Skip to content

CHANGELOG.md

Latest commit

 

History

History
276 lines (200 loc) · 9.19 KB

CHANGELOG.md

File metadata and controls

276 lines (200 loc) · 9.19 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog.

[26.03]

Added

  • MLEK Zephyr module to allow platform-agnostic library to be used in Zephyr-based projects
  • Linker script override options at global and use-case level
  • Configure-time and runtime detection of the model's intended target to highlight any mismatches with build parameters
  • Improved visual output for the ExecuTorch Conformer ASR use case
  • AGENTS.md to support coding agents
  • Pre-commit configuration

Changed

  • Updated to 26.02 NPU components and dependencies (core-driver, core-platform, Vela 5.0.0, CMSIS-6, CMSIS-DSP, CMSIS-NN, TensorFlow Lite Micro and ExecuTorch 1.1.0)
  • Refactored the source and CMake layout to separate application and library code to better-support inclusion in external projects
  • Refactored Python tooling into an installable mlek_tools package
  • Improved detection and reporting of ExecuTorch model memory mode, memory use and ops
  • Improved error handling when ExecuTorch model memory requirements exceed limits

Fixed

  • Replaced the Yolo-Fastest model download URL with a more suitable alternative

[25.12]

Added

  • Support for ExecuTorch framework alongside TensorFlow Lite Micro
  • ASR use case updated to allow building with Conformer model via ExecuTorch
  • Image Classification use case support for ExecuTorch version of MobileNet V2 and DeiT tiny
  • Multi-threaded model download and optimisation in set up script

Changed

  • Updated to 25.08 NPU components and dependencies (core-driver, core-platform, Vela 4.4.1, CMSIS-6, CMSIS-DSP, CMSIS-NN, TensorFlow Lite Micro and ExecuTorch 1.0.0)

[25.05]

Added

  • Support for LLVM-based Arm Toolchain for Embedded (ATfE) versions 19.1.5, 20.0.0 and 20.1.0
  • Interactive mode option to allow user control when running use cases

Changed

  • Updated bare-metal toolchain files to always add function and data sections for both toolchains
  • Updated CMSIS-DSP build to only build required sources by default
  • General documentation updates: fixed broken links, added macOS support commands for FPGA deployment, updated URLs to GitLab

[25.02]

Changed

  • Updated to 25.02 NPU components and dependencies (core-driver, core-platform, Vela 4.2.0, CMSIS-6, CMSIS-DSP, CMSIS-NN and TensorFlow Lite Micro)
  • Minor update to generic BSP with CMSIS_device_header definition preferred to CPU_HEADER_FILE
  • Removed Arm Virtual Hardware references from documentation

Fixed

  • CMSIS-NN usage in TensorFlow Lite Micro (bug introduced in Sep 2024)

[24.11]

Added

  • Support for Arm Corstone-320 FVP, including Arm Ethos-U85 NPU
  • HAL APIs for obtaining camera and audio data for use case apps
  • Support for Arm Virtual Streaming Interface (VSI) FVP feature as camera and audio data source

Changed

  • Updated to 24.11 NPU components and dependencies (core-driver, core-platform, Vela 4.1.0, CMSIS-6, CMSIS-DSP, CMSIS-NN and TensorFlow Lite Micro)
  • Moved to C11 and C++17 standards

Removed

  • Dropped CMSIS-5 support; now using CMSIS-6 exclusively

[24.08]

Added

  • BatchMatMul to all-ops resolver
  • Setup script now allows download location to be specified

Changed

  • Updated to 24.08 NPU components and dependencies (core-driver, core-platform, Vela 4.0.0, CMSIS, CMSIS-NN and TensorFlow Lite Micro)
  • Test file generation supports types other than int8

[24.05]

Added

  • Support for Arm Corstone-315 targets

Changed

  • Updated to 24.05 NPU components and dependencies (core-driver, core-platform, Vela 3.12.0, CMSIS, CMSIS-NN and TensorFlow Lite Micro)
  • CMSIS pack refresh (version from 23.02 added)
  • CMake user configuration options refactored
  • Dockerfile updates including FVP versions downloaded
  • Minor improvements to ease integration for cases where this project is a dependency

[23.11]

Added

  • Support for Arm Cortex-M85 with GCC

Changed

  • Updated to 23.11 NPU components and dependencies (core-driver, core-platform, Vela 3.10.0, CMSIS, CMSIS-NN and TensorFlow Lite Micro)
  • Minimum Python version increased to 3.10
  • Updated Python files to conform to Pylint

[23.08]

Changed

  • Updated to 23.08 NPU components and dependencies (core-driver, core-platform, Vela 3.9.0, CMSIS, CMSIS-NN and TensorFlow Lite Micro)
  • Updated Python package requirements for compatibility
  • Minimum Armclang version increased to 6.19

[23.05]

Changed

  • Updated to 23.05 NPU components and dependencies (core-driver, core-platform, Vela 3.8.0, CMSIS, CMSIS-NN and TensorFlow Lite Micro)
  • Minor updates to CMSIS-pack dependencies
  • Updated Python package requirements for compatibility
  • Minimum Python version increased to 3.9

