release-notes.rst 3.99 KB
Newer Older
1
.. project/release-notes.rst:
2 3 4 5

Release Notes
#############

6 7 8
nGraph is provided as source code, APIs, build scripts, and some binary formats 
for various Compiler stack configurations and use cases. 

9 10 11
For downloads formatted as ``.zip`` and ``tar.gz``, see 
https://github.com/NervanaSystems/ngraph/releases.

12 13 14 15 16
This page includes additional documentation updates.

We are pleased to announce the release of version |version|-doc.


17 18 19
Core updates for |version|
~~~~~~~~~~~~~~~~~~~~~~~~~~~

Leona C's avatar
Leona C committed
20 21 22 23 24 25
+ More ONNX ops
+ Elementwise divide defaults to Python semantics
+ GenerateMask seed optional
+ Graph visualization improvements
+ Preserve control dependencies in more places
+ GetOutputElement has single input
26

27

Leona C's avatar
Leona C committed
28 29
.. Latest doc updates
.. ~~~~~~~~~~~~~~~~~~
30

Leona C's avatar
Leona C committed
31
.. + Add instructions how to build ``NGRAPH_PLAIDML`` backend.
32 33


34
.. important:: Pre-releases (``-rc-0.*``) have newer features, and are less stable.  
35 36


37 38 39
Changelog on Previous Releases
==============================

40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59
0.23
----

+ More ONNX ops
+ Elementwise divide defaults to Python semantics
+ GenerateMask seed optional
+ Document new debug tool
+ Graph visualization improvements
+ Note deprecation of MXNet's ``ngraph-mxnet`` PyPI
+ Note default change to `svg` files for graphs and visualization
+ Add more prominent tips for contributors who find the doc-contributor-README
+ Better GSG / Install Guide structure.
+ Added group edits and new illustrations from PR 2994 to `introduction.rst`.
+ Ensure ngraph-bridge link in README goes to right place.
+ Make project `extras` their own subdirectory with index to help organize them.
+ **Known Issues**
  
  - When using TensorFlow\* v1.14.0 with ```ngraph-bridge`` v0.16.0rc0 and CPU
    backend, we saw notable to severe decreases in throughput in many models.

60 61 62 63 64 65 66 67 68
0.22
----

+ More ONNX ops
+ Optimizations
+ Don't reseed RNG on each use
+ Initial doc and API for IntelGPU backend 
+ DynamicBackend API

69 70 71

0.21
----
72 73 74 75 76 77 78 79 80

+ The offset argument in tensor reads and writes has been removed
+ Save/load API
+ More ONNX ops
+ Better tensor creation
+ More shape support
+ Provenance improvements
+ offset arg for tensor creation is deprecated
+ static linking support
81 82 83 84 85 86 87 88 89 90 91
+ Initial test of 0.21-doc
+ Updated :doc:`doc-contributor-README` for new community-based contributions. 
+ Added instructions on how to test or display the installed nGraph version.
+ Added instructions on building nGraph bridge (ngraph-bridge).
+ Updated Backend Developer Guides and ToC structure.
+ Tested documentation build on Clear Linux OS; it works.
+ Fixed a few links and redirs affected by filename changes.
+ Some coding adjustments for options to render math symbols, so they can be 
  documented more clearly and without excessive JS (see replacements.txt).
+ Consistent filenaming on all BE indexes.
+ Removed deprecated TensorAPI.
92 93 94 95 96 97 98 99 100 101 102


0.20
----

+ Save/load API
+ More ONNX ops
+ Better tensor creation
+ More shape support
+ Provenance improvements

103 104 105 106

0.19
----

107 108 109 110 111 112 113 114 115 116 117
+ More dynamic shape preparation
+ Distributed interface factored out
+ fp16 and bfloat16 types
+ codegen execution parameterized by context
+ NodeMap, NodeVector, ParameterVector, ResultVector now vectors
  
  - ``node_vector.hpp`` replaced by ``node.hpp``
  - ``op/parameter_vector.hpp`` replaced by ``op/parameter.hpp``
  - ``op/result_vector.hpp`` replaced by ``op/result.hpp``

+ Additional ONNX ops
118 119 120 121 122
+ Add graph visualization tools to doc
+ Update doxygen to be friendlier to frontends



123 124 125
0.18
----

126 127 128 129 130 131 132 133
+ Python formatting issue
+ mkl-dnn work-around
+ Event tracing improvements
+ Gaussian error function
+ Begin tracking framework node names
+ ONNX quantization
+ More fusions

134

135 136 137
0.17
----

138 139 140 141
+ Allow negative padding in more places
+ Add code generation for some quantized ops
+ Preliminary dynamic shape support
+ initial distributed ops
142 143
+ Pad op takes CoordinateDiff instead of Shape pad values to allow for negative 
  padding.
144

145

146 147
0.16
----
L.S. Cook's avatar
L.S. Cook committed
148

149 150 151 152 153 154 155
+ NodeInput and NodeOutput classes prepare for simplifications of Node
+ Test improvements
+ Additional quantization ops
+ Performance improvements
+ Fix memory leak
+ Concat optimization
+ Doc updates