Qualcomm AI Engine Direct - Docs Update

.claude/skills/qualcomm/SKILL.md

@@ -83,12 +83,12 @@ LD_LIBRARY_PATH=/path/to/executorch/build-x86/lib:/path/to/qnn_sdk/lib/x86_64-li
 PYTHONPATH=$(dirname $EXECUTORCH_ROOT) \
 python backends/qualcomm/tests/test_qnn_delegate.py \
     TestQNNFloatingPointOperator.test_qnn_backend_abs \
-    -H $HOST -s $DEVICE_SERIAL -m SM8850 -b build-android -a /path/to/artifacts
+    --host $HOST --device $DEVICE_SERIAL --soc_model SM8850 --build_folder build-android -a /path/to/artifacts
 ```
 
 > **Note (build from source):** Set `PYTHONPATH` to the parent directory of the executorch repo root. Required because `executorch.examples.qualcomm` lives in the source tree and is not installed into site-packages.
 
-Required flags: `-m` (SoC model), `-b` (Android build dir). Optional: `-s` (device serial), `-H` (host), `-a` (artifact dir), `-c` (compile only), `-x` (run on x86_64).
+Required flags: `--soc_model` (SoC model), `--build_folder` (Android build dir). Optional: `--device` (device serial), `--host` (host), `-a` (artifact dir), `--compile_only`, `--enable_x86_64`.
 
 **Test classes:**
 

.claude/skills/qualcomm/qnn_intermediate_debugger.md

@@ -137,12 +137,12 @@ Present the result as a fenced bash block, prefixed by a one-line note of what c
 >
 > ```bash
 > python -m examples.qualcomm.util_scripts.my_model_debug \
->     -H $HOST -s $DEVICE_SERIAL -m $SOC_MODEL -b build-android \
+>     --host $HOST --device $DEVICE_SERIAL --soc_model $SOC_MODEL --build_folder build-android \
 >     -d /path/to/dataset -a ./my_model_debug \
 >     --dump_intermediate_outputs
 > ```
 
-If the user did not give you a runnable original command in step 1 (e.g. they pasted only the script path), do **not** fabricate values for `-H` / `-s` / `-m` / `-b` / `-a` / `-d`. Stop and ask before printing — wrong device or SoC values waste a full export + on-device run.
+If the user did not give you a runnable original command in step 1 (e.g. they pasted only the script path), do **not** fabricate values for `--host` / `--device` / `--soc_model` / `--build_folder` / `-a` / `-d`. Stop and ask before printing — wrong device or SoC values waste a full export + on-device run.
 
 After the command runs, the artifact dir will contain SVG / CSV reports — green nodes pass, red nodes fail the comparator threshold. That's the first place to look for the layer that introduces the gap.
 

.wiki/backends/qnn/overview.md

@@ -66,7 +66,7 @@ There are two code paths for exporting LLMs to QNN. **Use `examples/qualcomm/oss
 ```bash
 # CORRECT path for QNN LLM export
 python examples/qualcomm/oss_scripts/llama/llama.py \
-  -b build-android -m SM8650 \
+  --build_folder build-android --soc_model SM8650 \
   --decoder_model qwen3-0_6b \
   --model_mode hybrid --max_seq_len 1024 --prefill_ar_len 128 \
   --prompt "..." --tasks wikitext --limit 1 --compile_only
@@ -131,7 +131,7 @@ Use `--compile_only` to export, quantize, and compile a `.pte` without running o
 
 ```bash
 python examples/qualcomm/oss_scripts/llama/llama.py \
-  -b build-android -m SM8650 --compile_only \
+  --build_folder build-android --soc_model SM8650 --compile_only \
   --decoder_model qwen3-0_6b ...
 ```
 

