vectorize KPLS: `tf.vectorized_map` fallback cause #291

bhack · 2022-04-10T13:13:20Z

This is the current list of fallback we have on master with the related cause for our KPLs ~~(it needs tensorflow/tensorflow#55562 or tf-nightly when merged)~~:

pytest -rP keras_cv/layers/preprocessing/ | grep while_loop | sort | uniq

WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting AdjustHue cause Input "delta" of op 'AdjustHue' expected to be loop invariant.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting AdjustSaturation cause Input "scale" of op 'AdjustSaturation' expected to be loop invariant.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting DepthwiseConv2dNative cause Input "filter" of op 'DepthwiseConv2dNative' expected to be not loop invariant.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting DynamicPartition cause there is no register converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting DynamicStitch cause there is no register converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting HistogramFixedWidth cause there is no register converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting ImageProjectiveTransformV3 cause there is no register converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting RandomShuffle cause there is no register converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting StridedSlice cause Input "input" of op 'StridedSlice' expected to be not loop invariant.

You can find exactly which test is involed running:
pytest -rP keras_cv/layers/preprocessing/

Some of these could be fixed with a refactor in our code, others need a contribution (trivial or not) for a registered converter (e.g. See #259 (comment))

I think that we could care about this:

monitor how many fallback we are accumulating directly in the CI, with a Github Action, so that we could check this at the PR stage
contribute the (trivial?) converters that we care about in https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/ops/parallel_for/pfor.py#L920-L997
(extra) extend the same policy to jit_compile when we have our compile API ready

The text was updated successfully, but these errors were encountered:

bhack · 2022-04-10T13:58:59Z

/cc @wangpengmit @rohan100jain

bhack · 2022-04-11T14:25:48Z

Just for reference I copy-paste the current Pfor heuristic:

https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/ops/parallel_for/pfor.py#L3372-L3389

  The conversion process uses a simple greedy heuristic. It walks the loop body
  and tries to express the functionality of running each node in a loop with a
  new set of nodes. When converting an op several cases are possible:
  - The op is not inside the loop body. Hence it can be used as is.
  - The op does not depend on the iteration number and is stateless. In this
    case, it can be used as is.
  - The op is not stateful, and depends on iteration number only through control
    dependencies. In this case, we can create a single op with same inputs and
    attributes, but with "converted" control dependencies.
  - The op is not stateful, and all its inputs are loop invariant. In this
    case, similar to above, we can create a single op with same inputs and
    attributes, but with "converted" control dependencies.
  - The op is stateful or at least one of the inputs is not loop invariant. In
    this case, we run the registered converter for that op to create a set of
    converted ops. All nodes in the set will have converted control dependencies
    corresponding to control dependencies of the original op. If the op returned
    multiple outputs, "converted outputs" could be produced by different ops in
    this set.

bhack · 2022-04-11T15:43:47Z

I want to add an example picking the first fallback in the list that has already a registered converter:

WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting AdjustHue cause Input "delta" of op 'AdjustHue' expected to be loop invariant.

This is the converter:

@RegisterPFor("AdjustHue")
def _convert_adjust_hue(pfor_input):
  images = pfor_input.stacked_input(0)
  delta = pfor_input.unstacked_input(1)
  return wrap(gen_image_ops.adjust_hue(images, delta), True)

So we cannot supersed the limit of the original design of tf.image.adjust_hue where it is already vectorized on the batch dimension but not on the delta arg as it is just float value:

def adjust_hue(image, delta, name=None):
  ....
  ....
  Args:
    image: RGB image or images. The size of the last dimension must be 3.
    delta: float.  How much to add to the hue channel.

