update moe training tests with scaled grouped mm test and use verbose mode (#3451)

Author: danielvegamyhre

test/prototype/moe_training/test_everything.sh

@@ -12,8 +12,9 @@ IS_ROCM=$(rocm-smi --version || true)
 # These tests do not work on ROCm yet
 if [ -z "$IS_ROCM" ]
 then
-pytest test/prototype/moe_training/test_kernels.py -s
-pytest test/prototype/moe_training/test_training.py -s
+pytest test/prototype/moe_training/test_kernels.py -s -v
+pytest test/prototype/moe_training/test_scaled_grouped_mm.py -s -v 
+pytest test/prototype/moe_training/test_training.py -s -v
 ./test/prototype/moe_training/test_fsdp.sh
 ./test/prototype/moe_training/test_tp.sh
 ./test/prototype/moe_training/test_fsdp_tp.sh

test/prototype/moe_training/test_fsdp.sh

@@ -1 +1 @@
-torchrun --nproc_per_node=2 --local-ranks-filter=0 -m pytest test/prototype/moe_training/test_fsdp.py -s
+torchrun --nproc_per_node=2 --local-ranks-filter=0 -m pytest test/prototype/moe_training/test_fsdp.py -s -v

test/prototype/moe_training/test_fsdp_tp.sh

@@ -1 +1 @@
-torchrun --nproc_per_node=4 --local-ranks-filter=0 -m pytest test/prototype/moe_training/test_fsdp_tp.py -s
+torchrun --nproc_per_node=4 --local-ranks-filter=0 -m pytest test/prototype/moe_training/test_fsdp_tp.py -s -v

test/prototype/moe_training/test_tp.sh

@@ -1 +1 @@
-torchrun --nproc_per_node=2 --local-ranks-filter=0 -m pytest test/prototype/moe_training/test_tp.py -s
+torchrun --nproc_per_node=2 --local-ranks-filter=0 -m pytest test/prototype/moe_training/test_tp.py -s -v