Provide clear guidance on required QNN versions and SoC serials

[guangy10]

📚 The doc issue

The potential doc improvement is from this GitHub Issue: #8640.

Today in our tutorial for QNN (https://pytorch.org/executorch/main/build-run-qualcomm-ai-engine-direct-backend.html), it doesn't provide a clear guidance on how each SoC number and supported QNN versions should be used given a specific model.

@winskuo-quic wdyt?

Suggest a potential alternative/fix

No response

cc @mergennachin @byjlw

[guangy10]

cc: @GregoryComer as you are also working on documentation for better UX

[cccclai]

QNN version is unrelated to SoCs. I think we should just use qnn 2.28, as described in https://pytorch.org/executorch/main/build-run-qualcomm-ai-engine-direct-backend.html

[guangy10]

QNN version is unrelated to SoCs. I think we should just use qnn 2.28, as described in https://pytorch.org/executorch/main/build-run-qualcomm-ai-engine-direct-backend.html

Do we expect all models can work on all listed SoCs? If yes, that's good, otherwise, let's be explicit about it. I recall I was having hard time to figure it out in the past.

[cccclai]

QNN version is unrelated to SoCs. I think we should just use qnn 2.28, as described in https://pytorch.org/executorch/main/build-run-qualcomm-ai-engine-direct-backend.html

Do we expect all models can work on all listed SoCs? If yes, that's good, otherwise, let's be explicit about it. I recall I was having hard time to figure it out in the past.

Hmm, what does all models mean? I think many PyTorch models are supported by the QNN delegates, but not all models (I wish)

[guangy10]

QNN version is unrelated to SoCs. I think we should just use qnn 2.28, as described in https://pytorch.org/executorch/main/build-run-qualcomm-ai-engine-direct-backend.html

Do we expect all models can work on all listed SoCs? If yes, that's good, otherwise, let's be explicit about it. I recall I was having hard time to figure it out in the past.

Hmm, what does all models mean? I think many PyTorch models are supported by the QNN delegates, but not all models (I wish)

For any model supported by QNN, do we expect it can work on any of the listed SoCs?

[winskuo-quic]

QNN version is unrelated to SoCs. I think we should just use qnn 2.28, as described in https://pytorch.org/executorch/main/build-run-qualcomm-ai-engine-direct-backend.html

Do we expect all models can work on all listed SoCs? If yes, that's good, otherwise, let's be explicit about it. I recall I was having hard time to figure it out in the past.

Hmm, what does all models mean? I think many PyTorch models are supported by the QNN delegates, but not all models (I wish)

For any model supported by QNN, do we expect it can work on any of the listed SoCs?

@guangy10,
Yes, I believe that any model supported by QNN should work on any SoCs list.