Aditya Agrawal adityaiitb

Save a saved_model

#!/usr/bin/env python3

import tensorflow as tf
import numpy as np

Download the CUDA toolkit i.e. runfile(local) from Nvidia.
Download the cuDNN .tgz file from Nvidia. You can use this hack from Nvidia Developer Forum to wget a cuDNN file.
Install CUDA toolkit (includes cuBLAS).

export CUDA_ROOT=/where/to/install/cuda
bash cuda_<version>.run --silent \
  --toolkit --toolkitpath=$CUDA_ROOT \
  --samples --samplespath=$CUDA_ROOT/samples 
  --tmpdir=$HOME/tmp

The .npz file format is used to save several numpy arrays into a single zipped file. Documentation.

import numpy as np

a = np.array([42])
b = np.array([2, 3, 5, 7, 9]) 

with open("foo.npz", "wb") as f:

Saved Model CLI

Saved Model CLI (show)

saved_model_cli show --dir resnet/1

saved_model_cli show --dir resnet/1 --tag_set serve

A simple example of how to use PyTorch ATen directly using C++.

ps2epsi foo.ps
#ps2pdf14 -dEPSCrop -dAutoRotatePages=/None -dPDFSETTINGS=/prepress -dEmbedAllFonts=true foo.epsi foo.pdf
ps2pdf14 -dEPSCrop -dAutoRotatePages=/All -dPDFSETTINGS=/prepress -dEmbedAllFonts=true foo.epsi foo.pdf

Using `pdftk`

touch out.pdf
pdftk foo.pdf cat 1E output out.pdf

Using `pdf270`

pdf270 --suffix 'turned' foo.pdf
mv foo-turned.pdf foo.pdf

DEB files are ar archives, which contain three files:

debian-binary
control.tar.gz
data.tar.gz

Unpacking this file is a two step process.

ar vx mypackage.deb
tar -xzvf data.tar.gz

	# Tensor names, shape and dtype of each shard
	python -m tensorflow.python.tools.inspect_checkpoint --file_name path/variables/variables

	# Tensor names, shape and dtype
	python -m tensorflow.python.tools.inspect_checkpoint --file_name path/variables/variables --all_tensor_names

	# Tensor names, shapes, dtype and value
	python -m tensorflow.python.tools.inspect_checkpoint --file_name path/variables/variables --all_tensors

	#!/usr/bin/env python

	# python train.py -train_steps 1 -data data/demo -gpu_ranks 0

	import subprocess
	from itertools import product

	src_emb = [128, 1024]
	dst_emb = [128, 1024]
	hidden = [128, 1024]

Aditya Agrawal adityaiitb

Save a saved_model

Saved Model CLI (show)

Using pdftk

Using pdf270

Using `pdftk`

Using `pdf270`