refine example

examples/mnist/predict/job.prototxt

 dlnet_filepath: "./net.prototxt"
 resource_filepath: "./resource.prototxt"
 placement_filepath: "./placement.prototxt"
-model_load_snapshot_path: "/willzhang/snapshots/snapshot_10"
-piece_size: 1000
-max_data_id_length: 16
 global_fs_conf {
   hdfs_conf {
     namenode: "hdfs://192.168.1.11:9000"
   }
 }
+single_piece_size: 1000
+data_part_num: 6
+model_load_snapshot_path: "/willzhang/snapshots/snapshot_10"
+max_data_id_length: 16
 predict_conf {
-  total_data_num: 60000
 }

examples/mnist/predict/net.prototxt

 op {
   name: "feature"
-  data_loader_conf {
+  basic_data_loader_conf {
     out: "out"
     data_type: kFloat
     shape {
-      dim: 60000
       dim: 1
       dim: 28
       dim: 28
@@ -15,11 +14,10 @@ op {
 
 op {
   name: "label"
-  data_loader_conf {
+  basic_data_loader_conf {
     out: "out"
     data_type: kInt32
     shape {
-      dim: 60000
     }
     data_dir: "/mnist/label/"
   }
@@ -56,10 +54,10 @@ op {
 }
 
 op {
-  name: "record"
-  record_conf {
+  name: "print"
+  print_conf {
     lbn: "label/out"
     lbn: "softmax/out"
-    record_path: "/willzhang/predict_result"
+    print_path: "/willzhang/predict_result"
   }
 }

examples/mnist/predict/placement.prototxt

@@ -2,12 +2,13 @@ placement_group {
   op_set {
     op_name: "feature"
     op_name: "label"
-    op_name: "record"
+    op_name: "print"
   }
   parallel_conf {
     policy: kDataParallel
-    device_name: "192.168.1.11:persistence"
-    device_name: "192.168.1.13:persistence"
+    device_name: "192.168.1.11:persistence:2"
+    device_name: "192.168.1.12:persistence:2"
+    device_name: "192.168.1.13:persistence:2"
   }
 }
 
@@ -19,7 +20,8 @@ placement_group {
   }
   parallel_conf {
     policy: kDataParallel
-    device_name: "192.168.1.11:0-3"
-    device_name: "192.168.1.13:0-3"
+    device_name: "192.168.1.11:gpu:0-1"
+    device_name: "192.168.1.12:gpu:0-1"
+    device_name: "192.168.1.13:gpu:0-1"
   }
 }

examples/mnist/predict/resource.prototxt

 machine {
   addr: "192.168.1.11"
-  port: 6666
+  port: 6789
   name: "192.168.1.11"
 }
-
+machine {
+  addr: "192.168.1.12"
+  port: 6789
+  name: "192.168.1.12"
+}
 machine {
   addr: "192.168.1.13"
-  port: 6666
+  port: 6789
   name: "192.168.1.13"
 }
 
-device_num_per_machine: 4
-device_type: kCPU
+cpu_device_num: 20
+gpu_device_num: 2
+persistence_worker_num: 6
+boxing_worker_num: 6
+comm_net_worker_num: 4

examples/mnist/train/placement.prototxt

@@ -5,7 +5,9 @@ placement_group {
   }
   parallel_conf {
     policy: kDataParallel
-    device_name: "192.168.1.11:0-5"
+    device_name: "192.168.1.11:persistence:2"
+    device_name: "192.168.1.12:persistence:2"
+    device_name: "192.168.1.13:persistence:2"
   }
 }
 
@@ -17,6 +19,8 @@ placement_group {
   }
   parallel_conf {
     policy: kDataParallel
-    device_name: "192.168.1.11:0-3"
+    device_name: "192.168.1.11:gpu:0-1"
+    device_name: "192.168.1.12:gpu:0-1"
+    device_name: "192.168.1.13:gpu:0-1"
   }
 }

examples/mnist/train/resource.prototxt

@@ -3,9 +3,19 @@ machine {
   port: 6789
   name: "192.168.1.11"
 }
+machine {
+  addr: "192.168.1.12"
+  port: 6789
+  name: "192.168.1.12"
+}
+machine {
+  addr: "192.168.1.13"
+  port: 6789
+  name: "192.168.1.13"
+}
 
-device_num_per_machine: 4
-device_type: kCPU
-persistence_worker_num: 1
-boxing_worker_num: 1
-comm_net_worker_num: 1
+cpu_device_num: 20
+gpu_device_num: 2
+persistence_worker_num: 6
+boxing_worker_num: 6
+comm_net_worker_num: 4

scripts/run.sh

 set -e
 set -x
 
-declare -a hosts=("192.168.1.11" "192.168.1.13")
+declare -a hosts=("192.168.1.11" "192.168.1.12" "192.168.1.13")
 
 ONEFLOW_CMD='nohup ./oneflow -logtostderr=0 -log_dir=./log -v=0 -logbuflevel=-1 -job_conf_filepath=./job.prototxt'