Skip to content

Decompose after export in export_llama #15951

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

Draft
wants to merge 1 commit into
base: main
Choose a base branch
from
+2 −9
Draft

Decompose after export in export_llama #15951

Show file tree
Hide file tree
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
3 changes: 0 additions & 3 deletions examples/models/llama/source_transformation/quantize.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -194,9 +194,6 @@ def filter_fn(m, fqn):
),
filter_fn=filter_fn,
)

model = unwrap_tensor_subclass(model)

# TODO: deal with checkpoint / computation dtype decoupling.

if verbose:
Expand Down
8 changes: 2 additions & 6 deletions extension/llm/export/builder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,7 +38,6 @@
from torch.nn.attention import SDPBackend
from torchao.quantization.pt2e.quantize_pt2e import convert_pt2e, prepare_pt2e
from torchao.quantization.pt2e.quantizer import ComposableQuantizer, Quantizer
from torchao.utils import unwrap_tensor_subclass

FORMAT = "[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s"
logging.basicConfig(level=logging.INFO, format=FORMAT)
Expand Down Expand Up @@ -203,11 +202,6 @@ def _get_edge_config(self) -> EdgeCompileConfig:
return edge_config

def _export(self, module: Optional[torch.nn.Module] = None) -> ExportedProgram:
if module is not None:
unwrap_tensor_subclass(module)
else:
unwrap_tensor_subclass(self.model)

dynamic_shape = self._get_dynamic_shape()
# 1. torch.nn.attention.sdpa_kernel([SDPBackend.MATH]) is for bypassing the dynamo error when tracing
# 2. torch.no_grad() is for getting rid of the dropout (not sure why training ops will show up)
Expand All @@ -226,6 +220,8 @@ def _export(self, module: Optional[torch.nn.Module] = None) -> ExportedProgram:
dynamic_shapes=dynamic_shape,
strict=True,
)
# Functionalize the graph, and decompose subclasses from torchao quantize.
exported_module = exported_module.run_decompositions({})
return exported_module

def export(self) -> "LLMEdgeManager":
Expand Down
Loading