add seed parameter

MxNet/Classification/RN50v1.5/MxNet_Imagenet.sh

@@ -3,9 +3,10 @@ for (( i = 0; i < 3; i++ )); do # 3 runs
     mkdir /imagenet/MxNet/run_"${i}"/
     echo "Created Folder run_${i}"
     ./runner -n 2 -b 192 --num-epochs 90 --mode train_val \
+    --save-frequency 1 \
     --amp --dllogger-log /imagenet/MxNet/run_"${i}"/run_"${i}".log \
     --workspace /imagenet/MxNet/run_"${i}"/ --data-backend synthetic \
-    --data-mxnet-threads 80 \
-    --data-root /imagenet --warmup 0
+    --data-root /imagenet --warmup 0 \
+    --seed "${i}"
     echo "Run ${i} done"
 done

PyTorch/Classification/ConvNets/PyTorch_Imagenet.sh

@@ -6,6 +6,7 @@ for (( i = 0; i < 3; i++ )); do # 3 runs
     python ./multiproc.py --nproc_per_node 2 ./main.py --arch resnet50 \
     --amp --data-backend pytorch -j 20 --epochs 90 -b 192 \
     --warmup 0 --raport-file /imagenet/PyTorch/run_"${i}"/run_"${i}".json \
-    --workspace /imagenet/PyTorch/run_"${i}" /imagenet
+    --workspace /imagenet/PyTorch/run_"${i}" /imagenet \
+    --seed "${i}"
     echo "Run ${i} done"
 done

TensorFlow2/Recommendation/DLRM/main.py

@@ -226,7 +226,8 @@ def main(argv):
         optimizers = [mlp_optimizer]
 
     elif FLAGS.optimizer == 'adam':
-        embedding_optimizer = tfa.optimizers.LazyAdam(lr=FLAGS.learning_rate)
+        embedding_optimizer = tfa.optimizers.LazyAdam(lr=FLAGS.learning_rate)0
+
         mlp_optimizer = tf.keras.optimizers.Adam(lr=FLAGS.learning_rate)
         if FLAGS.amp:
             embedding_optimizer = LossScaleOptimizer(embedding_optimizer,