[Kernel] Expand MoE weight loading + Add Fused Marlin MoE Kernel (#7527)
Co-authored-by: ElizaWszola <eliza@neuralmagic.com>
This commit is contained in:
co-authored by
ElizaWszola
parent
5844017285
commit
8678a69ab5
@@ -73,6 +73,7 @@ class MixtralMoE(nn.Module):
|
||||
self.hidden_size = hidden_size
|
||||
|
||||
# Gate always runs at half / full precision for now.
|
||||
|
||||
self.gate = ReplicatedLinear(hidden_size,
|
||||
num_experts,
|
||||
bias=False,
|
||||
|
||||
Reference in New Issue
Block a user