Skip to content

[ROCM][RL] Shuffle Weight In-Place to Preserve Parameter Attributes #21825

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
HaiShaw merged 2 commits into from
Apr 3, 2026
+5 −7
Merged

[ROCM][RL] Shuffle Weight In-Place to Preserve Parameter Attributes #21825

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f1f298c
1
zyzshishui Apr 3, 2026
156f6b9
1
zyzshishui Apr 3, 2026
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
12 changes: 5 additions & 7 deletions python/sglang/srt/layers/quantization/unquant.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,7 @@
LinearMethodBase,
QuantizeMethodBase,
)
from sglang.srt.layers.utils import MultiPlatformOp
from sglang.srt.layers.utils import MultiPlatformOp, copy_or_rebind_param
from sglang.srt.utils import (
cpu_has_amx_support,
get_bool_env_var,
Expand Down Expand Up @@ -233,14 +233,12 @@ def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
# because aiter CK kernels don't support all GEMM dimensions
_should_use_aiter_moe = _use_aiter and get_moe_runner_backend().is_auto()
if _should_use_aiter_moe:
layer.w13_weight = torch.nn.Parameter(
shuffle_weight(layer.w13_weight.data, (16, 16)),
requires_grad=False,
copy_or_rebind_param(
layer, "w13_weight", shuffle_weight(layer.w13_weight.data, (16, 16))
)
torch.cuda.empty_cache()
layer.w2_weight = torch.nn.Parameter(
shuffle_weight(layer.w2_weight.data, (16, 16)),
requires_grad=False,
copy_or_rebind_param(
layer, "w2_weight", shuffle_weight(layer.w2_weight.data, (16, 16))
)
torch.cuda.empty_cache()

Expand Down
Loading