backend.ai-kernels

Backend.AI agent kernels in various programming languages / toolkits and frameworks.

Supporting modes

Language	Version	Batch	Query	Input Hook	TTY	ETC
C	6.3	O	O	O		GCC compiler
C++ (C+14)	6.3	O	O	O		GCC compiler
Go	1.9	O	O
Haskell	8.2	O	O
Java	8.0	O	O
Linux Console	-		O	O	O	Bash On Ubuntu
Lua	5.1		O
Lua	5.2		O
Lua	5.3		O
Node.js	6		O
Octave	4.2		O
Python	2.7	O	O	O
Python	3.6	O	O	O
Rust	1.21	O	O
PHP	7.0		O
R	3.0		O			CRAN R

Deep-Learning Framework	Version	Batch	Query	Input Hook	TTY	ETC
TensorFlow	1.4	O	O	O		Bundled w/Keras 2
TensorFlow	1.3	O	O	O		Bundled w/Keras 2
PyTorch	0.2	O	O	O
Theano	0.9	O	O	O		Bundled w/Keras 2
CNTK	(WIP)	O	O	O		Bundled w/Keras 2

Build chain

Kernels have dependencies to reduce total amount of storage, especailly for faster agent / instance setup and cost reduction. Default docker images are:

• base-alpine (smallest base image)
• base-debian (easier to compile common frameworks with libraries)
• base-ubuntu (heaviest but best compatibility)

base-alpine	base-debian	base-ubuntu
base-python-minimal-2.7
base-python-minimal-3.6	base-python-minimal-3.6	base-python-minimal-3.6
base-python-wheels-2.7
base-python-wheels-3.6
c-gcc6.3-alpine
cpp-gcc6.3-alpine
java-8-alpine
rust-1.17-alpine
git-3.6-alpine
go-1.8-alpine
go-1.9-alpine
	haskell-ghc8.2-debian
	octave-4.2-debian
r-3.3-alpine
lua-5.1-alpine
lua-5.2-alpine
lua-5.3-alpine
php-7-alpine
nodejs-6-alpine
	julia-0.6-debian
r-3.3-alpine
	bazel-0.7-debian
	cuda-8.0-cudnn-6.0
	python-caffe-1.0-py36
	python-torch-0.2-py36
	python-torch-0.2-py36-gpu
		python-cntk-2.2-py36
	tensorflow-1.4-py36-dense
	tensorflow-1.4-py36-dense-gpu
	tensorflow-1.3-py36-dense
	tensorflow-1.3-py36-dense-gpu

For the complete build chain, please refer build.py for more information.