Skip to content

Quant tool: Add nodes_to_exclude in get_qnn_qdq_config #23779

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
jambayk merged 2 commits into from
Feb 27, 2025
Merged

Quant tool: Add nodes_to_exclude in get_qnn_qdq_config #23779

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fa76fc3
nodes_to_exclude in get_qnn_qdq_config
jambayk Feb 21, 2025
c85f5dd
lint fix
jambayk Feb 21, 2025
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
7 changes: 7 additions & 0 deletions onnxruntime/python/tools/quantization/execution_providers/qnn/quant_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -55,6 +55,7 @@ def get_qnn_qdq_config(
stride: int | None = None,
calibration_providers: list[str] | None = None,
op_types_to_quantize: list[str] | None = None,
nodes_to_exclude: list[str] | None = None,
) -> StaticQuantConfig:
"""
Returns a static quantization configuration suitable for running QDQ models on QNN EP.
Expand Down Expand Up @@ -122,6 +123,8 @@ def get_qnn_qdq_config(
calibration_providers: Execution providers to run the session during calibration. Default is None which uses
[ "CPUExecutionProvider" ].
op_types_to_quantize: If set to None, all operator types will be quantized except for OP_TYPES_TO_EXCLUDE
nodes_to_exclude: List of nodes names to exclude from quantization. The nodes in this list will be excluded from
quantization when it is not None.

Returns:
A StaticQuantConfig object
Expand Down Expand Up @@ -167,10 +170,13 @@ def get_qnn_qdq_config(
)

op_types_to_quantize_set = set(op_types_to_quantize) if op_types_to_quantize else None
nodes_to_exclude_set = set(nodes_to_exclude) if nodes_to_exclude else None

for node in model.graph.node:
if op_types_to_quantize_set and node.op_type not in op_types_to_quantize_set:
continue
if nodes_to_exclude_set and node.name in nodes_to_exclude_set:
continue
op_types.add(node.op_type)
qnn_compat.process_node(node)

Expand Down Expand Up @@ -201,6 +207,7 @@ def get_qnn_qdq_config(
op_types_to_quantize=op_types_to_quantize
if op_types_to_quantize
else list(op_types.difference(OP_TYPES_TO_EXCLUDE)),
nodes_to_exclude=nodes_to_exclude,
per_channel=per_channel,
use_external_data_format=(model_has_external_data or model.ByteSize() >= MODEL_SIZE_THRESHOLD),
calibration_providers=calibration_providers,
Expand Down