Releases: renesas-rz/meta-renesas-ai
Renesas RZ/G AI BSP v5.2.0
Change log:
- Upgraded all platforms to use the RZ/G BSP tag BSP-3.0.6-update2
- Upgraded TensorFlow Lite version to v2.15.1
- Upgraded ArmNN version to v24.05
- Upgraded Arm Compute library version to v24.06
- Improved the armnn-benchmark application so that it runs benchmarks on models specified in a configuration file instead of a hard-coded list in the application
- Enabled FP16 turbo mode by default when benchmarking with the tfLiteDelegateBenchmark application for Arm Cortex-A55 devices
- Various CI improvements
- Various code improvements and bug fixes
Supported AI tools:
- Arm Neural Network SDK v24.05 / Arm Compute Library v24.06
- ONNX Runtime v1.8.0
- TensorFlow Lite v2.15.1
Supported Platforms:
- Renesas RZ/G2H hihope-rzg2h
- Renesas RZ/G2M hihope-rzg2m
- Renesas RZ/G2N hihope-rzg2n
- Renesas RZ/G2E ek874
- Renesas RZ/G2L smarc-rzg2l
- Renesas RZ/G2LC smarc-rzg2lc
- Renesas RZ/G2UL smarc-rzg2ul
- Renesas RZ/V2L smarc-rzv2l
For all platforms use meta-renesas tag BSP-3.0.6-update2.
Renesas RZ/G AI BSP v5.1.0
Change log:
- Upgraded all platforms to use the RZ/G BSP tag BSP-3.0.4
- Added support for TensorFlow Lite Python bindings
- Added support for the RZ/V2L smarc-rzv2l EVK platform
- Changed architecture optimisations for the RZ/G2L family of devices when using the ArmNN framework. This has shown inference performance improvements of 50-60% when using float32 image recognition models
- Structural improvements to the AI framework recipes to better split out benchmarking support
- Various CI improvements
- Various code improvements and bug fixes
Supported AI tools:
- Arm Neural Network SDK / Arm Compute Library v22.02
- ONNX Runtime v1.8.0
- TensorFlow Lite v2.5.3
Supported Platforms:
- Renesas RZ/G2H hihope-rzg2h
- Renesas RZ/G2M hihope-rzg2m
- Renesas RZ/G2N hihope-rzg2n
- Renesas RZ/G2E ek874
- Renesas RZ/G2L smarc-rzg2l
- Renesas RZ/G2LC smarc-rzg2lc
- Renesas RZ/G2UL smarc-rzg2ul
- Renesas RZ/V2L smarc-rzv2l
For all platforms use meta-renesas tag BSP-3.0.4.
Renesas RZ/G AI BSP v5.0.0
Change log:
- Upgraded all platforms to use the new "unified" meta-renesas BSP, BSP-3.0.0
- All platforms now use the same versions of all frameworks and dependencies
- All CI build and test scripts are now included in meta-renesas-ai
- The same Yocto configuration templates are now used for all platforms, configured by the set-config-files.sh script
- Various code improvements and bug fixes
Supported AI tools:
- Arm Neural Network SDK / Arm Compute Library v22.02
- ONNX Runtime v1.8.0
- TensorFlow Lite v2.5.3
Supported Platforms:
- Renesas RZ/G2H hihope-rzg2h
- Renesas RZ/G2M hihope-rzg2m
- Renesas RZ/G2N hihope-rzg2n
- Renesas RZ/G2E ek874
- Renesas RZ/G2L smarc-rzg2l
- Renesas RZ/G2LC smarc-rzg2lc
- Renesas RZ/G2UL smarc-rzg2ul
For all platforms use meta-renesas tag BSP-3.0.0.
Renesas RZ/G AI BSP v4.7.0
Change log:
- Upgraded base Renesas RZ/G2L BSP to tag rzg2l_bsp_v1.4
- Added RZ/G2UL smarc-rzg2ul support
- Added support for ArmNN / Arm Compute Library v22.02
- Improved TFLite delegate testing application to support both XNNPACK and ArmNN delegates
- Various code improvements and bug fixes
Supported AI tools:
- Arm Neural Network SDK / Arm Compute Library v22.02, v21.05
- ONNX Runtime v1.8.0
- TensorFlow Lite v2.5.3, v2.3.1
Supported Platforms:
- Renesas RZ/G2H hihope-rzg2h
- Renesas RZ/G2M hihope-rzg2m
- Renesas RZ/G2N hihope-rzg2n
- Renesas RZ/G2E ek874
- Renesas RZ/G2L smarc-rzg2l
- Renesas RZ/G2LC smarc-rzg2lc
- Renesas RZ/G2UL smarc-rzg2ul
For RZ/G2 platforms use meta-rzg2 tag BSP-1.0.10-update1.
