Skip to content

[Feature]:Allow for Granite MoE Hybrid models with _only_ shared experts. #19652

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Jun 16, 2025
Merged

[Feature]:Allow for Granite MoE Hybrid models with _only_ shared experts. #19652

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
90d14e9
Allow for models with _only_ shared experts.
shawntan Jun 14, 2025
893c562
Formatting.
shawntan Jun 16, 2025
1602eab
Fix torch removal.
shawntan Jun 16, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
64 changes: 40 additions & 24 deletions vllm/model_executor/models/granitemoehybrid.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,13 +67,15 @@ def __init__(self,
activation=config.hidden_act,
quant_config=quant_config)

self.block_sparse_moe = GraniteMoeMoE(
num_experts=config.num_local_experts,
top_k=config.num_experts_per_tok,
hidden_size=config.hidden_size,
intermediate_size=config.intermediate_size,
quant_config=quant_config,
prefix=f"{prefix}.block_sparse_moe")
self.block_sparse_moe = None
if getattr(config, "num_local_experts", 0) > 0:
self.block_sparse_moe = GraniteMoeMoE(
num_experts=config.num_local_experts,
top_k=config.num_experts_per_tok,
hidden_size=config.hidden_size,
intermediate_size=config.intermediate_size,
quant_config=quant_config,
prefix=f"{prefix}.block_sparse_moe")

self.shared_mlp = None if \
getattr(config, 'shared_intermediate_size', 0) == 0 \
Expand Down Expand Up @@ -105,13 +107,19 @@ def forward(
residual = hidden_states
hidden_states = self.post_attention_layernorm(hidden_states)
if self.shared_mlp is None:
hidden_states = self.block_sparse_moe(hidden_states)
if self.block_sparse_moe is not None:
hidden_states = self.block_sparse_moe(hidden_states)
# else: skip
else:
# create a copy since block_sparse_moe modifies in-place
moe_hidden_states = hidden_states.clone()
moe_hidden_states = self.block_sparse_moe(moe_hidden_states)
hidden_states = moe_hidden_states + self.shared_mlp(hidden_states)
del moe_hidden_states
if self.block_sparse_moe is not None:
moe_hidden_states = hidden_states.clone()
moe_hidden_states = self.block_sparse_moe(moe_hidden_states)
hidden_states = moe_hidden_states + self.shared_mlp(
hidden_states)
del moe_hidden_states
else:
hidden_states = self.shared_mlp(hidden_states)
hidden_states = residual + hidden_states * self.residual_multiplier

return hidden_states, residual
Expand All @@ -137,13 +145,15 @@ def __init__(
quant_config=quant_config,
prefix=f"{prefix}.self_attn")

self.block_sparse_moe = GraniteMoeMoE(
num_experts=config.num_local_experts,
top_k=config.num_experts_per_tok,
hidden_size=config.hidden_size,
intermediate_size=config.intermediate_size,
quant_config=quant_config,
prefix=f"{prefix}.block_sparse_moe")
self.block_sparse_moe = None
if getattr(config, "num_local_experts", 0) > 0:
self.block_sparse_moe = GraniteMoeMoE(
num_experts=config.num_local_experts,
top_k=config.num_experts_per_tok,
hidden_size=config.hidden_size,
intermediate_size=config.intermediate_size,
quant_config=quant_config,
prefix=f"{prefix}.block_sparse_moe")

self.shared_mlp = None if \
getattr(config, 'shared_intermediate_size', 0) == 0 \
Expand Down Expand Up @@ -178,13 +188,19 @@ def forward(
residual = hidden_states
hidden_states = self.post_attention_layernorm(hidden_states)
if self.shared_mlp is None:
hidden_states = self.block_sparse_moe(hidden_states)
if self.block_sparse_moe is not None:
hidden_states = self.block_sparse_moe(hidden_states)
# else: skip
else:
# create a copy since block_sparse_moe modifies in-place
moe_hidden_states = hidden_states.clone()
moe_hidden_states = self.block_sparse_moe(moe_hidden_states)
hidden_states = moe_hidden_states + self.shared_mlp(hidden_states)
del moe_hidden_states
if self.block_sparse_moe is not None:
moe_hidden_states = hidden_states.clone()
moe_hidden_states = self.block_sparse_moe(moe_hidden_states)
hidden_states = moe_hidden_states + self.shared_mlp(
hidden_states)
del moe_hidden_states
else:
hidden_states = self.shared_mlp(hidden_states)
hidden_states = residual + hidden_states * self.residual_multiplier

return hidden_states, residual
Expand Down