CNNScoreVariants problem? Slow or stopped?

Dear all, I'm working trying to use CNN to annotate my VCF, adopting wdl from gatk4-cnn-variant-filter in GitHub (gatk-workflows)

My project is a custom target capture panel and I've scattered my bams (each 2megabytes) + vcfs (each 38 kilobytes)
and each pair is fed to CNNScoreVariants. There are no error messages but the process is not consuming any CPU and looks like it's stuck for half a day without any progress.

Would truly appreciate if you can offer your expert advice. Thank you!

Command as follow: (gatk4 was installed by conda)

gatk --java-options -Xmx5G \
CNNScoreVariants \
-I /mnt/operation/RedCellNGS/Processed_bam_vcf_1/bam/cromwell-executions/CNN_filtering/982a2730-51ee-4e06-aa2a-6f8e9ab0ab3d/call-CNNScoreVariants/shard-0/inputs/-1372610583/11H28680-2_bamout.bam \
-R /mnt/operation/RedCellNGS/Processed_bam_vcf_1/bam/cromwell-executions/CNN_filtering/982a2730-51ee-4e06-aa2a-6f8e9ab0ab3d/call-CNNScoreVariants/shard-0/inputs/131803870/Homo_sapiens_assembly38.fasta \
-V /mnt/operation/RedCellNGS/Processed_bam_vcf_1/bam/cromwell-executions/CNN_filtering/982a2730-51ee-4e06-aa2a-6f8e9ab0ab3d/call-CNNScoreVariants/shard-0/inputs/-1372610583/11H28680-2_hc4.vcf.gz \
-O 11H28680-2_cnn_annotated.vcf.gz \
-L /mnt/operation/RedCellNGS/Processed_bam_vcf_1/bam/cromwell-executions/CNN_filtering/982a2730-51ee-4e06-aa2a-6f8e9ab0ab3d/call-CNNScoreVariants/shard-0/inputs/1460496992/0000-scattered.interval_list \
--tensor-type read_tensor \
--inference-batch-size 2 \
--transfer-batch-size 2

Stderr content is in next post