Koan-Sin Tan

Results 238 comments of Koan-Sin Tan

> Yes, we can run x86 based apps on windows_arm64 devices. But the problem is our "SNPE on Windows" libs are arm64 based. Not sure how to link x86 based...

> It seems like Windows for ARM64 now supports seamless interop between native ARM64 and emulated x86_64 code: https://blogs.windows.com/windowsdeveloper/2021/06/28/announcing-arm64ec-building-native-and-interoperable-apps-for-windows-11-on-arm/ I think we should be able to build the app in...

Flutter doesn't support windows arm64 (yet). The EC version ARM64 lib not supported by QTI yet. (There are potential hurdles ahead. We don't known we don't know). Compiling loadgen for...

I finally managed to build and run the Windows x86_64 version of the app on an Windows 10 machine with. - bazel 5.0, - tensorflow 2.8.0, - msys2, - Python...

@freedomtan to send a PR for TensorFlow 2.8.0 and Bazel 5.0.0. We'll try to merge it after we test all the backends are valid.

One thing annoying is that, although Android NDK is not used to build Windows app, we do require the path to ANDROID_NDK_HOME.

> how to execute .hwx model? That's a billion dollar question. Fortunately, @geohot answered that already, if you don't know where to start, please check https://github.com/geohot/tinygrad/blob/master/accel/ane/3_run/test.mm

No idea. To use IOReport, it seems you need root privilege or you have to disable amfi `amfi_get_out_of_my_way=1` on iOS. I don't have access to either jailbroken iPhone 11 or...

> Hi @badger200 , when I ran the test_ioreport binary on the iPhone13, it caused the following error: > > ``` > iPhone-13:/var/tmp/test root# ./test_ioreport > dyld[861]: Library not loaded:...

a minimal script to reproduce the issue ```python import keras import keras_nlp import tensorflow as tf os.environ["KAGGLE_USERNAME"] = '....' os.environ["KAGGLE_KEY"] = '...' os.environ["KERAS_BACKEND"] = "tensorflow" gemma_lm = keras_nlp.models.GemmaCausalLM.from_preset("gemma_2b_en") tf.saved_model.save(gemma_lm.backbone, '/tmp/gemma_saved_model/')...