2023-12-14 09:52:05 +00:00
|
|
|
|
|
|
|
|
2023-12-15 08:38:51 +00:00
|
|
|
NUM_GPU=8
|
2023-12-14 09:52:05 +00:00
|
|
|
MODEL="8b"
|
|
|
|
SEQ_LENGTH=2048
|
|
|
|
BATCH_SIZE=1
|
|
|
|
LR=0.00001
|
|
|
|
|
|
|
|
# ep zero
|
|
|
|
# torchrun --standalone --nproc_per_node $NUM_GPU train.py \
|
|
|
|
# --num_epoch 1 \
|
|
|
|
# --model_name $MODEL \
|
|
|
|
# --plugin "ep_zero" \
|
|
|
|
# --batch_size $BATCH_SIZE \
|
|
|
|
# --lr $LR \
|
|
|
|
# --zero_stage 1 \
|
|
|
|
# --extra_dp_size 2
|
|
|
|
|
|
|
|
# ep
|
2023-12-15 08:38:51 +00:00
|
|
|
torchrun --standalone --nproc_per_node $NUM_GPU train.py \
|
2023-12-14 09:52:05 +00:00
|
|
|
--num_epoch 1 \
|
|
|
|
--model_name $MODEL \
|
2023-12-15 08:38:51 +00:00
|
|
|
--plugin "ep" \
|
2023-12-14 09:52:05 +00:00
|
|
|
--batch_size $BATCH_SIZE \
|
|
|
|
--lr $LR \
|
2023-12-15 08:38:51 +00:00
|
|
|
--zero_stage 2
|
2023-12-14 09:52:05 +00:00
|
|
|
|
|
|
|
# hybrid
|
|
|
|
# torchrun --standalone --nproc_per_node $NUM_GPU train.py \
|
|
|
|
# --num_epoch 1 \
|
|
|
|
# --model_name $MODEL \
|
|
|
|
# --plugin "hybrid" \
|
|
|
|
# --batch_size $BATCH_SIZE \
|
|
|
|
# --lr $LR \
|
|
|
|
# --zero_stage 1 \
|
|
|
|
# --pp_size 2 \
|
|
|
|
# --dp_size 1 \
|
|
|
|
# --ep_size 2 \
|