when training with datasets with truncation set to "split" and when the datalength is greater than the featuers length configured, the features_offset variable is left un-initialized and is later used un-initialized which crashes the script. This commit fixes that...
when training with datasets with truncation set to "split" and when the datalength is greater than the featuers length configured, the features_offset variable is left un-initialized and is later used un-initialized which crashes the script. This commit fixes that...