zhreshold

python mxnet/example/image-classification/train_imagenet.py --network shufflenet --data-train ~/efs/users/joshuazz/data/imagenet/record/train_480_q95.rec --data-val ~/efs/users/joshuazz/data/imagenet/record/val_256_q90.rec --batch-size 512 --gpus 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 --num-epochs 150 --lr-step-epochs 30,60,90 --min-random-scale 0.533 --lr 0.01 --disp-batches 100 --top-k 5 --data-nthreads 32 --random-mirror 1 --max-random-shear-ratio 0 --max-random-rotate-angle 0 --max-random-h 0 --max-random-l 0 --max-random-s 0 --model-prefix model/shufflenet  | tee ~/efs/users/joshuazz/temp/train_imagenet_logs/shufflenet.log

zhreshold

name: "shufflenet"
#  transform_param {
#    scale: 0.017
#    mirror: false
#    crop_size: 224
#    mean_value: [103.94,116.78,123.68]
#  }
input: "data"
input_shape { 
  dim: 1

zhreshold

{
  "nvidia-titan-x": {
    "devices": [
      {
        "cores": "3072",
        "memory": "12GB",
        "memory_bandwith": "336.5GB/s",
        "name": "Nvidia Titan X",
        "quantity": 1
      }

zhreshold

(function() {

// Chart design based on the recommendations of Stephen Few. Implementation
// based on the work of Clint Ivy, Jamie Love, and Jason Davies.
// http://projects.instantcognition.com/protovis/bulletchart/
d3.bullet = function() {
  var orient = "left", // TODO top & bottom
      reverse = false,
      duration = 0,
      ranges = bulletRanges,

zhreshold

python train_imagenet.py --data-train ~/data/train.rec --data-val ~/data/val.rec --gpus 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 --data-nthreads 32 --network resnet101_v2 --batch-size 256 --top-k 5 --model-prefix model/resnet101_v2 --min-random-scale 0.533 --max-random-shear-ratio 0 --max-random-rotate-angle 0 --max-random-h 0 --max-random-l 0 --max-random-s 0 --lr-step-epochs 30,60,90 --num-epochs 120 --rgb-std '58.395,57.12,57.375'

zhreshold

import mxnet as mx
import numpy as np
from timeit import default_timer as timer

def get_bench_net(num_hidden=10000):
  data = mx.sym.var('data')
  fc = mx.sym.FullyConnected(data, num_hidden=num_hidden)
  return fc

num_out = 10000

zhreshold

from mxnet.gluon import HybridBlock

class Concat(HybridBlock):
  """Concat operation for multiple inputs."""
  def __init__(self, dim=1, **kwargs):
    super(Concat, self).__init__(**kwargs)
    self._kwargs = {'dim': dim}
    
  def hybrid_forward(self, F, *args):
    return F.concat(*args, name='fwd', **self._kwargs)

zhreshold

This is a demo onnx model for super resolution.

zhreshold

{0: 'tench, Tinca tinca',
 1: 'goldfish, Carassius auratus',
 2: 'great white shark, white shark, man-eater, man-eating shark, Carcharodon carcharias',
 3: 'tiger shark, Galeocerdo cuvieri',
 4: 'hammerhead, hammerhead shark',
 5: 'electric ray, crampfish, numbfish, torpedo',
 6: 'stingray',
 7: 'cock',
 8: 'hen',
 9: 'ostrich, Struthio camelus',

zhreshold

name: "GoogleNet"
layer {
  name: "data"
  type: "Input"
  top: "data"
  input_param { shape: { dim: 10 dim: 3 dim: 224 dim: 224 } }
}
layer {
  name: "conv1/7x7_s2"
  type: "Convolution"