IMHO this is going to point again to the original topic related to what kind of empiric evidence we are going to sustain within batch randomization, with the current native ops set we have in TF, vs other alternative solutions like #146 (comment) at least as a workaround to not generate a random augmentation for each sample in the batch for each op.

bhack · 2022-04-13T15:54:07Z

As tf.image.* is also currently MIA/orphan I don't think that we could extend the API and kernels to add the non scalar support of delta, scale etc.. params

bhack · 2022-04-15T19:00:17Z

Do you have talked about this internally between teams? Any action item?

wangpengmit · 2022-04-15T20:51:20Z

We (the TF core team) don't monitor keras-team/keras-cv . The issue needs to be reposted to tensorflow/tensorflow to be under our radar.

bhack · 2022-04-15T20:58:23Z

@wangpengmit I understand but this is really a boundary topic. What kind of ticket do you want for TF (core)? Just for the missing converters?

wangpengmit · 2022-04-15T21:10:11Z

It can be just a title and a link to this one. TF has a system for alerting TF developers of tensorflow/tensorflow issues.

bhack · 2022-04-15T21:15:50Z

Ok let see what we can do in TF at tensorflow/tensorflow#55639 and what we need to change in Keras-cv

…e conversions. Reference issues: keras-team/keras-cv#264, keras-team/keras-cv#291 PiperOrigin-RevId: 443809170

bhack · 2022-04-28T13:33:42Z

I update the fallback list on the current master b542921:

WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting AdjustContrastv2 cause Input "contrast_factor" of op 'AdjustContrastv2' expected to be loop invariant.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting AdjustHue cause Input "delta" of op 'AdjustHue' expected to be loop invariant.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting AdjustSaturation cause Input "scale" of op 'AdjustSaturation' expected to be loop invariant.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting Bitcast cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting DepthwiseConv2dNative cause Input "filter" of op 'DepthwiseConv2dNative' expected to be not loop invariant.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting DynamicPartition cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting DynamicStitch cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting HistogramFixedWidth cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting ImageProjectiveTransformV3 cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting RandomShuffle cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting RngReadAndSkip cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting StatelessRandomGetKeyCounter cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting StatelessRandomUniformFullIntV2 cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting StatelessRandomUniformV2 cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1082 Using a while_loop for converting StridedSlice cause Input "input" of op 'StridedSlice' expected to be not loop invariant.

bhack · 2022-05-24T23:17:27Z

@qlzh727 Just an extra note, as you have already seen in tensorflow/tensorflow#56242

We need to extend this analysis also to the Keras repo for:
https://github.com/keras-team/keras/blob/master/keras/layers/preprocessing/image_preprocessing.py

qlzh727 · 2022-05-25T18:59:38Z

Ack. Will take a look later this week.

LukeWood · 2022-05-25T22:57:00Z

@qlzh727 Just an extra note, as you have already seen in tensorflow/tensorflow#56242

We need to extend this analysis also to the Keras repo for: https://github.com/keras-team/keras/blob/master/keras/layers/preprocessing/image_preprocessing.py

We could write a script to iterate the symbols in the keras_cv.layers.preprocessing namespace. That should cover all KPLs from keras core and keras cv as we alias the core ones there

bhack · 2022-05-25T23:07:55Z

We could write a script to iterate the symbols in the keras_cv.layers.preprocessing namespace. That should cover all KPLs from keras core and keras cv as we alias the core ones there

Yes, I used some quick and dirty bash command locally to produce these "reports".
A new CI Action it would be ideal so that we have also a detector for new introduced fallbacks in candidate PRs.

(extra) extend the same policy to jit_compile when we have #165 (comment)

This also would be nice to have later when we have the mentioned API in Keras master.

wangpengmit · 2022-07-25T21:48:25Z

CCing @ishark .

Mirandatz · 2022-08-29T18:17:22Z

Does this problem still exist?

bhack · 2022-08-29T18:36:41Z

Does this problem still exist?

Yes, this is the updated list today only for the preprocessing layers

pytest -rP keras_cv/layers/preprocessing/ | grep while_loop | sort | uniq

WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting AdjustContrastv2
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting AdjustHue
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting AdjustSaturation
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting Bitcast
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting CropAndResize
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting DecodeJpeg
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting DepthwiseConv2dNative
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting EncodeJpegVariableQuality
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting HistogramFixedWidth
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting ImageProjectiveTransformV3
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting RandomShuffle
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting RngReadAndSkip
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting StatelessRandomGetKeyCounter
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting StatelessRandomUniformFullIntV2
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting StatelessRandomUniformV2

Mirandatz · 2023-02-02T03:38:33Z

If you, like me, is still facing this issue, my "workaround" is:

pip uninstall keras -y
pip install keras==2.8.0

LukeWood · 2023-02-02T04:17:37Z

Hey @Mirandatz - we rolled back changes to the KPLs in the most recent keras release, and we'll be fixing them forwards in KerasCV! Sorry for any inconvenience.

