馃摎 The doc issue
Summary
The "Deploying and running on device" section of the Qualcomm AI Engine Backend tutorial tells readers to compile DeepLabV3 with -b build-android -m SM8550.
https://docs.pytorch.org/executorch/stable/backends-qualcomm.html#aot-compile-a-model
Those short options were removed on 2026-06-16 by PR #19660. The tutorial was never updated, so some of the documented commands now fail at argument parsing.
This reproduces on v1.4.0 and on current HEAD of main.
What the docs say
Tutorial, AOT compile step:
main @ 42ebbc3d:
|
```bash |
|
cd $EXECUTORCH_ROOT |
|
|
|
python -m examples.qualcomm.scripts.deeplab_v3 -b build-android -m SM8550 --compile_only --download |
|
``` |
v1.4.0:
|
```bash |
|
cd $EXECUTORCH_ROOT |
|
|
|
python -m examples.qualcomm.scripts.deeplab_v3 -b build-android -m SM8550 --compile_only --download |
|
``` |
cd $EXECUTORCH_ROOT
python -m examples.qualcomm.scripts.deeplab_v3 -b build-android -m SM8550 --compile_only --download
The on-device step further down has the same problem with -s and -t:
|
The model is merely executed. If we want to feed real inputs and get model outputs, we can use |
|
```bash |
|
cd $EXECUTORCH_ROOT |
|
# android |
|
python -m examples.qualcomm.scripts.deeplab_v3 -b build-android -m SM8550 --download -s <device_serial> |
|
# (optional) linux embedded |
|
python -m examples.qualcomm.scripts.deeplab_v3 -b build-oe-linux -m SXR1230P --download -s <device_serial> -t aarch64-oe-linux-gcc-9.3 |
|
``` |
Line 292 needs a second fix beyond the flags: aarch64-oe-linux-gcc-9.3 is not a valid --target
value either. The choices are aarch64-android, aarch64-oe-linux-gcc9.3, and
aarch64-oe-linux-gcc11.2, so the extra hyphen is wrong in long form too:
|
parser.add_argument( |
|
"--target", |
|
help="Target platform for deployment", |
|
choices=[ |
|
"aarch64-android", |
|
"aarch64-oe-linux-gcc9.3", |
|
"aarch64-oe-linux-gcc11.2", |
|
], |
|
default="aarch64-android", |
|
type=str, |
|
) |
What happens when you run it
Running that first command:
$ cd $EXECUTORCH_ROOT
$ python -m examples.qualcomm.scripts.deeplab_v3 -b build-android -m SM8550 --compile_only --download
usage: deeplab_v3.py [-h] [--config_file CONFIG_FILE] [--soc_model SOC_MODEL]
[--build_folder BUILD_FOLDER] [--host HOST]
[--online_prepare] [--ip IP] [--port PORT]
[--skip_delegate_node_ids SKIP_DELEGATE_NODE_IDS]
[--skip_delegate_node_ops SKIP_DELEGATE_NODE_OPS]
[--compile_only] [--device DEVICE]
[--backend {htp,gpu,lpai}] [--shared_buffer]
[--skip_push] [--dump_intermediate_outputs]
[--profile_level {0,2,3}] [--enable_x86_64] [--ci]
[--seed SEED]
[--target {aarch64-android,aarch64-oe-linux-gcc9.3,aarch64-oe-linux-gcc11.2}]
[--pre_gen_pte PRE_GEN_PTE]
[--direct_build_folder DIRECT_BUILD_FOLDER]
[--htp_performance_mode {...}]
[-a ARTIFACT] [-d]
deeplab_v3.py: error: unrecognized arguments: -b build-android -m SM8550
$ echo $?
2Suggest a potential alternative/fix
The new correct command would be:
python -m examples.qualcomm.scripts.deeplab_v3 \
--build_folder build-android --soc_model SM8550 --compile_only --download(and I ran this one & it succeeds)
The same change affects other documents:
docs/source/llm/build-run-llama3-qualcomm-ai-engine-direct-backend.md
L70,
L84backends/qualcomm/README.md
L123backends/qualcomm/debugger/README.md
L16,
L262,
L291-L293examples/qualcomm/custom_op/README.md (4 command blocks)
L293-L299,
L304-L309,
L318-L324,
L329-L334examples/qualcomm/qaihub_scripts/llama/README.md
L33,
L56examples/qualcomm/qaihub_scripts/stable_diffusion/README.md
L35.claude/skills/qualcomm/SKILL.md
L80-L86,
L91
-- the second is prose listing -m/-b/-s/-H as the flags to use. Note -a there is still
valid: test_qnn_delegate.py registers its own -r, -a, -i, -p, -n, -e, -d, -F.
cc @cbilgin @psiddh
馃摎 The doc issue
Summary
The "Deploying and running on device" section of the Qualcomm AI Engine Backend tutorial tells readers to compile DeepLabV3 with
-b build-android -m SM8550.https://docs.pytorch.org/executorch/stable/backends-qualcomm.html#aot-compile-a-model
Those short options were removed on 2026-06-16 by PR #19660. The tutorial was never updated, so some of the documented commands now fail at argument parsing.
This reproduces on v1.4.0 and on current HEAD of
main.What the docs say
Tutorial, AOT compile step:
main@42ebbc3d:executorch/docs/source/backends-qualcomm.md
Lines 149 to 153 in 42ebbc3
v1.4.0:executorch/docs/source/backends-qualcomm.md
Lines 149 to 153 in 3dd7ccd
The on-device step further down has the same problem with
-sand-t:executorch/docs/source/backends-qualcomm.md
Lines 286 to 293 in 42ebbc3
Line 292 needs a second fix beyond the flags:
aarch64-oe-linux-gcc-9.3is not a valid--targetvalue either. The choices are
aarch64-android,aarch64-oe-linux-gcc9.3, andaarch64-oe-linux-gcc11.2, so the extra hyphen is wrong in long form too:executorch/backends/qualcomm/export_utils.py
Lines 851 to 861 in 42ebbc3
What happens when you run it
Running that first command:
Suggest a potential alternative/fix
The new correct command would be:
python -m examples.qualcomm.scripts.deeplab_v3 \ --build_folder build-android --soc_model SM8550 --compile_only --download(and I ran this one & it succeeds)
The same change affects other documents:
docs/source/llm/build-run-llama3-qualcomm-ai-engine-direct-backend.mdL70,
L84
backends/qualcomm/README.mdL123
backends/qualcomm/debugger/README.mdL16,
L262,
L291-L293
examples/qualcomm/custom_op/README.md(4 command blocks)L293-L299,
L304-L309,
L318-L324,
L329-L334
examples/qualcomm/qaihub_scripts/llama/README.mdL33,
L56
examples/qualcomm/qaihub_scripts/stable_diffusion/README.mdL35
.claude/skills/qualcomm/SKILL.mdL80-L86,
L91
-- the second is prose listing
-m/-b/-s/-Has the flags to use. Note-athere is stillvalid:
test_qnn_delegate.pyregisters its own-r,-a,-i,-p,-n,-e,-d,-F.cc @cbilgin @psiddh