Philipp Moritz

No, it has been flaking quite a bit, I'm re-running to make sure it didn't get more flaky :)

On H100s, changing the number of SMs to 256 brought the best improvement in terms of throughtput for me (but still not quite matching current master). It was measured with...

As a workaround, this can be used: ``` python from numpy.testing import assert_equal def ray_assert_equal(a, b, attrs=[]): attrs.extend(["ray_deallocator", "ray_objectid"]) a = a.__dict__ b = b.__dict__ A = dict((i, a[i]) for...

Thanks for removing the layer name until the weight refactor is ready @dsikka

Fixed in triton 2.3.1 now :)

- You should do `clear output` in jupyter before committing the notebook to make sure there is no extra output. - `There are two types of views:` -> `There are...

Hey, I managed to reproduce the problem you are running into, most likely there is a problem in the way the CentOS 6 binaries are built. We are looking into...

Currently, ppc64le is not supported. If you get TensorFlow to build with JavaCPP on ppc64le, you can follow these instructions to create JARs: https://github.com/amplab/SparkNet/blob/master/doc/creating-jars.md

Thanks @jsongcse, I tried your new build.sbt and it works! How can I tell build-cpu.sbt instead of build.sbt? I didn't find anything about how to use multiple .sbt files in...

Have you tried using SBT? That is the way we are building the project. The maven file was contributed by @javadba and if it is currently broken, we should update...