Hi @Sergii Ivakhno, What DBR version are you running for this notebook?
NOTE:- Make sure to create Databricks Runtime ML for the cluster and attach it to this notebook. (You cannot run this exercise in the standard Databricks runtime without "ML.")
ML runtime is optimized for deep learning, and all related components (TensorFlow, Horovod, Keras, XGBoost, etc.) are already built-in. (You don’t need to install these components by yourself.)
Built-in HorovodRunner on ML runtime helps Horovod to run on Apache Spark™. (Horovod (by Uber) has an efficient parameter-sharing mechanism and is beneficial for scaling.)