Skip to content

fix(te-plugin): handle TE 2.15+ tuple return from _Linear / _GroupedLinear #1481

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
kevalmorabia97 merged 1 commit into from
May 13, 2026
+8 −0
Merged

fix(te-plugin): handle TE 2.15+ tuple return from _Linear / _GroupedLinear #1481

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
8 changes: 8 additions & 0 deletions modelopt/torch/quantization/plugins/transformer_engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,6 +93,10 @@ def te_quantized_linear_fn(package, func_name, self, *args, **kwargs):
new_args[weight_pos] = self.weight_quantizer(args[weight_pos])
new_args[inp_pos] = self.input_quantizer(args[inp_pos])
output = getattr(package, func_name)(*new_args, **kwargs)
# TE 2.15+ returns `(out, new_weight_workspace)`; TE <= 2.14 returns just `out`.
# Only the activation tensor participates in output quantization.
if isinstance(output, tuple):
return (self.output_quantizer(output[0]), *output[1:])
return self.output_quantizer(output)

# Override the quantized linear function
Expand Down Expand Up @@ -181,6 +185,10 @@ def te_grouped_quantized_linear_fn(package, func_name, self, *args):
for i in range(weights_start, weights_start + num_gemms):
new_args[i] = self.weight_quantizer(args[i])
output = getattr(package, func_name)(*new_args)
# TE 2.15+ returns `(out, new_workspaces)`; TE <= 2.14 returns just `out`.
# Only the activation tensor participates in output quantization.
if isinstance(output, tuple):
return (self.output_quantizer(output[0]), *output[1:])
return self.output_quantizer(output)

# Override the quantized linear function
Expand Down
Loading