1 one (1.22.0) bionic; urgency=medium
3 * Introduce new optimization options: `unroll_unidirseqlstm`, `forward_transpose_op`, `fold_fully_connected`, `fuse_prelu`
4 * Support more Ops for fake quantization: `Depth2Space`, `Space2Depth`, `Pack`, `Unpack`, `Abs`
5 * Support more Ops for quantization: `Abs`, `ReduceProd`
6 * Introduce _visq_ tool for quantization error visualization
7 * Introduce _Environment_ section into configuration file
8 * Improve speed of `convert_nchw_to_nhwc` option
9 * Support `Add`, `Mul` of index-type (int32, int64) tensors in _one-quantize_
10 * Support ubuntu 20.04
12 -- seongwoo <mhs4670go@naver.com> Fri, 24 Mar 2023 13:58:16 +0900
14 one (1.21.0) bionic; urgency=medium
16 * Support unrolling of LSTM and RNN Ops in `one-import-onnx` tool
17 * Introduced new tools `one-infer`, `circle-operator`, `circle-interpreter`
18 * Introduced `Workflow`(WIP) in `one-cmds`
19 * New option `quant_config` in `one-quantize`
20 * New option `fake_quantize` in `one-quantize`
21 * More Ops supported: Densify
22 * More Ops for quantization: ReduceMax
23 * More Ops for mixed-precision quantization (MPQ): LeakyRelu, Neg, Relu6, Squeeze
24 * More Ops for `convert_nchw_to_nhwc` option: LogSoftmax, ReduceMax, SplitV, Softmax
25 * New optimization options in `one-optimize`: `replace_non_const_fc_with_bmm`, `resolve_customop_splitv`, `fold_densify`
26 * Improved reshape elimination in `convert_nchw_to_nhwc` option.
27 * Support fusion of Channel-wise Add + Relu with TConv
28 * Support negative axis in ArgMin/Max
29 * Show errors for unrecognized options in `one-optimize`
30 * Fix shape inference for `StridedSlice`
31 * Fix FuseBatchNormWithTConvPass to support TConv with bias
32 * Deprecate `--O1` option in `circle2circle`
34 * Support limited Float16 for kernels constants with dequantization to Float32
36 -- seongwoo <mhs4670go@naver.com> Wed, 06 Sep 2022 12:00:00 +0900
38 one (1.20.0) bionic; urgency=medium
40 * luci-interpreter supports multiple kernels with PAL layer including Cortext-M
41 * luci-interpreter supports integer tensor for partly kernels
42 * luci import support constant without coping to reduce memory for luci-interpreter
43 * Reduce duplicate codes to package released modules
44 * Limited support for ONNX LSTM/RNN unrolling while importing
45 * Limited support for ARM32 cross build
46 * Support new operator: SVDF
47 * New virtual CircleVariable to support tensor with variable
48 * Support quantization of BatchMatMul Op
49 * Support mixed(UINT8 + INT16) quantization
50 * Support backward propagation of quantization parameters
51 * Upgrade default python to version 3.8
52 * Support TensorFlow 2.8.0, ONNX-TF 1.10.0, ONNX 1.11.0
53 * Upgrade circle schema to follow tflite schema v3b
54 * Refactor to mio-tflite280, mio-circle04 with version and helpers methods
55 * Use one flatbuffers 2.0 version
56 * Drop support for TensorFlow 1.x
57 * Fix for several bugs, performance enhancements, and typos
59 -- seongwoo <mhs4670go@naver.com> Tue, 26 Apr 2022 12:00:00 +0900
61 one (1.19.0) bionic; urgency=medium
63 * `circle-quantizer` supports input/output type option
64 * Introduce configuration file for optimization options
66 -- seongwoo <mhs4670go@naver.com> Wed, 10 Nov 2021 15:53:39 +0900
68 one (1.18.0) bionic; urgency=medium
70 * More optimization pass
72 -- seongwoo <mhs4670go@naver.com> Fri, 15 Oct 2021 15:23:20 +0900
74 one (1.17.0) bionic; urgency=medium
76 * More optimization pass
77 * Add new InstanceNorm pattern in `FuseInstanceNormPass`
79 * Introduce `onecc` driver to `one-cmds`
80 * Introduce `one-profile` driver to `one-cmds`
82 -- seongwoo <mhs4670go@naver.com> Fri, 20 Aug 2021 17:50:20 +0900
84 one (1.16.1) bionic; urgency=medium
86 * Extends the point where `one-codegen` finds backends.
88 -- seongwoo chae <mhs4670go@naver.com> Wed, 26 May 2021 18:06:53 +0900
90 one (1.16.0) bionic; urgency=low
94 -- seongwoo chae <mhs4670go@naver.com> Mon, 26 Apr 2021 14:34:57 +0900