For RZ/G2L platforms use meta-rzg2 tag rzg2l_bsp_v1.4.
Renesas RZ/G AI BSP v4.6.0
Change log:
- Upgraded base Renesas RZ/G2L BSP to tag rzg2l_bsp_v1.3-update2
- Upgraded base Renesas RZ/G2 BSP to tag BSP-1.0.10-update1
- Added support for TensorFlow Lite v2.5.3
- Added XNNPACK support for TensorFlow Lite on the RZ/G2L family of platforms
- Dropped support for RZ/G1 platforms
- Various code improvements and bug fixes
Supported AI tools:
- Arm Neural Network SDK / Arm Compute Library v21.05
- ONNX Runtime v1.8.0
- TensorFlow Lite v2.5.3, v2.3.1
Supported Platforms:
- Renesas RZ/G2H hihope-rzg2h
- Renesas RZ/G2M hihope-rzg2m
- Renesas RZ/G2N hihope-rzg2n
- Renesas RZ/G2E ek874
- Renesas RZ/G2L rzg2l-smarc evk
- Renesas RZ/G2LC rzg2lc-smarc evk
For RZ/G2 platforms use meta-rzg2 tag BSP-1.0.10-update1.
For RZ/G2L platforms use meta-rzg2 tag rzg2l_bsp_v1.3-update2.
Renesas RZ/G AI BSP v4.5.0
Change log:
- Upgrade base Renesas RZ/G2L BSP to tag rzg2l_bsp_v1.3-update1
- Add Renesas RZ/G2LC rzg2lc-smarc evk support
- Add ArmNN Mali GPU support for RZ/G2L and RZ/G2LC
Supported AI tools:
- Arm Neural Network SDK v21.05
- ONNX Runtime v1.8.0
- TensorFlow Lite v2.3.1
Supported Platforms:
- Renesas RZ/G2H hihope-rzg2h
- Renesas RZ/G2M hihope-rzg2m
- Renesas RZ/G2N hihope-rzg2n
- Renesas RZ/G2E ek874
- Renesas RZ/G2L rzg2l-smarc evk
- Renesas RZ/G2LC rzg2lc-smarc evk
- Renesas RZ/G1H iwg21m
- Renesas RZ/G1M iwg20m-g1m
- Renesas RZ/G1E iwg22m
For RZ/G1 platforms use meta-renesas tag certified-linux-v2.1.9.
For RZ/G2 platforms use meta-rzg2 tag BSP-1.0.8.
For the RZ/G2L platform use meta-rzg2 tag rzg2l_bsp_v1.3-update1.
Renesas RZ/G AI BSP v4.4.0
Change log:
- Upgrade base Renesas RZ/G1 BSP to tag certified-linux-v2.1.9
- Upgrade base Renesas RZ/G2 BSP to tag BSP-1.0.8
- Upgrade base Renesas RZ/G2L BSP to tag rzg2l_bsp_v1.3
- Drop GCC v7.3 (meta-rzg1-gcc-linaro) support for RZ/G1 devices
- Improve sample build script
- Add ArmNN delegate header files to dev package
- Correct MobileNet v3 model data types in meta-benchmark/armnn
Supported AI tools:
- Arm Neural Network SDK v21.05
- ONNX Runtime v1.8.0
- TensorFlow Lite v2.3.1
Supported Platforms:
- Renesas RZ/G2H hihope-rzg2h
- Renesas RZ/G2M hihope-rzg2m
- Renesas RZ/G2N hihope-rzg2n
- Renesas RZ/G2E ek874
- Renesas RZ/G2L rzg2l-smarc evk
- Renesas RZ/G1H iwg21m
- Renesas RZ/G1M iwg20m-g1m
- Renesas RZ/G1E iwg22m
For RZ/G1 platforms use meta-renesas tag certified-linux-v2.1.9.
For RZ/G2 platforms use meta-rzg2 tag BSP-1.0.8.
