modify resent to run many loops quickly

Taylor Robie · Taylor Robie · commit 49d7c0ea7245 · 2018-09-24T11:40:07.000-07:00
diff --git a/official/resnet/resnet_run_loop.py b/official/resnet/resnet_run_loop.py
@@ -25,6 +25,7 @@
 
 import math
 import os
+import sys
 
 # pylint: disable=g-bad-import-order
 from absl import flags
@@ -389,6 +390,12 @@ def resnet_main(
       This is only used if flags_obj.export_dir is passed.
   """
 
+  print("\n" * 3)
+  tf.logging.info("TF version: {}".format(tf.VERSION))
+  tf.logging.info("TF git version: {}".format(tf.GIT_VERSION))
+  print("\n" * 3)
+  sys.stdout.flush()
+
   model_helpers.apply_clean(flags.FLAGS)
 
   # Using the Winograd non-fused algorithms provides a small performance boost.
@@ -487,7 +494,9 @@ def input_fn_eval():
 
     if num_train_epochs:
       classifier.train(input_fn=lambda: input_fn_train(num_train_epochs),
-                       hooks=train_hooks, max_steps=flags_obj.max_train_steps)
+                       hooks=train_hooks, steps=5)
+
+    continue  # We don't need eval for this repro.
 
     tf.logging.info('Starting to evaluate.')
 
diff --git a/official/resnet/run_repro.sh b/official/resnet/run_repro.sh
@@ -0,0 +1,12 @@
+#!/usr/bin/env bash
+set -e
+
+pushd ../..
+export PYTHONPATH=$(pwd)
+popd
+
+TIME=$(date +%s)
+
+python3 imagenet_main.py --clean --data_dir /imn/imagenet/combined/ \
+  --model_dir /tmp/garden_imagenet --train_epochs 30 --batch_size 2048 --resnet_version 1 \
+  --resnet_size 50 --dtype fp16 --num_gpus 8 |& tee imagenet_garden_${TIME}.log
