From 333770a695b72cb4930361e00b82924db8e501fa Mon Sep 17 00:00:00 2001 From: Tim Hall Date: Fri, 27 Nov 2020 12:32:27 +0000 Subject: MLBEDSW-3606: Update release notes - Updated for 20.11 release Signed-off-by: Tim Hall Change-Id: I24332b83a72250dfc64a245be796ed0f618a4234 --- RELEASES.md | 37 +++++++++++++++++++++++++++++++++---- 1 file changed, 33 insertions(+), 4 deletions(-) diff --git a/RELEASES.md b/RELEASES.md index c88c0335..ac10e8b4 100644 --- a/RELEASES.md +++ b/RELEASES.md @@ -5,19 +5,48 @@ main feature changes, interface changes and reported defects that have been fixed. The version numbering adheres to the [semantic versioning](https://semver.org/) scheme. -## Release 2.0.0 +## Release 2.0.0 - 30/11/2020 **Main feature changes:** -* New Vela configuration file format +* Advanced Spilling for Ethos-U65 Dedicated SRAM +* Specific network performance improvements for Ethos-U65 Dedicated SRAM +* General performance improvements for different networks and system +configurations +* Support SOFTMAX batching +* Improved supported operator checking and reporting +* Improved pass-through of Ethos-U hardware unsupported operators +* Enhanced performance estimations +* Extended external API +* New Vela configuration file format (with example `vela.ini`) +* Updated support to TensorFlow Lite 2.3 +* Made 8-bit TANH and LOGISTIC bit-exact and faster +* Generation of a network debug database to aid profiling and debug **Interface changes:** -* Non-backwards compatible changes to the Vela configuration file -* Addition of CLI options: `--memory-mode` +* Non-backwards compatible changes to the Vela configuration (.ini) file +* Removal of CLI options: + * `--show-minimum-possible-allocation`, `--ifm-ofm-overlap` + * `--permanent-storage`, `--global-memory-clock-scale` +* Change to CLI options: + * `--config` (support multiple) + * `--accelerator-config` (change 'yoda' to 'ethos-u65') +* Addition of CLI options: + * `--api-version`, `--supported-ops-report`, `--enable-debug-db` + * `--verbose-config`, `--cache-bias-scale-tensor`, `--memory-mode` + * `--cpu-tensor-alignment` +* Addition of External APIs: + * Command Stream generation, Driver Payload creation, and Finding a Block Config **Reported defect fixes:** +* Add support for 128-Byte alignment (`--cpu-tensor-alignment`) (MLCE-221) +* Bug with SPLIT and non-unique scale & bias tensors (MLCE-234) +* Bug with overlapping tensor buffers in CONCATENATION (MLCE-246) +* Bug with a non-splitting SPLIT (MLCE-247) +* Regression in STRIDED_SLICE between v1.0.0 and v1.2.0 (MLCE-252) +* Add batch support to SOFTMAX (MLCE-265) * Vela config file examples (MLCE-277) ## Release 1.2.0 - 31/08/2020 -- cgit v1.2.1