.wiki/quantization/debugging.md

@@ -22,7 +22,7 @@ Quantized LLMs (especially 1B parameter models with 4-bit weights) frequently pr
 ```bash
 # Use the QNN-specific script with better quantization
 python examples/qualcomm/oss_scripts/llama/llama.py \
-  --compile_only -m SM8750 \
+  --compile_only --soc_model SM8750 \
   --model_mode hybrid \
   --decoder_model <model> \
   ...

backends/qualcomm/README.md

@@ -120,7 +120,7 @@ PRs are always welcome to help improve the codebase in a comprehensive manner. B
     ```bash
     cd $PATH_TO_EXECUTORCH
     # example usage of performing unit test
-    python backends/qualcomm/tests/test_qnn_delegate.py -k TestQNNQuantizedOperator.test_qnn_backend_layer_norm -s $DEVICE_SERIAL -m SM8650 -b build-android/ -a $PATH_TO_TEST_ARTIFACTS
+    python backends/qualcomm/tests/test_qnn_delegate.py -k TestQNNQuantizedOperator.test_qnn_backend_layer_norm --device $DEVICE_SERIAL --soc_model SM8650 --build_folder build-android/ -a $PATH_TO_TEST_ARTIFACTS
     ```
     The test graph is expected to have 1 delegated node with only placeholders / output nodes being left. Check the execution report for more information.
 

backends/qualcomm/debugger/README.md

@@ -13,7 +13,7 @@ pip install qairt-visualizer
 ## Quick start
 This command launches an interactive GUI interface to visualize the `optrace` and `QHAS` results.
 ```
-python -m examples.qualcomm.util_scripts.qairt_visualizer_demo -H ${host} -s {device} -m ${SOC_MODEL} -b build-android -a ${path_to_output_folder} --online_prepare
+python -m examples.qualcomm.util_scripts.qairt_visualizer_demo --host ${host} --device {device} --soc_model ${SOC_MODEL} --build_folder build-android -a ${path_to_output_folder} --online_prepare
 ```
 - If online prepare mode is `enabled`, the following artifacts will be generated:
     - `model`.dlc
@@ -259,7 +259,7 @@ An Inception_V3 demo script is provided at [qnn_intermediate_debugger_demo.py](.
 Before running, ensure the dataset is downloaded. An example dataset can be retrieved [here](https://www.kaggle.com/datasets/ifigotin/imagenetmini-1000).
 
 ```bash
-python -m examples.qualcomm.util_scripts.qnn_intermediate_debugger_demo -b build-android -s $DEVICE_SERIAL -m $SOC_MODEL -d path/to/imagenet/val --dump_intermediate_outputs
+python -m examples.qualcomm.util_scripts.qnn_intermediate_debugger_demo --build_folder build-android --device $DEVICE_SERIAL --soc_model $SOC_MODEL -d path/to/imagenet/val --dump_intermediate_outputs
 ```
 
 ## Limitations
@@ -290,7 +290,7 @@ The difference between the two values represents heap consumed during context ex
 ```bash
 python backends/qualcomm/tests/test_qnn_delegate.py \
     TestQNNQuantizedUtils.test_qnn_backend_runtime_option_heap_profile \
-    -b build-android -H ${HOST} -s ${SN} -m ${SOC_MODEL}
+    --build_folder build-android --host ${HOST} --device ${SN} --soc_model ${SOC_MODEL}
 ```
 
 See [test_qnn_delegate.py](../tests/test_qnn_delegate.py) for the full test implementation.

docs/source/backends-qualcomm.md

@@ -149,7 +149,7 @@ We use deeplab-v3-resnet101 as an example in this tutorial. Run below commands t
 ```bash
 cd $EXECUTORCH_ROOT
 
-python -m examples.qualcomm.scripts.deeplab_v3 -b build-android -m SM8550 --compile_only --download
+python -m examples.qualcomm.scripts.deeplab_v3 --build_folder build-android --soc_model SM8550 --compile_only --download
 ```
 
 You might see something like below:
@@ -287,9 +287,9 @@ The model is merely executed. If we want to feed real inputs and get model outpu
 ```bash
 cd $EXECUTORCH_ROOT
 # android
-python -m examples.qualcomm.scripts.deeplab_v3 -b build-android -m SM8550 --download -s <device_serial>
+python -m examples.qualcomm.scripts.deeplab_v3 --build_folder build-android --soc_model SM8550 --download --device <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
+python -m examples.qualcomm.scripts.deeplab_v3 --build_folder build-oe-linux --soc_model SXR1230P --download --device <device_serial> --target aarch64-oe-linux-gcc9.3
 ```
 The `<device_serial>` can be found by `adb devices` command.
 

docs/source/llm/build-run-llama3-qualcomm-ai-engine-direct-backend.md

@@ -67,7 +67,7 @@ To export with the Qualcomm AI Engine Direct Backend, ensure the following:
 
