Skip to content

Fix shared experts fusion error #6289

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 7 commits into from
May 30, 2025
Merged

Fix shared experts fusion error #6289

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
5b271d7
fix shared experts fusion error without quantization
lambert0312 May 14, 2025
1e4a680
lint
lambert0312 May 14, 2025
9bea8e8
Merge branch 'main' into fix_shared_experts_fusion_without_quantization
lambert0312 May 15, 2025
17dff9d
Merge branch 'main' into fix_shared_experts_fusion_without_quantization
Alcanderian May 16, 2025
4c8792f
Merge branch 'main' into fix_shared_experts_fusion_without_quantization
lambert0312 May 18, 2025
f90de29
Merge branch 'main' into fix_shared_experts_fusion_without_quantization
lambert0312 May 29, 2025
3d42d94
Merge branch 'main' into fix_shared_experts_fusion_without_quantization
zhyncs May 30, 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
50 changes: 38 additions & 12 deletions python/sglang/srt/models/deepseek_v2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1903,23 +1903,49 @@ def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]], is_nextn=Fal
if self.n_share_experts_fusion > 0:
weights_list = list(weights)
weights_dict = dict(weights_list)
if self.quant_config is None or self.quant_config.get_name() == "w8a8_int8":
suffix_list = [
"down_proj.weight",
"down_proj.weight_scale",
"gate_proj.weight",
"gate_proj.weight_scale",
"up_proj.weight",
"up_proj.weight_scale",
]
if self.quant_config is not None:
if self.quant_config.get_name() == "w8a8_int8":
suffix_list = [
"down_proj.weight",
"down_proj.weight_scale",
"gate_proj.weight",
"gate_proj.weight_scale",
"up_proj.weight",
"up_proj.weight_scale",
]
elif (
self.quant_config.get_name() == "fp8"
or self.quant_config.get_name() == "blockwise_int8"
):
suffix_list = [
"down_proj.weight",
"down_proj.weight_scale_inv",
"gate_proj.weight",
"gate_proj.weight_scale_inv",
"up_proj.weight",
"up_proj.weight_scale_inv",
]
elif self.quant_config.get_name() == "awq":
suffix_list = [
"down_proj.qweight",
"down_proj.qzeros",
"down_proj.scales",
"gate_proj.qweight",
"gate_proj.qzeros",
"gate_proj.scales",
"up_proj.qweight",
"up_proj.qzeros",
"up_proj.scales",
]
else:
raise ValueError(
f"Unsupported shared expert fusion for quantization: {self.quant_config.get_name()}."
)
else:
suffix_list = [
"down_proj.weight",
"down_proj.weight_scale_inv",
"gate_proj.weight",
"gate_proj.weight_scale_inv",
"up_proj.weight",
"up_proj.weight_scale_inv",
]
names_to_remove = []

Expand Down
Loading