Skip to content

[Quantization] Fix ModelOpt pre-quantized loading #14188

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
yzhautouskay wants to merge 1 commit into
base: main
Choose a base branch
from
+7 −1
Open

[Quantization] Fix ModelOpt pre-quantized loading #14188

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: 7 additions & 1 deletion src/diffusers/quantizers/modelopt/modelopt_quantizer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,7 +69,11 @@ def check_if_quantized_param(

module, tensor_name = get_module_from_name(model, param_name)
if self.pre_quantized:
return True
# ModelOpt restoration recreates quantizer state such as `_amax` and
# `_scale` as buffers. Let the regular low-memory loader materialize
# those buffers on their target device. Only parameters need the
# wrapper-aware assignment below.
return tensor_name in module._parameters
elif is_quantized(module) and "weight" in tensor_name:
return True
return False
Expand All @@ -92,6 +96,8 @@ def create_quantized_param(
dtype = kwargs.get("dtype", torch.float32)
module, tensor_name = get_module_from_name(model, param_name)
if self.pre_quantized:
if tensor_name not in module._parameters:
raise ValueError(f"Expected {param_name} to be a parameter in the restored ModelOpt graph.")
module._parameters[tensor_name] = torch.nn.Parameter(param_value.to(device=target_device))
else:
set_module_tensor_to_device(model, param_name, target_device, param_value, dtype)
Expand Down
Loading