 ```bash
 # export llama
-python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --checkpoint consolidated.00.pth --params params.json --tokenizer_model tokenizer.model --decoder_model llama3_2-3b_instruct --model_mode kv --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1 --compile_only
+python examples/qualcomm/oss_scripts/llama/llama.py --build_folder build-android --device ${SERIAL_NUM} --soc_model ${SOC_MODEL} --checkpoint consolidated.00.pth --params params.json --tokenizer_model tokenizer.model --decoder_model llama3_2-3b_instruct --model_mode kv --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1 --compile_only
 ```
 Note: end-to-end [instructions](https://github.com/pytorch/executorch/blob/main/examples/qualcomm/oss_scripts/llama/README.md)
 
@@ -81,7 +81,7 @@ Note: end-to-end [instructions](https://github.com/pytorch/executorch/blob/main/
 **3.3 Run model**
 ```bash
 # Run llama
-python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --checkpoint consolidated.00.pth --params params.json --tokenizer_model tokenizer.model --decoder_model llama3_2-3b_instruct --model_mode kv --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1 --pre_gen_pte ${PATH_TO_ARTIFACT}
+python examples/qualcomm/oss_scripts/llama/llama.py --build_folder build-android --device ${SERIAL_NUM} --soc_model ${SOC_MODEL} --checkpoint consolidated.00.pth --params params.json --tokenizer_model tokenizer.model --decoder_model llama3_2-3b_instruct --model_mode kv --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1 --pre_gen_pte ${PATH_TO_ARTIFACT}
 ```
 
 ## What is coming?

examples/qualcomm/custom_op/README.md

@@ -292,9 +292,9 @@ Registers `torch.ops.my_ops.mul3.default` (multiply by 3) and delegates it via `
 ```bash
 python3 examples/qualcomm/custom_op/custom_ops_1.py \
   --build_folder build-android \
-  -s <device_serial> \
-  -H <host> \
-  -m SM8650 \
+  --device <device_serial> \
+  --host <host> \
+  --soc_model SM8650 \
   --op_package_dir examples/qualcomm/custom_op/example_op_package_htp/ExampleOpPackage \
   --build_op_package
 ```
@@ -303,7 +303,7 @@ python3 examples/qualcomm/custom_op/custom_ops_1.py \
 ```bash
 python3 examples/qualcomm/custom_op/custom_ops_1.py \
   --build_folder build-x86 \
-  -m SM8650 \
+  --soc_model SM8650 \
   --op_package_dir examples/qualcomm/custom_op/example_op_package_htp/ExampleOpPackage \
   --build_op_package \
   --enable_x86_64
@@ -317,9 +317,9 @@ Registers `torch.ops.my_ops.split_custom.default` (splits a tensor into two halv
 ```bash
 python3 examples/qualcomm/custom_op/custom_ops_2.py \
   --build_folder build-android \
-  -s <device_serial> \
-  -H <host> \
-  -m SM8650 \
+  --device <device_serial> \
+  --host <host> \
+  --soc_model SM8650 \
   --op_package_dir examples/qualcomm/custom_op/example_op_package_htp_multi_output/SplitCustomOpPackage \
   --build_op_package
 ```
@@ -328,7 +328,7 @@ python3 examples/qualcomm/custom_op/custom_ops_2.py \
 ```bash
 python3 examples/qualcomm/custom_op/custom_ops_2.py \
   --build_folder build-x86 \
-  -m SM8650 \
+  --soc_model SM8650 \
   --op_package_dir examples/qualcomm/custom_op/example_op_package_htp_multi_output/SplitCustomOpPackage \
   --build_op_package \
   --enable_x86_64

examples/qualcomm/qaihub_scripts/llama/README.md

