From ccbbdab894b30c7ddf135b889a7dba26b1f813ae Mon Sep 17 00:00:00 2001
From: Wu <fewu@38f9d34e1390.ant.amazon.com>
Date: Thu, 6 Jun 2019 14:14:36 -0700
Subject: [PATCH] minor change for EFA running

---
 infra/ami/train_efa.sh | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/infra/ami/train_efa.sh b/infra/ami/train_efa.sh
index 86eb342e..a3c6135c 100644
--- a/infra/ami/train_efa.sh
+++ b/infra/ami/train_efa.sh
@@ -27,18 +27,18 @@ mpirun -np ${NUM_GPU} \
 -x LD_LIBRARY_PATH \
 -x PATH \
 -x NCCL_SOCKET_IFNAME=^docker0,lo \
--x NCCL_MIN_NRINGS=8 \
+-x NCCL_MIN_NRINGS=13 \
 -x NCCL_DEBUG=INFO \
 -x TENSORPACK_FP16=1 \
 -x HOROVOD_CYCLE_TIME=0.5 \
 -x HOROVOD_FUSION_THRESHOLD=67108864 \
-python3 tensorpack-mask-rcnn/MaskRCNN/train.py \
+python3 /home/ec2-user/tensorpack-mask-rcnn/MaskRCNN/train.py \
 --fp16 \
 --throughput_log_freq ${THROUGHPUT_LOG_FREQ} \
 --config \
 MODE_MASK=True \
 MODE_FPN=True \
-DATA.BASEDIR=/data \
+DATA.BASEDIR=/home/ec2-user/data \
 DATA.TRAIN='["train2017"]' \
 DATA.VAL='("val2017",)' \
 TRAIN.BATCH_SIZE_PER_GPU=${BATCH_SIZE_PER_GPU} \
@@ -46,7 +46,7 @@ TRAIN.LR_EPOCH_SCHEDULE='[(8, 0.1), (10, 0.01), (12, None)]' \
 TRAIN.EVAL_PERIOD=12 \
 RPN.TOPK_PER_IMAGE=True \
 PREPROC.PREDEFINED_PADDING=True \
-BACKBONE.WEIGHTS=/data/pretrained-models/ImageNet-R50-AlignPadding.npz \
+BACKBONE.WEIGHTS=/home/ec2-user/data/pretrained-models/ImageNet-R50-AlignPadding.npz \
 BACKBONE.NORM=FreezeBN \
 TRAINER=horovod
 #For 32x4