master 0.65
ml server 2020-01-20 19:09:22 -07:00
parent b7bdc2521e
commit ecbc5fe3e8
3 changed files with 7 additions and 24 deletions

View File

@ -134,7 +134,10 @@ firefox https://github.com/bazelbuild/bazel/releases
# Install Tensorflow
git clone tensorflow...
cd tensorflow
git checkout remotes/origin/r2.1
git checkout v2.1.0
bazel clean
# Get flags to pass:
grep flags -m1 /proc/cpuinfo | cut -d ":" -f 2 | tr '[:upper:]' '[:lower:]' | { read FLAGS; OPT="-march=native"; for flag in $FLAGS; do case "$flag" in "sse4_1" | "sse4_2" | "ssse3" | "fma" | "cx16" | "popcnt" | "avx" | "avx2") OPT+=" -m$flag";; esac; done; MODOPT=${OPT//_/\.}; echo "$MODOPT"; }
./configure
# Run Bazel to build pip package. Takes nearly 2 hours to build.
bazel build --config=opt //tensorflow/tools/pip_package:build_pip_package

View File

@ -18,8 +18,8 @@
HOSTNUM=`hostname | sed -e 's/ml//g'`
#export TF_CONFIG='{"cluster": {"worker": [ "ml0-int:2222", "ml1-int:2222", "ml2-int:2222", "ml3-int:2222", "ml4-int:2222", "ml5-int:2222"]}, "task": {"index": '$HOSTNUM', "type": "worker"}}'
#export TF_CONFIG='{"cluster": {"worker": [ "ml1-int:2222", "ml2-int:2222", "ml3-int:2222", "ml4-int:2222", "ml5-int:2222"]}}'
export TF_CONFIG='{"cluster": {"chief": [ "ml0-int:2222" ], "worker": [ "ml1-int:2222", "ml2-int:2222", "ml3-int:2222", "ml4-int:2222", "ml5-int:2222"]}, "task": {"index": '$HOSTNUM', "type": "worker"}}'
export TF_CONFIG='{"cluster": {"worker": [ "ml1-int:2222", "ml2-int:2222", "ml3-int:2222", "ml4-int:2222", "ml5-int:2222"]}}'
#export TF_CONFIG='{"cluster": {"chief": [ "ml0-int:2222" ], "worker": [ "ml1-int:2222", "ml2-int:2222", "ml3-int:2222", "ml4-int:2222", "ml5-int:2222"]}, "task": {"index": '$HOSTNUM', "type": "worker"}}'
echo $TF_CONFIG
python3 wut-worker-mas.py

View File

@ -26,11 +26,9 @@ from tensorflow.python.keras.preprocessing import image
from tensorflow.python.keras.preprocessing.image import img_to_array
from tensorflow.python.keras.preprocessing.image import ImageDataGenerator
from tensorflow.python.keras.preprocessing.image import load_img
#import tensorflow.python.distribute.cluster_resolver
#from tensorflow.python.distribute.cluster_resolver import TFConfigClusterResolver
#from tensorflow.python.distribute.cluster_resolver.TFConfigClusterResolver
tf.keras.backend.clear_session()
tf.config.optimizer.set_jit(True)
options = tf.data.Options()
os.environ["TF_CONFIG"] = json.dumps({
"cluster": {
@ -39,27 +37,16 @@ os.environ["TF_CONFIG"] = json.dumps({
},
"task": {"type": "chief", "index": 0 },
})
#os.environ["TF_CONFIG"] = json.dumps({
# "cluster": {
# "worker": [ "ml1-int:2222", "ml2-int:2222", "ml3-int:2222", "ml4-int:2222", "ml5-int:2222" ]
# }#,
# #"task": {"type": "worker", "index": 0 },
#})
print("Tensorflow Version: ", tf.__version__)
print("Num GPUs Available: ", len(tf.config.experimental.list_physical_devices('GPU')))
print("Num CPUs Available: ", len(tf.config.experimental.list_physical_devices('CPU')))
#with tf.device("GPU:0"):
# tf.ones(()) # Make sure we can run on GPU
# This ensures that XLA and ptxas work well together, and helps with scaling.
print("XLA_FLAGS='{}'".format(os.getenv("XLA_FLAGS")))
IMG_HEIGHT = 416
IMG_WIDTH= 804
batch_size = 32
epochs = 4
BUFFER_SIZE = 10000
NUM_WORKERS = 6
GLOBAL_BATCH_SIZE = 64 * NUM_WORKERS
@ -68,12 +55,9 @@ GLOBAL_BATCH_SIZE = 64 * NUM_WORKERS
POSITIVE_DIRECTORY = '/home/jebba/devel/spacecruft/satnogs-wut/data/pos'
pos_dir = '/home/jebba/devel/spacecruft/satnogs-wut/data/posdir'
from tensorflow.python.distribute.cluster_resolver import SimpleClusterResolver
strategy = tf.distribute.experimental.MultiWorkerMirroredStrategy(
tf.distribute.experimental.CollectiveCommunication.RING)
def get_bytes_and_label(filepath):
raw_bytes = tf.io.read_file(filepath)
label = tf.strings.regex_full_match(
@ -117,13 +101,9 @@ dataset = dataset.map(process_image, num_parallel_calls=AUTOTUNE)
dataset = dataset.batch(batch_size=32)
dataset = dataset.prefetch(buffer_size=AUTOTUNE)
os.makedirs(profile_dir, exist_ok=True)
# tf.data.Dataset.from_generator
tf.config.optimizer.set_jit(True)
#tf.summary.trace_on(profiler=True)