Skip to content

[Bugfix] Fix marlin moe fallback logic for llama4 #18042

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 2 commits into from
May 13, 2025
Merged

[Bugfix] Fix marlin moe fallback logic for llama4 #18042

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4a797b3
Fix marlin moe fallback logic for llama4
mgoin May 13, 2025
e2287a4
Add llama 4 to weight loading test
mgoin May 13, 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
3 changes: 2 additions & 1 deletion tests/weight_loading/models-large.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,4 +4,5 @@ compressed-tensors, nm-testing/Mixtral-8x7B-Instruct-v0.1-W8A16-quantized, main
compressed-tensors, nm-testing/test-w4a16-mixtral-actorder-group, main
gptq_marlin, TheBloke/Mixtral-8x7B-v0.1-GPTQ, main
gptq_marlin, TheBloke/Mixtral-8x7B-v0.1-GPTQ, gptq-8bit-128g-actorder_True
awq_marlin, casperhansen/deepseek-coder-v2-instruct-awq, main
awq_marlin, casperhansen/deepseek-coder-v2-instruct-awq, main
compressed-tensors, RedHatAI/Llama-4-Scout-17B-16E-Instruct-quantized.w4a16, main
2 changes: 1 addition & 1 deletion vllm/model_executor/layers/fused_moe/layer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -480,6 +480,7 @@ def __init__(
self.custom_routing_function = custom_routing_function
self.scoring_func = scoring_func
self.e_score_correction_bias = e_score_correction_bias
self.apply_router_weight_on_input = apply_router_weight_on_input
self.activation = activation

if self.scoring_func != "softmax" and not self.use_grouped_topk:
Expand All @@ -498,7 +499,6 @@ def __init__(
self.quant_method = quant_config.get_quant_method(self, prefix)
assert self.quant_method is not None

self.apply_router_weight_on_input = apply_router_weight_on_input
moe_quant_params = {
"num_experts": self.local_num_experts,
"hidden_size": hidden_size,
Expand Down
12 changes: 9 additions & 3 deletions vllm/model_executor/layers/quantization/utils/marlin_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -171,13 +171,19 @@ def check_moe_marlin_supports_layer(layer: LinearBase, group_size: int) \
-> bool:
hidden_size = layer.hidden_size
intermediate_size_per_partition = layer.intermediate_size_per_partition
# apply_router_weight_on_input is not supported for moe marlin
supports_router_weight = not layer.apply_router_weight_on_input
# moe marlin requires the activation to be silu
supports_activation = layer.activation == "silu"

# gate-up: (n, k) = (intermediate_size_per_partition * 2, hidden_size)
# down: (n, k) = (hidden_size, intermediate_size_per_partition)
# moe marlin requires n % 128 == 0 and k % 64 == 0
return hidden_size % 128 == 0 and \
intermediate_size_per_partition % max(64, group_size) == 0 and \
group_size in [-1, 32, 64, 128]
supports_shape = hidden_size % 128 == 0 and \
intermediate_size_per_partition % max(64, group_size) == 0
supports_group_size = group_size in [-1, 32, 64, 128]
return supports_shape and supports_group_size and \
supports_router_weight and supports_activation


def marlin_make_workspace(output_size_per_partition: int,
Expand Down