV4MoE)
|
||||
torch.nn.Module.__init__(moe)
|
||||
moe.use_mega_moe = True
|
||||
moe.use_native_mega_moe = True
|
||||
moe.gate = FakeGate()
|
||||
moe.experts = FakeExperts()
|
||||
moe.shared_experts = FakeSharedExperts()
|
||||
|
||||
Reference in New Issue
Block a user