Måns Nilsson

Conflict with bazel scripts and installed tflite_micro wheel

1

This comment summarize the problem: https://github.com/tensorflow/tflite-micro/pull/2545#issuecomment-2082843877 How to reproduce: Make sure tflite_micro is not installed. Either of these should run successfully, ./tensorflow/lite/micro/tools/ci_build/test_generate_integration_tests.sh ./tensorflow/lite/micro/tools/ci_build/test_generate_micro_mutable_op_resolver_tests.sh Now pip install tflite_micro and try to...

BUG=https://github.com/tensorflow/tflite-micro/issues/2564

Python dependency issue for aarch64

1

[tensorflow==2.15.0.post1 ](https://github.com/tensorflow/tflite-micro/blob/c01ca97f52ca451bfba714d7eb8ae0349d85f537/third_party/python_requirements.txt#L704) is not existing for aarch64. I am proposing to do a new update which should probably solve it: `bazel run //third_party:python_requirements.update` Some more background on the issue: https://github.com/tensorflow/tflite-micro/issues/2499#issuecomment-2009260738

Adds per channel quantization support for fully connnected reference and cmsis-nn kernel. BUG=missing support for FC per channel quantization

Output diff when enabling CMSIS_NN_USE_REQUANTIZE_INLINE_ASSEMBLY

2

https://github.com/ARM-software/CMSIS-NN/pull/171 introduce optional flag CMSIS_NN_USE_REQUANTIZE_INLINE_ASSEMBLY. When this is enabled there are uplifts and as well as downlifts depending on the target and model. For one particular combination that is Arm...

Printfs not showing

2

I noticed that printfs are not coming through, although context-ReportError seems to work. For example (after cloning tensorflow), put this in Prepare() in tensorflow/lite/micro/kernels/cmsis-nn/conv.cc printf("printf: this is not seen\n"); context->ReportError(context,...

Arm backend: test without output re-order workaround

4

By default outputs are re-ordered to correct order during TOSA lowering. However this is seen as a workaround as it should not be needed. Furthermore the output issue is not...

Tensorflow-cpu package does not support aarch64

2

This PR https://github.com/tensorflow/tflite-micro/pull/3261/ replace tensorflow with tensorflow-cpu to save memory. Problem is that tensorflow-cpu package is not supported on aarch64. The tensorflow-cpu package missing aarch64 support issue has been raised...