Deepspeed with SLURM, not seeing deepspeed distributed and only seeing 1 GPU working instead of multigpu

[mars747]

When I run
python -u tutorial_train.py --lr 2e-5 --gpus 8 --batch_size 12
logged into a slurm node, it starts on each GPU and each GPU would write log images.

But when I run the job with sbatch using a script like this:

#!/bin/bash
#SBATCH --partition=V100x8
#SBATCH --gres=gpu:8
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=40
#SBATCH --output=logs/job_%j.out
#SBATCH --error=logs/job_%j.err

export PATH=/tools/cuda_10.0.130/bin:$PATH
export PATH=/tools/anaconda3-5.1.0/bin:$PATH

source activate automatic1111_new

module load cuda/11.7.0
module load cudnn/8.6.0
module load nccl/2.12.12

python -u tutorial_train.py --lr 1e-1 --gpus 8 --batch_size 12 

I only see log images from 1 gpu and I don't see log messages like this:

initializing deepspeed distributed: GLOBAL_RANK: 4, MEMBER: 5/8
initializing deepspeed distributed: GLOBAL_RANK: 5, MEMBER: 6/8
initializing deepspeed distributed: GLOBAL_RANK: 1, MEMBER: 2/8
initializing deepspeed distributed: GLOBAL_RANK: 7, MEMBER: 8/8
initializing deepspeed distributed: GLOBAL_RANK: 6, MEMBER: 7/8
initializing deepspeed distributed: GLOBAL_RANK: 2, MEMBER: 3/8
initializing deepspeed distributed: GLOBAL_RANK: 3, MEMBER: 4/8

Do I need to do something extra for Deepspeed to launch other than the normal sbatch script?