Skip to content
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

Jump to bottom

Fix subgraph quantization regression in onnxruntime 1.17 #19421

Merged
merged 4 commits into from
Feb 13, 2024
Merged

Fix subgraph quantization regression in onnxruntime 1.17 #19421

Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
a3cec52
fix subgraph quantization bug
fxmarty Feb 5, 2024
f01260a
precise comment
fxmarty Feb 5, 2024
818d5b0
add test
fxmarty Feb 6, 2024
a9a1f24
lint
fxmarty Feb 8, 2024
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
Next Next commit
fix subgraph quantization bug
  • Loading branch information
@fxmarty
fxmarty committed Feb 5, 2024
commit a3cec5218b8092efb37ddcc4f0c1a985fcf14e75
11 changes: 9 additions & 2 deletions onnxruntime/python/tools/quantization/onnx_quantizer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1336,9 +1336,16 @@ def _dequantize_value(self, value_name):
if (value_name in self.quantized_value_map) and (value_name not in self.generated_value_names):
quantized_value = self.quantized_value_map[value_name]
# Add DequantizeLinear Node for this input

scale_init = find_by_name(quantized_value.scale_name, self.model.initializer())
# axis is not specified so scale_init must be a scalar.
assert onnx.numpy_helper.to_array(scale_init).size == 1

# In case we are working with subgraphs, the graph `producer_name` is set to `"onnx-quantizer"` in the `quantize_subgraph` method. In this case, the scale initializer may be on the top level graph, so this check can not be done.
if (
self.model.model.producer_name != "onnx-quantizer"
or (self.model.model.producer_name == "onnx-quantizer" and scale_init is not None)
):
# axis is not specified so scale_init must be a scalar.
assert onnx.numpy_helper.to_array(scale_init).size == 1

dqlinear_name = value_name + "_DequantizeLinear"
dqlinear_node = self.model.find_node_by_name(dqlinear_name, self.new_nodes, self.model.graph())
Expand Down