Skip to content

Cortex-M: preserve calibration input layout #22572

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
rascani wants to merge 1 commit into
base: main
Choose a base branch
from
+5 −10
Draft

Cortex-M: preserve calibration input layout #22572

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
2 changes: 0 additions & 2 deletions backends/arm/scripts/aot_arm_compiler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -981,8 +981,6 @@ def _to_channels_last(x):
calibration_samples = [example_inputs]

for sample in calibration_samples:
if not args.cortex_m_explicit_layout:
sample = tuple(_to_channels_last(x) for x in sample)
prepared(*sample)

model_quant = convert_pt2e(prepared)
Expand Down
5 changes: 1 addition & 4 deletions backends/cortex_m/test/models/test_mobilenet_v2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,10 +37,7 @@
}

# Use larger sample set for calibration to get better quantization
calibration_samples = [
(torch.randn(1, 3, 224, 224).to(memory_format=torch.channels_last),)
for _ in range(100)
]
calibration_samples = [(torch.randn(1, 3, 224, 224),) for _ in range(100)]

test_cases = {
"mobilenet_v2": McuTestCase(
Expand Down
5 changes: 1 addition & 4 deletions backends/cortex_m/test/models/test_mobilenet_v3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,10 +39,7 @@
}

# Use bigger sample set for calibration.
calibration_samples = [
(torch.randn(1, 3, 232, 232).to(memory_format=torch.channels_last),)
for i in (range(100))
]
calibration_samples = [(torch.randn(1, 3, 232, 232),) for _ in range(100)]

test_cases = {
"mobilenet_v3_small": McuTestCase(
Expand Down
3 changes: 3 additions & 0 deletions docs/source/backends/arm-cortex-m/arm-cortex-m-overview.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -100,6 +100,9 @@ quantized = convert_pt2e(prepared)
quantized_exported_program = torch.export.export(quantized, (example_input,))
```

Calibration observes logical tensor values, so calibration inputs do not need
to use the same memory format as the export example.

### 2. Lower to edge and apply Cortex-M passes

Lower to the edge dialect with the backend's `EdgeCompileConfig`, then run the `CortexMPassManager` to replace quantized subgraphs with CMSIS-NN operator implementations:
Expand Down
Loading