For the RZ/G2L platform use meta-rzg2 tag rzg2l_bsp_v1.3.
Renesas RZ/G AI BSP v4.3.0
Main features of this release:
- Upgrade ArmNN SDK to v21.05
- Add ONNX Runtime support for the Renesas RZ/G2L rzg2l-smarc evk platform
- Reduce the number of supported frameworks
- Other improvements and minor fixes
Supported AI tools:
- Arm Neural Network SDK v21.05
- ONNX Runtime v1.8.0
- TensorFlow Lite v2.3.1
Supported Platforms:
- Renesas RZ/G2H hihope-rzg2h
- Renesas RZ/G2M hihope-rzg2m
- Renesas RZ/G2N hihope-rzg2n
- Renesas RZ/G2E ek874
- Renesas RZ/G2L rzg2l-smarc evk
- Renesas RZ/G1H iwg21m
- Renesas RZ/G1M iwg20m-g1m
- Renesas RZ/G1E iwg22m
For RZ/G1 platforms use meta-renesas tag certified-linux-v2.1.8.
For RZ/G2 platforms use meta-rzg2 tag BSP-1.0.7.
For the RZ/G2L platform use meta-rzg2 branch dunfell/rzg2l commit 2313d60eb75e5c86ce3e42ad378c8473f8e95c88.
Renesas RZ/G AI BSP v4.2.0
Main features of this release:
- Upgrade ONNX Runtime to v1.8.0
- Add support for the Renesas RZ/G2L rzg2l-smarc evk platform for the ArmNN and TensorFlow Lite frameworks
- Improved various benchmarking applications
- Other improvements and minor fixes
Supported AI tools:
- Arm Neural Network SDK v21.02
- Caffe2 v0.8.1
- Google Coral USB TPU (frogfish)
- ONNX Runtime v1.8.0
- OpenCV v4.1.1
- PyTorch v1.5.1
- TensorFlow v2.3.1
- TensorFlow Lite v2.3.1
Supported Platforms:
- Renesas RZ/G2H hihope-rzg2h
- Renesas RZ/G2M hihope-rzg2m
- Renesas RZ/G2N hihope-rzg2n
- Renesas RZ/G2E ek874
- Renesas RZ/G2L rzg2l-smarc evk
- Renesas RZ/G1H iwg21m
- Renesas RZ/G1M iwg20m-g1m
- Renesas RZ/G1E iwg22m
For RZ/G1 platforms use meta-renesas tag certified-linux-v2.1.8.
For RZ/G2 platforms use meta-rzg2 tag BSP-1.0.7.
For the RZ/G2L platform use meta-rzg2 branch dunfell/rzg2l commit 2313d60eb75e5c86ce3e42ad378c8473f8e95c88.
Renesas RZ/G AI BSP v4.1.0
Main features of this release:
- Upgrade ArmNN SDK and Arm Compute Library to v21.02
- Add support and sample application for ArmNN TfLite Delegate
- Add support in armnnBenchmark sample application for ArmNN 'fp16-turbo' and 'fast-math' modes
- Add ability to set the compute backend in the armnnBenchmark sample application
- Upgrade base Renesas RZ/G1 BSP to tag certified-linux-v2.1.8
- Upgrade base Renesas RZ/G2 BSP to tag BSP-1.0.7
- Changed the configuration schema in meta-benchmarks. Sample configuration templates are now provided per platform for two builds:
- armnn+coral+tf+tfl
- caffe2+onnx+opencv+pytorch
- Other improvements and minor fixes
Supported AI tools:
- Arm Neural Network SDK v21.02
- Caffe2 v0.8.1
- Google Coral USB TPU (frogfish)
- ONNX Runtime v1.1.2
- OpenCV v4.1.1
- PyTorch v1.5.1
- TensorFlow v2.3.1
- TensorFlow Lite v2.3.1
Supported Platforms:
- Renesas RZ/G2H hihope-rzg2h
- Renesas RZ/G2M hihope-rzg2m
- Renesas RZ/G2N hihope-rzg2n
- Renesas RZ/G2E ek874
- Renesas RZ/G1H iwg21m
- Renesas RZ/G1M iwg20m-g1m
- Renesas RZ/G1E iwg22m
For RZ/G1 platforms use meta-renesas tag certified-linux-v2.1.8.
For RZ/G2 platforms use meta-rzg2 tag BSP-1.0.7.