diff options
author | Tim Hall <tim.hall@arm.com> | 2023-11-16 12:10:57 +0000 |
---|---|---|
committer | Tim Hall <tim.hall@arm.com> | 2023-11-16 12:48:26 +0000 |
commit | 5aa9ae24ed834a3aeba4917823437bac75e14ab3 (patch) | |
tree | 1ecf66c9e298436c786c4013efbaf4151d0c4e4c | |
parent | 2742947db9dbec68f0ae978e70068b44651d6a8a (diff) | |
download | ethos-u-vela-5aa9ae24ed834a3aeba4917823437bac75e14ab3.tar.gz |
MLBEDSW-8240: Document reference comparison point
- Updated TensorFlow Support section
Change-Id: Ic2551f44e7dfa996a5dcc8840d480b7985415a0a
Signed-off-by: Tim Hall <tim.hall@arm.com>
-rw-r--r-- | README.md | 8 |
1 files changed, 5 insertions, 3 deletions
@@ -45,9 +45,11 @@ The tool has limited functionality for compiling a (EXPERIMENTAL). ## TensorFlow Support -Vela is tested by comparing the numerical behaviour of the optimised operators -against that of the corresponding TensorFlow Lite reference kernels. The -following list indicates which version is used for comparison: +Vela is tested by comparing the bit exact numerical behaviour of the Ethos-U +optimised operators against that of the corresponding TensorFlow Lite reference +kernels (or TensorFlow Lite for Microcontrollers reference kernels in the case +of the UNIDIRECTIONAL_SEQUENCE_LSTM operator). The following list indicates +which version is used for comparison: * Vela 3.10.0 to current supports TensorFlow 2.14 * Vela 3.9.0 supports TensorFlow 2.12 |