[23.02]

Changed

  • Updated to 23.02 NPU components and dependencies (core-driver, core-platform, Vela 3.7.0, CMSIS, CMSIS-DSP, CMSIS-NN and TensorFlow Lite Micro)
  • Improvement to PMU counters: reducing cache maintenance burden for Arm Cortex-M55 CPUs when NPU is used
  • Inclusive language updates
  • Documentation updates including links to AN555 resources
  • Updates to Python requirements file for virtual environment

[22.11]

Added

  • Formal support for CMake presets
  • Support for Arm Compiler 6.19
  • Published CMSIS-Pack under resources

Changed

  • Updated to 22.11 NPU components and dependencies (core-driver, core-platform, Vela 3.6.0, CMSIS, TensorFlow Lite Micro)
  • CMSIS-NN added as a new dependency (moved out of core CMSIS repository)
  • Updates for Arm Corstone-310 platform

Fixed

  • Bug fix for building with NPU disabled

[22.08]

Added

  • Formal support for Arm Corstone-310 Arm Virtual Hardware and FPGA implementations
  • CMake presets
  • Provision to allow platform-agnostic use case API to be packaged as CMSIS pack

Changed

  • Updated to 22.08 NPU components and dependencies (core-driver, core-platform, Vela 3.5.0, CMSIS, TensorFlow Lite Micro)

Fixed

  • Arm Ethos-U65 NPU default AXI burst length

[22.05]

Added

  • Support for Arm Cortex-M85 (Armclang 6.18 required)

Changed

  • Updated to 22.05 NPU components and dependencies (core-driver, core-platform, Vela 3.4.0, CMSIS, TensorFlow Lite Micro)
  • Restructured repository sources to allow generation of CMSIS packs
  • Minimum Python version reduced to 3.7 (aligned with Vela 3.4.0)
  • Minimum CMake version increased to 3.21.0

Fixed

  • Various bug fixes

[22.02]

Added

  • Object Detection use case

Changed

  • Updated to 22.02 NPU components and dependencies (core-driver, core-platform, Vela 3.3.0, CMSIS, TensorFlow Lite Micro)
  • Replaced DSCNN with MicroNet for KWS and KWS_ASR use cases
  • Minimum CMake version increased to 3.15.6, Armclang to 6.16, Python to 3.8
  • Initial restructuring of repository sources

Fixed

  • Various minor bug fixes

[21.11]

Added

  • Support for different memory modes: Shared_Sram, Dedicated_Sram and Sram_Only
  • Noise Reduction use case
  • Dynamic load support for FVP for inference runner use case

Changed

  • Updated to 21.11 NPU components and dependencies (core-software, core-driver, Vela 3.2.0, CMSIS, TensorFlow Lite Micro)
  • Updated support for Arm GNU Embedded Toolchain 10.3-2021.07 and Arm Compiler 6.17
  • Changes to support AN552 design (new Arm Corstone-300 implementation)

[21.08]

Added

  • Visual Wake Word use case

Changed

  • Updated to 21.05 NPU components (core-software, core-driver, Vela 3.0.0)
  • TensorFlow submodule changed to https://github.com/tensorflow/tflite-micro
  • Image classification model changed (from uint8 to int8 datatype)
  • Added support for Corstone-300 + Arm Ethos-U65

[21.05]

Added

  • Script to download and optimize default models
  • Script to run default build flow
  • Model for Anomaly Detection use case
  • Support for build with Arm GNU Embedded Toolchain (10.2.1)

Deprecated

  • Support for target subsystem SSE-200

[21.03]

Added

  • Simple platform support
  • Model conditioning examples
  • Tests for native platform
  • Anomaly detection use case

Changed

  • Build changed to use sources of the dependency libraries

[20.11]

Added

  • SSE-200 and SSE-300 system support
  • Models with multiple output tensors support
  • Generic inference runner use case
  • ASR triggered by KWS in combined use case (kws_asr)

Changed

  • Build cmake parameters changed: TARGET_SUBSYSTEM added, TARGET_PLATFORM accepted values changed

Removed

  • Support for simple fixed virtual platform for Arm Ethos-U55 and Arm Cortex-M55

[20.09]

Added

  • Speech recognition use case

Changed

  • Updated to TensorFlow Lite Micro version > 2.3.0
  • Updated NPU Fastmodel version to r0p2-00eac0-rc4

[20.08]

Added

  • Keyword spotting use case
  • Person detection use case

Known Issues

  • Telnet connection to FastModel environment may stop responding after some period of inactivity

[20.05]

Added

  • MPS3 FPGA build support
  • Configurable timing adaptor
  • Active and Idle cycle counts for NPU and CPU profiling report
  • Windows support for build scripts

Changed

  • FastModel environment built with FastModel Tools v11.10.22
  • Source code structure and build scripts refactored to support multiple ML use cases
  • Using EAC NPU software model and drivers

Known Issues

  • Telnet connection to FastModel environment may stop responding after some period of inactivity