Mirandatz · 2023-02-04T16:29:01Z

@LukeWood

I tried the most recent version of TensorFlow, using the docker image they provide, and the issue still there.

>>> tf.__version__
'2.11.0'
>>> tf.keras.__version__
'2.11.0'

LukeWood · 2023-02-08T05:19:19Z

x

@LukeWood Thank you for the recommendation. I rewrote my augmentation following your recommendation to create the layers outside of the function, as shown below, but I am still seeing the 9 minute epochs, versus the 21 second epochs I expect.

def build_model():
    backbone = tf.keras.applications.efficientnet_v2.EfficientNetV2B0(
        weights='imagenet', input_shape=IMAGE_SIZE, include_top=False)
    backbone.trainable = False

    ## Input 1

    old_inputs = layers.Input(shape=IMAGE_SIZE, name='input_oldest')
    # First input for 'data_a'
    inputs = scaler(old_inputs)
    # inputs = augment_inputs(inputs)
    inputs = tf.keras.layers.RandomZoom(height_factor=(-0.20, 0))(inputs)
    inputs = tf.keras.layers.RandomFlip()(inputs)
    inputs = tf.keras.layers.RandomRotation(factor=0.1)(inputs)
    inputs = tf.keras.layers.RandomTranslation(height_factor=0.1,
                                               width_factor=0.1)(inputs)

    oldest_outputs = tf.keras.layers.GlobalAveragePooling2D()(backbone(inputs, training=False))
    #... rest of the model

    return model, backbone


model, backbone = build_model()
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.01),
              loss=tf.keras.losses.BinaryCrossentropy(),
              metrics=METRICS)

model.fit(train,
          validation_data=validation,
          epochs=50

latest release should fix this in keras core, working on a fix in KerasCV

LukeWood · 2023-02-09T02:13:16Z

Does this problem still exist?

Yes, this is the updated list today only for the preprocessing layers

pytest -rP keras_cv/layers/preprocessing/ | grep while_loop | sort | uniq

WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting AdjustContrastv2
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting AdjustHue
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting AdjustSaturation
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting Bitcast
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting CropAndResize
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting DecodeJpeg
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting DepthwiseConv2dNative
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting EncodeJpegVariableQuality
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting HistogramFixedWidth
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting ImageProjectiveTransformV3
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting RandomShuffle
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting RngReadAndSkip
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting StatelessRandomGetKeyCounter
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting StatelessRandomUniformFullIntV2
WARNING  tensorflow:pfor.py:1081 Using a while_loop for converting StatelessRandomUniformV2

@bhack any way to produce a list of files generating these warnings? I will open tickets for these when we find all of the ones that need to be vectorized (like #1378)

bhack · 2023-02-09T02:20:50Z

Mhh, is the warning I have introduced in Tensorflow still controllable now?

If yes It seems that you need to enable the warn param and re-execute Keras-CV tests

bhack · 2023-02-09T02:38:31Z

About the filename you need to play with the regular expression to keep also the filename as it was in the output (in the case you want to automate that summary).

LukeWood · 2023-02-09T03:56:19Z

Mhh, is the warning I have introduced in Tensorflow still controllable now?

If yes It seems that you need to enable the warn param and re-execute Keras-CV tests

Awesome, yeah we can find layers we need to vectorize and disable the warning for layers that we can’t.

bhack · 2023-02-09T11:33:14Z

Awesome, yeah we can find layers we need to vectorize and disable the warning for layers that we can’t.

It would be nice if you will update/post the full list to have an complete fresh overview.

As I think Pfor converters are quite trivial/fast to contribute.

