Bug with tensorflow 2.5.0+nv21.08 => Need 2.60 - how install it in jetson nano?

Hello,

Under jupyterlab installed in my jetson nano, with tensorflow 2.5.0+nv21.08 if I do the following code my jupyterlab crash but it works under google colab (tensorflow 2.6.0). So I think that tf 2.5.0 is buggy ?

Seems that there is no version 2.6.0 available in jetson nano :
https://docs.nvidia.com/deeplearning/frameworks/install-tf-jetson-platform-release-notes/tf-jetson-rel.html#tf-jetson-rel

Is it possible to install it anyway ?

Thanks a lot

x = tf.convert_to_tensor([1,2],dtype="int32")
y = tf.zeros(shape=(2),dtype="int32")
z = tf.concat([x,y],axis=0)
tf.roll(z,1,axis=0)

Hi,

We check this issue with tensorflow-2.5.0+nv21.8 on both Nano and Xavier.
The sample can work well on Xavier but meet an “too many resources requested for launch” error on Nano.

Do you also observe the same error as below?

2021-09-07 10:45:22.596081: F tensorflow/core/kernels/roll_op_gpu.cu.cc:84] Non-OK-status: GpuLaunchKernel(RollKernel<T>, cfg.block_count, cfg.thread_per_block, 0, d.stream(), cfg.virtual_thread_count, num_dims, input, output, reinterpret_cast<const int32*>(dim_buf), reinterpret_cast<const int32*>(thres_buf), reinterpret_cast<const int64*>(range_buf)) status: Internal: too many resources requested for launch

If yes, this is a known issue and we are still working on this.
You can find more details in the below topic:

Thanks.

Thanks for your answer.
I have no debug message, only a pop up telling me that kernel crash…
Perhaps I can do something to get a debug message ?

There is no update from you for a period, assuming this is not an issue any more.
Hence we are closing this topic. If need further support, please open a new one.
Thanks

Hi,

Do you get any log on the jupyterlab console?
Thanks.