-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.sh
executable file
·155 lines (139 loc) · 5.54 KB
/
train.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
ulimit -c unlimited
[ -z "${version}" ] && version='v1'
[ -z "${lr}" ] && lr=2e-4
[ -z "${end_lr}" ] && end_lr=1e-9
[ -z "${warmup_steps}" ] && warmup_steps=150000
[ -z "${total_steps}" ] && total_steps=1500000
[ -z "${layers}" ] && layers=18
[ -z "${hidden_size}" ] && hidden_size=768
[ -z "${ffn_size}" ] && ffn_size=768
[ -z "${num_head}" ] && num_head=32
[ -z "${batch_size}" ] && batch_size=256
[ -z "${update_freq}" ] && update_freq=1
[ -z "${seed}" ] && seed=1
[ -z "${clip_norm}" ] && clip_norm=5
[ -z "${data_path}" ] && data_path='./datasets/'
[ -z "${save_path}" ] && save_path='./logs/'
[ -z "${dropout}" ] && dropout=0.0
[ -z "${act_dropout}" ] && act_dropout=0.1
[ -z "${attn_dropout}" ] && attn_dropout=0.1
[ -z "${weight_decay}" ] && weight_decay=0.0
[ -z "${sandwich_ln}" ] && sandwich_ln="false"
[ -z "${droppath_prob}" ] && droppath_prob=0.1
[ -z "${noise_scale}" ] && noise_scale=0.2
[ -z "${mode_prob}" ] && mode_prob="0.2,0.2,0.6"
[ -z "${dataset_name}" ] && dataset_name="PCQM4M-LSC-V2-3D"
[ -z "${add_3d}" ] && add_3d="true"
[ -z "${no_2d}" ] && no_2d="false"
[ -z "${num_3d_bias_kernel}" ] && num_3d_bias_kernel=128
[ -z "${MASTER_PORT}" ] && MASTER_PORT=10086
[ -z "${OMPI_COMM_WORLD_SIZE}" ] && OMPI_COMM_WORLD_SIZE=1
[ -z "$save_prefix" ] && save_prefix=visnet-${version}
echo -e "\n\n"
echo "==================================MP==========================================="
[ -z "${n_gpu}" ] && n_gpu=$(nvidia-smi -L | wc -l)
echo "MASTER_ADDR: ${MASTER_ADDR}"
echo "MASTER_PORT: ${MASTER_PORT}"
echo "NCCL_SOCKET_IFNAME: ${NCCL_SOCKET_IFNAME}"
echo "OMPI_COMM_WORLD_RANK: ${OMPI_COMM_WORLD_RANK}"
echo "OMPI_COMM_WORLD_SIZE: ${OMPI_COMM_WORLD_SIZE}"
if [[ -z "${OMPI_COMM_WORLD_SIZE}" ]]
then
ddp_options=""
else
if (( $OMPI_COMM_WORLD_SIZE == 1))
then
ddp_options=""
else
ddp_options="--nnodes=$OMPI_COMM_WORLD_SIZE --node_rank=$OMPI_COMM_WORLD_RANK --master_addr=$MASTER_ADDR"
fi
fi
echo "ddp_options: ${ddp_options}"
echo "==============================================================================="
hyperparams=lr-$lr-end_lr-$end_lr-tsteps-$total_steps-wsteps-$warmup_steps-L$layers-D$hidden_size-F$ffn_size-H$num_head-SLN-$sandwich_ln-BS$((batch_size*n_gpu*OMPI_COMM_WORLD_SIZE*update_freq))-SEED$seed-CLIP$clip_norm-dp$dropout-attn_dp$attn_dropout-wd$weight_decay-dpp$droppath_prob-noisescale-$noise_scale-mode_prob-${mode_prob}
save_dir=$save_path/$save_prefix-$hyperparams
tsb_dir=$save_dir/tsb
mkdir -p $save_dir
echo -e "\n\n"
echo "=====================================ARGS======================================"
echo "arg0: $0"
echo "seed: ${seed}"
echo "batch_size: $((batch_size*n_gpu*OMPI_COMM_WORLD_SIZE*update_freq))"
echo "n_layers: ${layers}"
echo "lr: ${lr}"
echo "warmup_steps: ${warmup_steps}"
echo "total_steps: ${total_steps}"
echo "clip_norm: ${clip_norm}"
echo "hidden_size: ${hidden_size}"
echo "ffn_size: ${ffn_size}"
echo "sandwich_ln: ${sandwich_ln}"
echo "num_head: ${num_head}"
echo "update_freq: ${update_freq}"
echo "dropout: ${dropout}"
echo "attn_dropout: ${attn_dropout}"
echo "act_dropout: ${act_dropout}"
echo "weight_decay: ${weight_decay}"
echo "droppath_prob: ${droppath_prob}"
echo "noise_scale: ${noise_scale}"
echo "mode_prob: ${mode_prob} for {2D+3D, 2D, 3D}"
echo "save_dir: ${save_dir}"
echo "tsb_dir: ${tsb_dir}"
echo "data_dir: ${data_path}"
echo "==============================================================================="
# ENV
echo -e "\n\n"
echo "======================================ENV======================================"
echo 'Environment'
ulimit -c unlimited;
echo '\n\nhostname'
hostname
echo '\n\nnvidia-smi'
nvidia-smi
echo '\n\nls -alh'
ls -alh
echo -e '\n\nls ~ -alh'
ls ~ -alh
echo "torch version"
python -c "import torch; print(torch.__version__)"
echo "==============================================================================="
echo -e "\n\n"
echo "==================================ACTION ARGS==========================================="
if ( $sandwich_ln == "true")
then
action_args="--sandwich-ln "
else
action_args=""
fi
echo "action_args: ${action_args}"
if ( $add_3d == "true")
then
add_3d_args="--add-3d"
else
add_3d_args=""
fi
echo "add_3d_args: ${add_3d_args}"
if ( $no_2d == "true")
then
no_2d_args="--no-2d"
else
no_2d_args=""
fi
echo "no_2d_args: ${no_2d_args}"
echo "========================================================================================"
export NCCL_ASYNC_ERROR_HADNLING=1
export OMP_NUM_THREADS=1
python -m torch.distributed.launch --nproc_per_node=$n_gpu --master_port=$MASTER_PORT $ddp_options train.py \
--user-dir $(realpath ./Global-ViSNet) \
--data-path $data_path \
--num-workers 16 --ddp-backend=legacy_ddp \
--dataset-name $dataset_name \
--save-interval 10 \
--batch-size $batch_size --data-buffer-size 20 \
--task graph_prediction --criterion graph_prediction --arch global_visnet_${version}_base --num-classes 1 \
--lr $lr --end-learning-rate $end_lr --lr-scheduler polynomial_decay --power 1 \
--warmup-updates $warmup_steps --total-num-update $total_steps --max-update $total_steps --update-freq $update_freq \
--encoder-layers $layers --encoder-attention-heads $num_head $add_3d_args $no_2d_args --num-3d-bias-kernel $num_3d_bias_kernel \
--encoder-embed-dim $hidden_size --encoder-ffn-embed-dim $ffn_size --droppath-prob $droppath_prob \
--attention-dropout $attn_dropout --act-dropout $act_dropout --dropout $dropout --weight-decay $weight_decay \
--optimizer adam --adam-betas '(0.9, 0.999)' --adam-eps 1e-8 $action_args --clip-norm $clip_norm \
--tensorboard-logdir $tsb_dir --save-dir $save_dir --fp16 --noise-scale $noise_scale --mode-prob $mode_prob