Skip to content
Pre-release
Pre-release

@dzakhar dzakhar released this Sep 16, 2021

Release Candidate 2 for Version 2.0

Release Notes

  1. Version 2.0

  2. This release supports following functional primitives

    • 2D Convolution
    • 2D Depthwise Convolution
    • 2D Transpose Convolution
    • 2D Group Convolution
    • Fully Connected layer
    • Max and average pooling
    • LSTM and GRU recurrent cells
    • RNN Dense layer
    • Elementwise (add, sub, mul, min, max)
    • Permute
    • Argmax
    • Data manipulation (concatenation, permute, 2D padding)
    • ReLU, Leaky ReLU, Parametric ReLU, ReLU1, ReLU6
    • Softmax, Sigmoid, TanH, L2 Normalization
    • Helper functions to copy (partial) tensors (mli_mov*)
  3. Supported data layout:

    • Data layout HWC (Height-Width-Channel)
  4. Supported data format:

    • Fixed point 8bit and 16bit (fx8 and fx16)
    • Signed asymmetric 8bit quantization (sa8)
    • Signed asymmetric datatype supports per-tensor or per channel quantization with 16bit scale factors.
  5. Slicing support: creation of sub-tensors and support for non-contiguous tensor data.

  6. Supported platforms:

    • VPX
    • x86 emulation
  7. Toolchains support:

Fixes and Improvements in the Release Candidate 2 Сompared to Release Candidate 1

  1. Add missing fields in some library interface enumerations.
  2. Remove -Hon=Long_enums compilation option from build scripts.
  3. Optimize code size of convert SAFX version.
  4. Improve accuracy for softmax kernel.
  5. Face Detect example uses NN model instead of hard-coded data for detecting faces on the input image.
  6. EMNIST Tutorial: clarify instructions and fix build process. Update requirements on dependent python modules.
  7. User Tests application: Fix assertion on scalar tensor for debug mode.
  8. CIFAR-10 Example: Fix fields filling to align with operands requirements from API side.
  9. Kernels input parameters checking for debug mode is improved and is better aligned with API specification.
  10. Documentation: Update sections hierarchy and clarify details on some aspects. Add more info in examples and master readme files.

Known Limitations

  1. embARC MLI 2.0 is partially optimized for ARC EMxD and ARC HSxD targets. Currently we recommend only building for VPX and x86 emulation targets. You can use MLI 1.1 for EM/HS targets.
Assets 3
Sep 13, 2021
snapshot 2021 ww37
Pre-release
Pre-release

@dzakhar dzakhar released this Aug 20, 2021

Release Candidate 1 for Version 2.0

Release Notes

  1. Version 2.0

  2. This release supports following functional primitives

    • 2D Convolution
    • 2D Depthwise Convolution
    • 2D Transpose Convolution
    • 2D Group Convolution
    • Fully Connected layer
    • Max and average pooling
    • LSTM and GRU recurrent cells
    • RNN Dense layer
    • Elementwise (add, sub, mul, min, max)
    • Permute
    • Argmax
    • Data manipulation (concatenation, permute, 2D padding)
    • ReLU, Leaky ReLU, Parametric ReLU, ReLU1, ReLU6
    • Softmax, Sigmoid, TanH, L2 Normalization
    • Helper functions to copy (partial) tensors (mli_mov*)
  3. Supported data layout:

    • Data layout HWC (Height-Width-Channel)
  4. Supported data format:

    • Fixed point 8bit and 16bit (fx8 and fx16)
    • Signed asymmetric 8bit quantization (sa8)
    • Signed asymmetric datatype supports per-tensor or per channel quantization with 16bit scale factors.
  5. Slicing support: creation of sub-tenors and support for non-contiguous tensor data.

  6. Supported platforms:

    • VPX
    • x86 emulation
  7. Toolchains support:

Known Limitations

  1. Face Detect example hard-codes some data, which means that it only works with supplied test image
Assets 3
Aug 3, 2021
snapshot 2021 ww32
Jun 29, 2021
snapshot 2021 ww27
Pre-release

@JaccovG JaccovG released this Apr 20, 2021

Release Notes

  1. Version 2.0 EA

    • This is the first early access release for MLI2.0 (MLI 2.0 EA)
    • Intended for VPX only, not for EM/HS cores
    • Not all kernels are fully optimized
  2. This release supports following functional primitives

    • 2D Convolution
    • 2D depthwise Convolution
    • 2D Transpose Convolution
    • 2D Group Convolution
    • Fully Connected layer
    • Max and average pooling
    • LSTM and GRU recurrent cells
    • RNN Dense layer
    • Elementwise (add, sub, mul, min, max)
    • Permute
    • Argmax
    • ReLU, Leaky ReLU, Parametric ReLU
    • Softmax, Sigmoid, TanH, L2 Normalization
    • Helper functions to copy (partial) tensors (mli_mov*)
  3. Supported data layout:

    • Data layout HWC (Height-Width-Channel)
  4. Supported data format:

    • Fixed point 8bit and 16bit (fx8 and fx16)
    • Signed asymmetric 8bit quantization (sa8)
    • Signed asymmetric datatype supports per-tensor or per channel quantization with 16bit scale factors.
  5. Slicing support: creation of sub-tenors and support for non-contiguous tensor data.

  6. Metaware compiler:

    • Recommended version: 2021.03 or newer.
  7. Supported platforms:

    • VPX
    • x86 emulation
  8. Documentation:

    • Readme on how to get started.
Assets 3