tensorflow-neuron for inf1 released in SDK 2.20.2 is not compatible with latest SDK 2.21 runtime

[jeffhataws]

Currently, the tensorflow-neuron for inf1 released in SDK 2.20.2 is not compatible with latest SDK 2.21 runtime. You may see the following error when executing a traced model:

........                                                                                                                                                     
Compiler status PASS                                                                                                                                         
2024-Dec-23 22:35:20.691409 28490:28590 ERROR  TDRV:arch_init                               Architecture not found 1                                         
                                                                                                                                                             
python3: /opt/workspace/KaenaRuntime/tdrv/tdrv_arch_type.c:96: tdrv_get_number_devices_on_instance: Assertion `device_count != 0' failed.                    

There was no update to tensorflow-neuron for inf1 in SDK 2.21.

A work-around is to use tensorflow-neuron for inf1 released in SDK 2.20.2 with runtime from SDK 2.20.2 only.

[cdbhatia-aws]

Here is a temporary work-around for this issue.

sudo apt-get remove -y aws-neuronx-dkms aws-neuronx-collectives aws-neuronx-runtime-lib aws-neuronx-tools

sudo apt-get install aws-neuronx-dkms=2.18.* -y
sudo apt-get install aws-neuronx-collectives=2.22.* -y
sudo apt-get install aws-neuronx-runtime-lib=2.22.* -y
sudo apt-get install aws-neuronx-tools=2.19.* -y