File size: 5,374 Bytes
14a7d24
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
156
157
158
159
160
161
162
163
164
165
166
167
168
169
#!/bin/bash

if [[ $DEBUG -eq 1 ]]; then
    set -x
    env
    #LOG_LEVEL:0 - TRACE, 1 - DEBUG, 2 - INFO, 3 - WARNING, 4 - ERROR, 5 - CRITICAL, 6 - OFF
    export LOG_LEVEL_ALL_HCL=2
else
    export LOG_LEVEL_ALL_HCL=6
fi

if [ -z $BASE_PATH ]; then
	BASE_PATH="$( cd "$(dirname "$(readlink -f ./defaults.cfg)" )" && pwd)"
	PYTHONPATH=${BASE_PATH}:$PYTHONPATH
fi

TRAIN_SCRIPT=${BASE_PATH}/TensorFlow/computer_vision/Resnets/resnet_keras/resnet_ctl_imagenet_main.py
PT_VERSION=`python3 -c 'import sys; print(f"{sys.version_info[0]}.{sys.version_info[1]}")'`
TF_VERSION=`python3 -c "import tensorflow as tf; print(tf.__version__.replace('.', '_'))"`
PATCH_PATH=/usr/local/lib/python${PT_VERSION}/dist-packages/habana_frameworks/tensorflow/tf${TF_VERSION}/lib/habanalabs
# This is required for HW profiling but does not hurt so we add it always
export PYTHONPATH=${PATCH_PATH}:${PYTHONPATH}

# Fixed varaibles, not inherited from launcher
export TF_ALLOW_CONTROL_EDGES_IN_HABANA_OPS=1
export EXPERIMENTAL_PRELOADING=1
export ENABLE_TENSORBOARD=false
export REPORT_ACCURACY_METRICS=true
export DIST_EVAL=true
export ENABLE_DEVICE_WARMUP=true
export TF_DISABLE_MKL=1
export SYNTHETIC_DATA=${SYNTHETIC_DATA}

if [[ $MODELING -eq 1 ]]; then
    ENABLE_CHECKPOINT=true
else
    ENABLE_CHECKPOINT=false
fi
if [[ $TF_BF16_CONVERSION -eq 1 ]]; then
    DATA_TYPE="bf16"
else
    DATA_TYPE="fp32"
fi
if [[  ${NO_EVAL} -eq 1 ]]; then
    SKIP_EVAL=true
else
    SKIP_EVAL=false
fi
if [[ ${USE_LARS_OPTIMIZER} -eq 1 ]]; then
	OPTIMIZER="LARS"
else
	OPTIMIZER="SGD"
fi
if [[ ${USE_HOROVOD} -eq 1 ]]; then
	DIST_EVAL=true
	USE_HOROVOD='--use_horovod'
else
	DIST_EVAL=false
	USE_HOROVOD=''
fi
if [[ ${SYNTHETIC_DATA} -eq 1 ]]; then
	SYNTHETIC_DATA=true
fi
if [[ -n ${NUM_ACCUMULATION_STEPS} ]]; then
	NUM_ACCUMULATION_STEPS="--num_acc_steps=${NUM_ACCUMULATION_STEPS}"
else
	NUM_ACCUMULATION_STEPS=""
fi

if [[ -n ${JPEG_IMAGENET_DIR} ]]; then
    JPEG_IMAGENET_DIR="--jpeg_data_dir=${JPEG_IMAGENET_DIR}"
fi

if [[ $SIGNALING_FROM_GRAPH -eq 1 ]]; then
    export HOROVOD_FUSION_THRESHOLD=0
    export TF_USE_SIGNALING_FROM_ENCAP_OP=1
else
    export TF_USE_SIGNALING_FROM_ENCAP_OP=0
fi

