Revert "[Kernel] Expand MoE weight loading + Add Fused Marlin MoE Kernel (#7527)" (#7764)

This commit is contained in:
Michael Goin
2024-08-22 03:42:14 +00:00
committed by GitHub
parent cde9183b40
commit aae74ef95c
15 changed files with 84 additions and 2374 deletions
+1 -1
View File
@@ -920,7 +920,7 @@ class JambaForCausalLM(nn.Module, HasInnerState):
weight_loader = param.weight_loader
weight_loader(param,
loaded_weight,
name,
weight_name,
shard_id=shard_id,
expert_id=expert_id)
break
-1
View File
@@ -73,7 +73,6 @@ 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,