Really interesting to hear about the AMX facility. If I train an AI model in Azure and export the container workload to run the model on some Azure Stack HCI on the Edge, will it automatically start to take advantage of the AMX tiles ?
Hi @bentaylor3408, great question! When you export a container workload with an AI model trained in Azure to run on Azure Stack HCI on the Edge, the model itself doesn't automatically take advantage of AMX tiles. To leverage the AMX tiles, you need to ensure that the software stack, including libraries and frameworks (like TensorFlow or PyTorch or openvino), supports AMX. Additionally, the deployment environment on Azure Stack HCI needs to be configured to utilize AMX. Make sure to use updated versions of these frameworks that have optimizations for AMX. Or use Openvino that offers support for AMX by default
Really interesting to hear about the AMX facility. If I train an AI model in Azure and export the container workload to run the model on some Azure Stack HCI on the Edge, will it automatically start to take advantage of the AMX tiles ?
As from my limited knowledge, it will not for now.
Hi @bentaylor3408, great question! When you export a container workload with an AI model trained in Azure to run on Azure Stack HCI on the Edge, the model itself doesn't automatically take advantage of AMX tiles. To leverage the AMX tiles, you need to ensure that the software stack, including libraries and frameworks (like TensorFlow or PyTorch or openvino), supports AMX. Additionally, the deployment environment on Azure Stack HCI needs to be configured to utilize AMX. Make sure to use updated versions of these frameworks that have optimizations for AMX. Or use Openvino that offers support for AMX by default