From 5ae49021983dd3da7a267a7e0105297e58ffb179 Mon Sep 17 00:00:00 2001 From: Ruoxin Sang Date: Tue, 19 Mar 2019 10:14:57 -0700 Subject: [PATCH] Document explicitly `iterations_per_loop` is not respected in TPUEstimator evaluation. PiperOrigin-RevId: 239215958 --- tensorflow/python/tpu/tpu_config.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/tensorflow/python/tpu/tpu_config.py b/tensorflow/python/tpu/tpu_config.py index 20946d7aa66..6ae65acb6cf 100644 --- a/tensorflow/python/tpu/tpu_config.py +++ b/tensorflow/python/tpu/tpu_config.py @@ -62,6 +62,8 @@ class TPUConfig( system before returning to CPU host for each `Session.run`. This means global step is increased `iterations_per_loop` times in one `Session.run`. It is recommended to be set as number of global steps for next checkpoint. + Note that in evaluation don't use this value, instead we run total eval + `steps` on TPU for a single `Session.run`. num_shards: (Deprecated, ignored by TPUEstimator). The number of model replicas in the system. For non-model-parallelism case, this number equals the total number of TPU cores. For