Skip to content

add chunk moe support on INC #2270

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

Open
wants to merge 1 commit into
base: r1-woq
Choose a base branch
from
Open

add chunk moe support on INC #2270

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
10 changes: 10 additions & 0 deletions neural_compressor/torch/algorithms/fp8_quant/_quant_common/helper_modules.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -841,6 +841,12 @@ def post_process(self):
self.w2_list[i].weight = torch.nn.Parameter(self.w2_list[i].weight.squeeze().t().contiguous())
htcore.mark_step()

def _get_extra_kwargs(self, tokens_num: int):
kwargs = {}
if hasattr(self.orig_mod, "_get_extra_kwargs"):
kwargs = self.orig_mod._get_extra_kwargs(tokens_num)
return kwargs

def forward_measure(
self,
x,
Expand Down Expand Up @@ -900,6 +906,9 @@ def forward_quant(
scale_w1 = [self.w13_list[i].scale_weight for i in experts_range]
scale_w2 = [self.w2_list[i].scale_weight for i in experts_range]
qinput = self.quant_input(hidden_states)

tokens_num, hidden_dim = hidden_states.shape
extra_kwargs = self._get_extra_kwargs(tokens_num)
output = self.dynamic_moe_op(
hidden_states=qinput,
expert_routing_table=expert_routing_table,
Expand All @@ -914,6 +923,7 @@ def forward_quant(
activation=activation,
experts_min=self.experts_min,
experts_max=self.experts_max,
**extra_kwargs
)
return output

Expand Down