Skip to content

T
Tflite Micro

码匠许师傅 / Tflite Micro
11 个月 前同步成功

通知 1
Star 0
Fork 0
T
Tflite Micro

前往新版Gitcode，体验更适合开发者的 AI 搜索 >>

切换分支/标签
  1. 21 9月, 2022 2 次提交
  3. 20 9月, 2022 4 次提交
    • T
      Syncing the test back to github. · 13bb1c28
      由 TANMAY DAS 提交于 
      BUG=b/247598846
      13bb1c28
    • R
      Access input/output tensor details from python · 5606c895
      由 Renjie 提交于 
      Inference on a quantized model requires information about 
* input tensor quantization parameters (dtype, scale, zero point): to convert the input float data into the quantized one
* output tensor quantization parameters: convert the quantized result back to float (i.e., sanity check with normal model output)

This PR allows users to access these information from the python interpreter. 

It also refactored the error handling mechanism following TF and TFLite.  

BUG=b/245831075, b/246651364
      5606c895
    • T
      Automated sync from github.com/tensorflow/tensorflow (#1423) · bc9dc295
      由 TFLM-bot 提交于
      bc9dc295
    • V
      TFLite Micro framework changes to support log refactor · 95761fcf
      由 Vamsi Krishna Manchala 提交于 
      This PR-

1. Changes the TFLite Micro Framework classes to eliminate the usage of `MicroErrorReporter` and `ErrorReporter` classes.
2. Keeps the version of the APIs, that expect ErrorReporter for backwards compatibility. 
	- This compatibility is limited to the bare minimum set of API currently being used.
	- The minimum support to the version of API expecting ErrorReporter will be removed as a follow-up PR, expected to be merged next week.
	- Any new applications or examples using TFLM should no longer be creating and passing-in an ErrorReporter. 


BUG=http://b/245802069
      95761fcf
  5. 16 9月, 2022 2 次提交
    • T
      Initial version of build structure change. · 4051ca97
      由 TANMAY DAS 提交于 
      This is a CL that allows us to build and run TFLM binaries from `tflite-micro` directory as well as from the parent directory of `tflite-micro`. 

Example : to run all the tests in TFLM from inside `tflite-micro` directory, we need to run the following command 
`make -f tensorflow/lite/micro/tools/make/Makefile test`

Now to run the tests from the parent directory of `tflite-micro`, we need to run the following command 
`make -f tflite-micro/tensorflow/lite/micro/tools/make/Makefile test TENSORFLOW_ROOT=tflite-micro/`

Please note that the '/' at the end of 'TENSORFLOW_ROOT=tflite-micro/' is mandatory. 

BUG=244204467
      4051ca97
    • T
      Automated sync from github.com/tensorflow/tensorflow · 6390ebee
      由 TFLM-bot 提交于 
      BUG=automated sync from upstream
NO_CHECK_TFLITE_FILES=automated sync from upstream
      6390ebee
  7. 15 9月, 2022 2 次提交
  9. 14 9月, 2022 1 次提交
  11. 11 9月, 2022 1 次提交
  13. 09 9月, 2022 3 次提交
    • R
      Format fix for #1388 · e1f37278
      由 Renjie 提交于 
      Google3 is not happy with the format and some bazel build dependency (Pillow). This PR fixes the python format and removed unnecessary dependency. 
BUG=cleanup
      e1f37278
    • A
      Fix the issue described in http://b/245773171 · 94b1e003
      由 Advait Jain 提交于 
      This PR fixes the issue for one particular target, but there are still others with the exact same error.

The mako template is already correct as can be seen: 
https://github.com/tensorflow/tflite-micro/blob/87d2db744611937f5dfb818ef639b84370cffe44/tensorflow/lite/micro/integration_tests/templates/integration_tests_cc.mako#L52-L54

What this means is that many of the integration_tests.cc need to be regenerated with the updated mako template.

Currently, we do not have an easy way to regenerate the integration_tests if all we have is a change to the mako template.  Instead, we need to perform the full step of going from the model and regenerating all the tflite and golden input and output csvs. That is a problem and should be fixed.

Additionally, the integration_test targets are currently cc_binary instead of cc_test which means that we do not have any CI coverage for them. We should revisit that decision as well.

BUG=http://b/245773171
      94b1e003
    • R
      Add MNIST LSTM model evaulation · 945d1976
      由 Renjie 提交于 
      # Evaluate trained LSTM model on MNIST dataset 

## evaluate.py

Given the trained .tflite model and a MNIST image path, output the predicted category (digit 0-9) probability. 

Example:
`bazel build tensorflow/lite/micro/examples/mnist_lstm:evaluate`
`bazel-bin/tensorflow/lite/micro/examples/mnist_lstm/evaluate --model_path='.tflite file path' --img_path='MNIST image path'`

## evaluate_test.py

Test the evaluation code under the following criteria: 
* Using random input to compare results from TFLite interpreter's results
* Using sample MNIST images to test model prediction accuracy 

Example:
bazel test tensorflow/lite/micro/examples/mnist_lstm:evaluate_test

## Issues 
Xtensa implementation is out of dated and not tested. http://b/230666079

BUG=http://b/242765271

As a side benfeit, this PR also confirms support for LSTMs (reference floating point). Fixes #1243 and fixes #920
      945d1976
  15. 07 9月, 2022 1 次提交
  17. 03 9月, 2022 1 次提交
  19. 02 9月, 2022 2 次提交
  21. 01 9月, 2022 6 次提交
  23. 31 8月, 2022 1 次提交
    • R
      Add MNIST LSTM model training · 639ae637
      由 rewu93 提交于 
      BUG=http://b/242765271

# Train/save/convert LSTM model on MNIST dataset 

Using bazel: 
`bazel build tensorflow/lite/micro/examples/mnist_lstm:train`
`bazel-bin/tensorflow/lite/micro/examples/mnist_lstm/train`

## Terminal arguments: 
* `--epoch` : number of epochs for model training, default 1 for fast trainning 
* `--save_dir`: the directory to save the trained model, default to `/tmp/trained_model`. 
* `--save_tf_model`: if specified, save the intermediate tensorflow model (debug usage) 

## Example: 
`bazel build tensorflow/lite/micro/examples/mnist_lstm:train`
`bazel-bin/tensorflow/lite/micro/examples/mnist_lstm/train --epoch=10 --save_dir=trained_model`

Code above will train a simple LSTM model for MNIST digits recognition with 10 epochs. The trained model is converted to tflite flatbuffer format, and the .tflite file is saved to the trained_model folder in the source directory. Note the intermediate tensorflow model is not saved in this case (pass --save_tf_model=True to enable it).
      639ae637
  25. 30 8月, 2022 3 次提交
  27. 26 8月, 2022 3 次提交
  29. 19 8月, 2022 1 次提交
    • A
      add flag for VHT-AMI to run armclang · dc25e1e3
      由 Annie Tallund 提交于 
      We need to differentiate between the compiler used in the two CI tests Cortex-M and Cortex-M on Arm Virtual Hardware, which is what this fix does. Cortex-M uses gcc and AVH is for armclang. Related to 

BUG=#1305
      dc25e1e3
  31. 18 8月, 2022 3 次提交
  33. 17 8月, 2022 3 次提交
  35. 16 8月, 2022 1 次提交