Hansong Zhang

Results 90 comments of Hansong Zhang

[Android] Error running a transformer decoder-based neural network

So it means the model can run if all inputs are ones if the dimension of the input tensor is correct, unless there is a dynamic shape in the model...

[Draft][XNNPACK] Share workspace across delegate instances

I will give a bit more patience for bot to merge this

Crash (native stack) for llama2 demo on Android

Hi @AmarOk1412 do you have the logs for the system? Any logs from `lmkd`? With Pixel 6 it might get OOM killed

Crash (native stack) for llama2 demo on Android

@AmarOk1412 Thank you for the updates! So you see no output tokens at all, and it just crashed?

Crash (native stack) for llama2 demo on Android

Hi @AmarOk1412 does it work for you on recent build?

Crash (native stack) for llama2 demo on Android

Could you please try with the `main` branch and remove `pip-out`?

Crash (native stack) for llama2 demo on Android

Hi @AmarOk1412 probably need to clean up everything like `pip-out` and `cmake-out*` when we switch between release versions 😅

[v0.4.0] Release Tracker

Link to landed trunk PR (if applicable): * https://github.com/pytorch/executorch/pull/5439 Link to release branch PR: * https://github.com/pytorch/executorch/pull/5456 Criteria Category: Docs -

Segmentation Fault when implementing llama/stories110M Android phone deployment

Hi @BESTTOOLBOX Thank you for reporting! This is a known issue and will be fixed by https://github.com/digantdesai/XNNPACK/pull/12 when we update the XNNPACK commit in ExecuTorch. You can also see https://github.com/pytorch/executorch/pull/4304/files...

Add Android test for #491

Instrumentation test is merged. Need to add to OSS CI. So far blocked on 1. Need to have a unified 2/3 runner to try with both 2 and 3 model....