Skip to content

Commit

Permalink
[CE] Add MP2-SP2-pp4-vpp2-SD2-stage1-mbs2-acc8 ce (PaddlePaddle#7774)
Browse files Browse the repository at this point in the history
  • Loading branch information
tianhaodongbd authored Jan 9, 2024
1 parent 393ac18 commit fc6ab70
Show file tree
Hide file tree
Showing 4 changed files with 95 additions and 5 deletions.
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,11 @@ mp_degree=2
pp_degree=1
sharding_degree=4
sharding=stage1
virtual_pp_degree=1
use_recompute=True
eval_freq=25
use_pipeline_parallel=False
sequence_parallel=False
bs_item=32
fp_item=bf16
run_mode=MP2-SD4-stage1-mbs4-acc2
Expand All @@ -26,8 +31,10 @@ max_iter=50000

model=gpt
micro_bs=4
acc=2
seed=3589

bash ./test_tipc/dygraph/hybrid_parallelism/ce_gpt/benchmark_common/prepare.sh
# run
bash ./test_tipc/dygraph/hybrid_parallelism/ce_gpt/benchmark_common/run_benchmark.sh ${model_item} ${fp_item} ${dp_degree} ${mp_degree} ${pp_degree} ${micro_bs} ${bs_item} ${run_mode} ${device_num} \
${max_iter} ${sharding} ${sharding_degree} 2>&1;
${max_iter} ${sharding} ${sharding_degree} ${virtual_pp_degree} ${use_recompute} ${eval_freq} ${use_pipeline_parallel} ${sequence_parallel} ${acc} ${seed} 2>&1;
Original file line number Diff line number Diff line change
@@ -0,0 +1,40 @@
# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

model_item=CE_gpt-345m_seqlen1024_pretrain
dp_degree=1
mp_degree=2
pp_degree=4
bs_item=32
fp_item=bf16
run_mode=MP2-SP2-PP4-VPP2-SD2-stage1-mbs2-acc8
device_num=N2C16
max_iter=50000
sharding=stage1
sharding_degree=2
virtual_pp_degree=2
use_recompute=True
eval_freq=25
use_pipeline_parallel=True
sequence_parallel=True

model=gpt
micro_bs=2
acc=8
seed=3589

bash ./test_tipc/dygraph/hybrid_parallelism/ce_gpt/benchmark_common/prepare.sh
# run
bash ./test_tipc/dygraph/hybrid_parallelism/ce_gpt/benchmark_common/run_benchmark.sh ${model_item} ${fp_item} ${dp_degree} ${mp_degree} ${pp_degree} ${micro_bs} ${bs_item} ${run_mode} ${device_num} \
${max_iter} ${sharding} ${sharding_degree} ${virtual_pp_degree} ${use_recompute} ${eval_freq} ${use_pipeline_parallel} ${sequence_parallel} ${acc} ${seed} 2>&1;
Original file line number Diff line number Diff line change
@@ -0,0 +1,40 @@
# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

model_item=gpt-345m_seqlen1024_pretrain
dp_degree=1
mp_degree=2
pp_degree=4
bs_item=32
fp_item=bf16
run_mode=MP2-SP2-PP4-VPP2-SD2-stage1-mbs2-acc8
device_num=N2C16
max_iter=100
sharding=stage1
sharding_degree=2

virtual_pp_degree=2
use_recompute=True
eval_freq=25
use_pipeline_parallel=True
sequence_parallel=True

model=gpt
micro_bs=2
acc=8

bash ./test_tipc/dygraph/hybrid_parallelism/ce_gpt/benchmark_common/prepare.sh
# run
bash ./test_tipc/dygraph/hybrid_parallelism/ce_gpt/benchmark_common/run_benchmark.sh ${model_item} ${fp_item} ${dp_degree} ${mp_degree} ${pp_degree} ${micro_bs} ${bs_item} ${run_mode} ${device_num} \
${max_iter} ${sharding} ${sharding_degree} ${virtual_pp_degree} ${use_recompute} ${eval_freq} ${use_pipeline_parallel} ${sequence_parallel} ${acc} 2>&1;
Original file line number Diff line number Diff line change
Expand Up @@ -37,11 +37,13 @@ function _set_params(){
sharding_degree=${12:-"1"}
num_workers=0 # (可选)
base_batch_size=$global_batch_size
virtual_pp_degree=${13:-"2"} # (可选) virtualpp数据并行度
vpp_degree=${13:-"1"} # (可选) virtualpp数据并行度
use_recompute=${14:-"True"} # (可选)是否打开recompute
eval_freq=${15:-"25"} # (可选)模型评估间隔
use_pipeline_parallel=${16:-"False"} # (可选)是否开启pipeline_parallel_config
sequence_parallel=${17:-"False"} # (可选)是否开启sequence_parallel
acc=${18:-"2"}
seed=${19:-"1234"}
# 以下为通用执行命令,无特殊可不用修改
model_name=${model_item}_bs${global_batch_size}_${fp_item}_${run_mode} # (必填) 且格式不要改动,与竞品名称对齐
device=${CUDA_VISIBLE_DEVICES//,/ }
Expand Down Expand Up @@ -108,10 +110,11 @@ function _train(){
--tensor_parallel_degree ${mp_degree} \
--pipeline_parallel_degree ${pp_degree} \
${pp_config_disable_partial_send_recv} \
--virtual_pp_degree ${vpp_degree} \
--sequence_parallel ${sequence_parallel} \
--split 949,50,1 \
--max_seq_length 1024 \
--seed 1234 \
--seed ${seed} \
--fuse_attention_qkv True \
--use_flash_attention True \
--bf16 ${bf16} \
Expand All @@ -125,7 +128,7 @@ function _train(){
--dataloader_num_workers 1 \
--eval_steps 1000 \
--disable_tqdm True \
--gradient_accumulation_steps 2 \
--gradient_accumulation_steps ${acc} \
--weight_decay 0.01\
--max_steps ${max_iter}\
--save_steps 5000\
Expand All @@ -150,7 +153,7 @@ function _train(){
run_pretrain.py ${train_cmd}"
workerlog_id=0
;;
DP8-mbs2-acc2|SD8-stage1-mbs2-acc2|SD8-stage2-mbs2-acc2|SD8-stage3-mbs2-acc2|MP2-SD4-stage1-mbs4-acc2|MP2-SP2-PP2-DP2-mbs8-acc2|MP8-mbs16-acc2|MP2-PP2-DP2-mbs8-acc2|MP2-PP2-SD2-Stage1-mbs8-acc2|MP2-SP2-PP2-SD2-Stage1-mbs8-acc2) echo "run run_mode: ${run_mode}"
DP8-mbs2-acc2|SD8-stage1-mbs2-acc2|SD8-stage2-mbs2-acc2|SD8-stage3-mbs2-acc2|MP2-SD4-stage1-mbs4-acc2|MP2-SP2-PP2-DP2-mbs8-acc2|MP8-mbs16-acc2|MP2-PP2-DP2-mbs8-acc2|MP2-PP2-SD2-Stage1-mbs8-acc2|MP2-SP2-PP2-SD2-Stage1-mbs8-acc2|MP2-SP2-PP4-VPP2-SD2-stage1-mbs2-acc8) echo "run run_mode: ${run_mode}"
train_cmd="python -m paddle.distributed.launch --log_dir=./mylog --devices=0,1,2,3,4,5,6,7 ${PADDLE_RANK_OPTION}\
run_pretrain.py ${train_cmd}"
workerlog_id=0
Expand Down

0 comments on commit fc6ab70

Please sign in to comment.