@@ -30,7 +30,7 @@ Please refer to [Check context binary version](../../README.md#check-context-bin
 #### Step4: Run default examples
 ```bash
 # AIHUB_CONTEXT_BINARIES: ${PATH_TO_AIHUB_WORKSPACE}/build/llama_v2_7b_chat_quantized
-python examples/qualcomm/qaihub_scripts/llama/llama2/qaihub_llama2_7b.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --context_binaries ${AIHUB_CONTEXT_BINARIES} --tokenizer_bin tokenizer.bin --prompt "What is Python?"
+python examples/qualcomm/qaihub_scripts/llama/llama2/qaihub_llama2_7b.py --build_folder build-android --device ${SERIAL_NUM} --soc_model ${SOC_MODEL} --context_binaries ${AIHUB_CONTEXT_BINARIES} --tokenizer_bin tokenizer.bin --prompt "What is Python?"
 ```
 
 ## Llama-3-8b-chat-hf
@@ -53,5 +53,5 @@ Please refer to [Check context binary version](../../README.md#check-context-bin
 #### Step4: Run default examples
 ```bash
 # AIHUB_CONTEXT_BINARIES: ${PATH_TO_AIHUB_WORKSPACE}/build/llama_v3_8b_chat_quantized
-python examples/qualcomm/qaihub_scripts/llama/llama3/qaihub_llama3_8b.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --context_binaries ${AIHUB_CONTEXT_BINARIES} --tokenizer_model tokenizer.model --prompt "What is baseball?"
+python examples/qualcomm/qaihub_scripts/llama/llama3/qaihub_llama3_8b.py --build_folder build-android --device ${SERIAL_NUM} --soc_model ${SOC_MODEL} --context_binaries ${AIHUB_CONTEXT_BINARIES} --tokenizer_model tokenizer.model --prompt "What is baseball?"
 ```

examples/qualcomm/qaihub_scripts/stable_diffusion/README.md

@@ -32,7 +32,7 @@ Please refer to [Check context binary version](../../README.md#check-context-bin
 #### Step5: Run default example
 In this example, we execute the script for 20 time steps with the `prompt` 'a photo of an astronaut riding a horse on mars':
 ```bash
-python examples/qualcomm/qaihub_scripts/stable_diffusion/qaihub_stable_diffusion.py -b build-android -m ${SOC_MODEL} --s ${SERIAL_NUM} --text_encoder_bin ${PATH_TO_TEXT_ENCODER_CONTEXT_BINARY} --unet_bin ${PATH_TO_UNET_CONTEXT_BINARY} --vae_bin ${PATH_TO_VAE_CONTEXT_BINARY} --vocab_json  ${PATH_TO_VOCAB_JSON_FILE} --num_time_steps 20 --prompt "a photo of an astronaut riding a horse on mars"
+python examples/qualcomm/qaihub_scripts/stable_diffusion/qaihub_stable_diffusion.py --build_folder build-android --soc_model ${SOC_MODEL} --device ${SERIAL_NUM} --text_encoder_bin ${PATH_TO_TEXT_ENCODER_CONTEXT_BINARY} --unet_bin ${PATH_TO_UNET_CONTEXT_BINARY} --vae_bin ${PATH_TO_VAE_CONTEXT_BINARY} --vocab_json  ${PATH_TO_VOCAB_JSON_FILE} --num_time_steps 20 --prompt "a photo of an astronaut riding a horse on mars"
 ```
 - Please replace `${PATH_TO_TEXT_ENCODER_CONTEXT_BINARY}`, `${PATH_TO_UNET_CONTEXT_BINARY}`, and `${PATH_TO_VAE_CONTEXT_BINARY}` with the actual paths to your AI Hub context binary files.
 - Please replace `${PATH_TO_VOCAB_JSON_FILE}` with the actual path to your vocab.json file.

examples/qualcomm/util_scripts/README.md

@@ -84,7 +84,7 @@ This section describes how to generate an ET record for a .pte program using the
     ```bash
     # Example usage to generate ET record and inspect execution statistics
     PYTHONPATH=.. python -m examples.qualcomm.util_scripts.gen_etrecord \
-      -b build-android \
+      --build_folder build-android \
       --device $DEVICE_SERIAL \
       --soc_model SM8750 \
     ```