Skip to content
Snippets Groups Projects
Commit 7b0ad6f0 authored by willzhang4a58's avatar willzhang4a58
Browse files

alexnet report

parent 7e57b20f
No related branches found
No related tags found
No related merge requests found
net: "./train.net"
resource: "./1_machine_4_gpu.resource"
placement: "./1_machine_1_gpu.placement"
other: "./piece_size_512.other"
placement_group {
op_set {
op_name: "decode"
op_name: "transpose"
}
parallel_conf {
policy: kDataParallel
device_name: "first:cpu:32"
}
}
placement_group {
op_set {
op_name: "conv1"
op_name: "pool1"
op_name: "conv2"
op_name: "pool2"
op_name: "conv3"
op_name: "conv4"
op_name: "conv5"
op_name: "pool5"
op_name: "dropout1"
op_name: "dropout2"
op_name: "softmax_loss"
}
parallel_conf {
policy: kDataParallel
device_name: "first:gpu:0"
}
}
placement_group {
op_set {
op_name: "fc1"
op_name: "fc2"
op_name: "fc3"
}
parallel_conf {
policy: kDataParallel
device_name: "first:gpu:0"
}
}
net: "./train.net"
resource: "./1_machine_4_gpu.resource"
placement: "./1_machine_2_gpu.placement"
other: "./piece_size_1024.other"
......@@ -5,17 +5,15 @@ placement_group {
}
parallel_conf {
policy: kDataParallel
device_name: "first:cpu:30"
device_name: "first:cpu:32"
}
}
placement_group {
op_set {
op_name: "conv1"
# op_name: "lrn1"
op_name: "pool1"
op_name: "conv2"
# op_name: "lrn2"
op_name: "pool2"
op_name: "conv3"
op_name: "conv4"
......
......@@ -4,7 +4,4 @@ machine {
name: "first"
}
cpu_device_num: 48
gpu_device_num: 2
persistence_worker_num: 48
comm_net_worker_num: 4
gpu_device_num: 4
use_rdma: false
use_rdma: true
globalfs_conf {
localfs_conf {
}
}
data_part_num: 30
piece_size: 480
model_load_snapshot_path: ""
max_data_id_length: 0
data_part_num: 32
piece_size: 1024
train_conf {
model_save_snapshots_path: "/tmp/snapshot"
num_of_batches_in_snapshot: 3000
staleness: 0
total_batch_num: 30000
default_initializer_conf {
random_normal_conf {
......@@ -18,7 +15,7 @@ train_conf {
std: 0.01
}
}
batch_size: 480
batch_size: 1024
l2: 0.0005
model_update_conf {
learning_rate: 0.01
......
use_rdma: true
globalfs_conf {
localfs_conf {
}
}
data_part_num: 32
piece_size: 512
train_conf {
model_save_snapshots_path: "/tmp/snapshot"
num_of_batches_in_snapshot: 3000
total_batch_num: 30000
default_initializer_conf {
random_normal_conf {
mean: 0.0
std: 0.01
}
}
batch_size: 1024
l2: 0.0005
model_update_conf {
learning_rate: 0.01
naive_conf {
}
learning_rate_decay {
exponential_conf {
decay_rate: 0.1
decay_batches: 10000
staircase: true
}
}
}
}
batch_size: 1024
gpu num | time per batch
:-------| :-------------
1 | 534ms
net: "./train.net"
resource: "./train.resource"
placement: "./train.placement"
other: "./train.other"
op {
name: "decode"
decode_ofrecord_conf {
data_dir: "/dataset/imagenet/train/30"
data_dir: "/dataset/imagenet/train/32"
blob {
name: "encoded"
data_type: kFloat
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment