[MoE] Enable Shared/Routed Overlap For Latent MoE (Nemotron-H) (#32790)

Signed-off-by: dafrimi <dafrimi@nvidia.com>
This commit is contained in:
danielafrimi
2026-02-02 09:18:50 -05:00
committed by GitHub
parent 9eb58f8cf1
commit 0aca8b8c62
4 changed files with 303 additions and 58 deletions
+30 -42
View File
@@ -188,10 +188,29 @@ class NemotronHMoE(nn.Module):
prefix=f"{prefix}.shared_experts",
)
if self.use_latent_moe:
self.fc1_latent_proj = ReplicatedLinear(
input_size=config.hidden_size,
output_size=self.moe_hidden_size,
bias=config.mlp_bias,
quant_config=quant_config,
disable_tp=self.is_sequence_parallel,
prefix=f"{prefix}.fc1_latent_proj",
)
self.fc2_latent_proj = ReplicatedLinear(
input_size=self.moe_hidden_size,
output_size=config.hidden_size,
bias=config.mlp_bias,
quant_config=quant_config,
disable_tp=self.is_sequence_parallel,
prefix=f"{prefix}.fc2_latent_proj",
)
else:
self.fc1_latent_proj = None
self.fc2_latent_proj = None
self.experts = SharedFusedMoE(
# TODO: make it possible for shared experts to have
# different input in SharedFusedMoE
shared_experts=self.shared_experts if not self.use_latent_moe else None,
shared_experts=self.shared_experts,
num_experts=config.n_routed_experts,
top_k=config.num_experts_per_tok,
hidden_size=self.moe_hidden_size,
@@ -211,30 +230,9 @@ class NemotronHMoE(nn.Module):
num_redundant_experts=self.n_redundant_experts,
is_sequence_parallel=self.is_sequence_parallel,
router_logits_dtype=router_logits_dtype,
routed_input_transform=self.fc1_latent_proj,
)
if self.use_latent_moe:
self.fc1_latent_proj = ReplicatedLinear(
input_size=config.hidden_size,
output_size=self.moe_hidden_size,
bias=config.mlp_bias,
quant_config=quant_config,
disable_tp=self.is_sequence_parallel,
prefix=f"{prefix}.fc1_latent_proj",
)
self.fc2_latent_proj = ReplicatedLinear(
input_size=self.moe_hidden_size,
output_size=config.hidden_size,
bias=config.mlp_bias,
quant_config=quant_config,
disable_tp=self.is_sequence_parallel,
prefix=f"{prefix}.fc2_latent_proj",
)
else:
self.fc1_latent_proj = None
self.fc2_latent_proj = None
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
num_tokens, hidden_dim = hidden_states.shape
hidden_states = hidden_states.view(-1, hidden_dim)
@@ -244,38 +242,28 @@ class NemotronHMoE(nn.Module):
# router_logits: (num_tokens, n_experts)
router_logits, _ = self.gate(hidden_states.to(dtype=torch.float32))
shared_output = None
if self.use_latent_moe:
if self.shared_experts is not None:
shared_output = self.shared_experts(hidden_states)
hidden_states, _ = self.fc1_latent_proj(hidden_states)
fused_moe_out = self.experts(
# SharedFusedMoE handles:
# - shared experts (with original hidden_states)
# - routed_input_transform (fc1_latent_proj) for latent MoE
# - multistream parallelism between shared and routed experts
shared_output, final_hidden_states = self.experts(
hidden_states=hidden_states, router_logits=router_logits
)
if self.use_latent_moe:
_, final_hidden_states = fused_moe_out
else:
shared_output, final_hidden_states = fused_moe_out
# Fix FP16 overflow
# See DeepseekV2DecoderLayer for more details.
if hidden_states.dtype != torch.float16:
final_hidden_states *= self.routed_scaling_factor
elif self.shared_experts is not None:
assert shared_output is not None
shared_output *= 1.0 / self.routed_scaling_factor
# TODO: currently latent up_proj is done before all-reduce for simplicity.
# if and when shared experts will be part of SharedFusedMoE,
# we should do the up_proj after all-reduce,
# to have the all-reduce in the smaller latent dimension.
# TODO: See SharedFusedMoE.apply_routed_input_transform
# for bandwidth optimization
if self.use_latent_moe:
final_hidden_states, _ = self.fc2_latent_proj(final_hidden_states)
if self.shared_experts is not None:
assert shared_output is not None
final_hidden_states += shared_output
if self.is_sequence_parallel: