Skip to content

ONE Release 1.21.0
Compare
Choose a tag to compare
@chunseoklee chunseoklee released this 07 Sep 04:35
· 1 commit to release/1.21.0 since this release
1.21.0
3d20228
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

Release Note 1.21.0

ONE Compiler

  • Support unrolling of LSTM and RNN Ops in one-import-onnx tool
  • Introduced new tools one-infer, circle-operator, circle-interpreter
  • Introduced Workflow(WIP) in one-cmds
  • New option quant_config in one-quantize
  • New option fake_quantize in one-quantize
  • More Ops supported: Densify
  • More Ops for quantization: ReduceMax
  • More Ops for mixed-precision quantization (MPQ): LeakyRelu, Neg, Relu6, Squeeze
  • More Ops for convert_nchw_to_nhwc option: LogSoftmax, ReduceMax, SplitV, Softmax
  • New optimization options in one-optimize: replace_non_const_fc_with_bmm, resolve_customop_splitv, fold_densify
  • Improved reshape elimination in convert_nchw_to_nhwc option.
  • Support fusion of Channel-wise Add + Relu with TConv
  • Support negative axis in ArgMin/Max
  • Show errors for unrecognized options in one-optimize
  • Fix shape inference for StridedSlice
  • Fix FuseBatchNormWithTConvPass to support TConv with bias
  • Deprecate --O1 option in circle2circle
  • Support gcc-11
  • Support limited Float16 for kernels constants with dequantization to Float32

ONE Runtime

Basic Multimodel nnpackage

  • Runtime supports to run nnpackage with two models

Channel Wise Quantization on Conv2D and Depthwise Conv2D

  • Conv2D and Depthwise Conv2D supports per-channel quantization of uint8 type.

Batch Execution with TRIX backend

  • TRIX backend supports batch execution which run in parallel with multicore
Assets 12
Loading