Compare commits

...
2 changed files with 1643 additions and 1 deletions
@@ -767,6 +767,13 @@
" strategy = tf.distribute.MultiWorkerMirroredStrategy()\n",
" logging.info(\"Multi-worker Strategy distributed training\")\n",
" logging.info('TF_CONFIG = {}'.format(os.environ.get('TF_CONFIG', 'Not found')))\n",
" # Single Machine, multiple TPU devices\n",
"elif args.distribute == 'tpu':\n",
" cluster_resolver = tf.distribute.cluster_resolver.TPUClusterResolver(tpu=\"local\")\n",
" tf.config.experimental_connect_to_cluster(cluster_resolver)\n",
" tf.tpu.experimental.initialize_tpu_system(cluster_resolver)\n",
" strategy = tf.distribute.TPUStrategy(cluster_resolver)\n",
" print(\"All devices: \", tf.config.list_logical_devices('TPU'))\n",
"\n",
"logging.info('num_replicas_in_sync = {}'.format(strategy.num_replicas_in_sync))\n",
"\n",
@@ -825,8 +832,11 @@
" else:\n",
" task_type, task_id = None, None\n",
"\n",
" if args.distribute==\"tpu\":\n",
" save_locally = tf.saved_model.SaveOptions(experimental_io_device='/job:localhost')\n",
" model.save(args.model_dir, options=save_locally)\n",
" # single, mirrored or primary for multiworker\n",
" if _is_chief(task_type, task_id):\n",
" elif _is_chief(task_type, task_id):\n",
" model.save(args.model_dir)\n",
" # non-primary workers for multi-workers\n",
" else:\n",
File diff suppressed because it is too large Load Diff