# clear cache
PROC_FS=${PROC_FS:-"/proc"}
sync && echo 3 > $PROC_FS/sys/vm/drop_caches

TRAIN_COMMAND="python3 ${TRAIN_SCRIPT}
    --model_dir=${WORK_DIR}
    --data_dir=${IMAGENET_DIR}
    ${JPEG_IMAGENET_DIR}
    --batch_size=${BATCH_SIZE}
    --distribution_strategy=off
    --num_gpus=0
    --data_format=channels_last
    --train_epochs=${TRAIN_EPOCHS}
    --train_steps=${TRAIN_STEPS}
    --experimental_preloading=${EXPERIMENTAL_PRELOADING}
    --log_steps=${DISPLAY_STEPS}
    --steps_per_loop=${STEPS_PER_LOOP}
    --enable_checkpoint_and_export=${ENABLE_CHECKPOINT}
    --enable_tensorboard=${ENABLE_TENSORBOARD}
    --epochs_between_evals=${EPOCHS_BETWEEN_EVALS}
    --base_learning_rate=${BASE_LEARNING_RATE}
    --warmup_epochs=${WARMUP_EPOCHS}
    --optimizer=${OPTIMIZER}
    --lr_schedule=polynomial
    --label_smoothing=${LABEL_SMOOTH}
    --weight_decay=${WEIGHT_DECAY}
    $NUM_ACCUMULATION_STEPS
    --single_l2_loss_op
    ${USE_HOROVOD}
    --modeling=${MODELING}
    --data_loader_image_type=${DATA_TYPE}
    --dtype=${DATA_TYPE}
    --eval_offset_epochs=${EVAL_OFFSET_EPOCHS}
    --report_accuracy_metrics=${REPORT_ACCURACY_METRICS}
    --dist_eval=${DIST_EVAL}
    --target_accuracy=${STOP_THRESHOLD}
    --enable_device_warmup=${ENABLE_DEVICE_WARMUP}
    --lars_decay_epochs=${LARS_DECAY_EPOCHS}
    --momentum=${LR_MOMENTUM}
    --skip_eval=${SKIP_EVAL}
    --use_synthetic_data=${SYNTHETIC_DATA}
    --dataset_cache=${DATASET_CACHE}
    --num_train_files=${NUM_TRAIN_FILES}
    --num_eval_files=${NUM_EVAL_FILES}
"
echo ${TRAIN_COMMAND}

echo "[run] General Settings:"
echo "[run] RESNET_SIZE" $RESNET_SIZE
echo "[run] IMAGENET_DIR" $IMAGENET_DIR
echo "[run] BATCH_SIZE"  $BATCH_SIZE
echo "[run] NUM_WORKERS" $NUM_WORKERS
echo "[run] TRAIN_EPOCHS" $TRAIN_EPOCHS
echo "[run] TRAIN_STEPS" $TRAIN_STEPS
echo "[run] DISPLAY_STEPS" $DISPLAY_STEPS
echo "[run] USE_LARS_OPTIMIZER" $USE_LARS_OPTIMIZER
echo "[run] CPU_BIND_TYPE" $CPU_BIND_TYPE
echo "[run] EPOCHS_BETWEEN_EVALS" $EPOCHS_BETWEEN_EVALS
echo "[run] TRAIN_AND_EVAL" $TRAIN_AND_EVAL
echo "[run] TF_BF16_CONVERSION" $TF_BF16_CONVERSION
echo "[run] DATASET_CACHE" $DATASET_CACHE
echo "[run] USE_HOROVOD" $USE_HOROVOD
echo
echo "[run] Learning Setting:"
echo "[run] WEIGHT_DECAY" $WEIGHT_DECAY
echo "[run] NUM_ACCUMULATION_STEPS" $NUM_ACCUMULATION_STEPS
echo "[run] LABEL_SMOOTH" $LABEL_SMOOTH
echo "[run] BASE_LEARNING_RATE" $BASE_LEARNING_RATE
echo "[run] WARMUP_EPOCHS" $WARMUP_EPOCHS
echo "[run] USE_MLPERF" $USE_MLPERF
echo "[run] NO_EVAL" $NO_EVAL
echo "[run] STOP_THRESHOLD" $STOP_THRESHOLD
echo "[run] LR_MOMENTUM" $LR_MOMENTUM
echo "[run] EVAL_OFFSET_EPOCHS" $EVAL_OFFSET_EPOCHS
echo "[run] LARS_DECAY_EPOCHS" $LARS_DECAY_EPOCHS
echo "[run] SYNTHETIC_DATA" $SYNTHETIC_DATA

if [[ ! -z $USE_HOROVOD ]] && [[ $CPU_BIND_TYPE == "numa" ]]; then
	LOCAL_SNC_VALUE=$(( OMPI_COMM_WORLD_LOCAL_RANK ))
	if [[ $HLS_TYPE == "HLS2" ]]; then
		export NUMA_MAPPING_DIR=$BASE_PATH
		bash list_affinity_topology_bare_metal.sh
		CPU_RANGE=`cat $NUMA_MAPPING_DIR/.habana_moduleID$LOCAL_SNC_VALUE`
	fi
	LD_PRELOAD=${PRELOAD_PATH} numactl --physcpubind=${CPU_RANGE} ${TRAIN_COMMAND}
else
	LD_PRELOAD=${PRELOAD_PATH} ${TRAIN_COMMAND}
fi