Then we will disabled the warning.

bhack · 2023-02-09T20:26:39Z

The updated list on master is:

WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting Bitcast cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting CropAndResize cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting DecodeJpeg cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting EncodeJpegVariableQuality cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting ImageProjectiveTransformV3 cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting RandomShuffle cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting RngReadAndSkip cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting StatelessRandomUniformV2 cause there is no registered converter for this op.

bhack · 2023-02-09T21:13:56Z

I don't know why but something was suppressed with only the warn so I needed to disable the fallback.

Here the list of params we to randomize over the batch but the op API has no support for the specific batched argument:

Here are the pfor conversion stack traces: Input "contrast_factor" of op 'AdjustContrastv2' expected to be loop invariant.
Here are the pfor conversion stack traces: Input "delta" of op 'AdjustHue' expected to be loop invariant.
Here are the pfor conversion stack traces: Input "filter" of op 'DepthwiseConv2dNative' expected to be not loop invariant.
Here are the pfor conversion stack traces: Input "scale" of op 'AdjustSaturation' expected to be loop invariant.

sebastian-sz · 2023-02-18T07:15:48Z

@bhack

The updated list on master is:

It seems like this list can be reproduced by just running:
pytest -rP keras_cv/layers/preprocessing/with_mixed_precision_test.py | grep while_loop | sort | uniq
or with_labels_test.py which is a a bit quicker.

Here is an analysis of each layer in with_mixed_precision_test.py and what fallbacks it reports:

ChannelShuffle:
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting RandomShuffle cause there is no registered converter for this op.

RandomCropAndResize:
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting CropAndResize cause there is no registered converter for this op.

RandomlyZoomedCrop
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting ImageProjectiveTransformV3 cause there is no registered converter for this op.

RandomTranslation
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting Bitcast cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting ImageProjectiveTransformV3 cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting RngReadAndSkip cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting StatelessRandomUniformV2 cause there is no registered converter for this op.

RandomJpegQuality
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting DecodeJpeg cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting EncodeJpegVariableQuality cause there is no registered converter for this op.

RandomShear
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting ImageProjectiveTransformV3 cause there is no registered converter for this op.

Mosaic 
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting CropAndResize cause there is no registered converter for this op.

RandomZoom
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting Bitcast cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting ImageProjectiveTransformV3 cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting RngReadAndSkip cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting StatelessRandomUniformV2 cause there is no registered converter for this op.

There might be more layers if uncovered by with_mixed_precision_test.py.

bhack · 2023-02-18T13:05:25Z

You can hack a file in your TF installation dir to check, if covering the converter, it doesn't trigger new issues https://github.com/tensorflow/tensorflow/blob/fc4ae9328b54a71da4954aa8ff75792f0352f1a7/tensorflow/python/ops/parallel_for/pfor.py#L1725:L1726

LukeWood · 2023-02-18T19:43:07Z

@bhack

The updated list on master is:

It seems like this list can be reproduced by just running: pytest -rP keras_cv/layers/preprocessing/with_mixed_precision_test.py | grep while_loop | sort | uniq or with_labels_test.py which is a a bit quicker.

Here is an analysis of each layer in with_mixed_precision_test.py and what fallbacks it reports:

ChannelShuffle:
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting RandomShuffle cause there is no registered converter for this op.

RandomCropAndResize:
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting CropAndResize cause there is no registered converter for this op.

RandomlyZoomedCrop
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting ImageProjectiveTransformV3 cause there is no registered converter for this op.

RandomTranslation
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting Bitcast cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting ImageProjectiveTransformV3 cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting RngReadAndSkip cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting StatelessRandomUniformV2 cause there is no registered converter for this op.

RandomJpegQuality
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting DecodeJpeg cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting EncodeJpegVariableQuality cause there is no registered converter for this op.

RandomShear
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting ImageProjectiveTransformV3 cause there is no registered converter for this op.

Mosaic 
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting CropAndResize cause there is no registered converter for this op.

RandomZoom
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting Bitcast cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting ImageProjectiveTransformV3 cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting RngReadAndSkip cause there is no registered converter for this op.
WARNING  tensorflow:pfor.py:1084 Using a while_loop for converting StatelessRandomUniformV2 cause there is no registered converter for this op.

There might be more layers if uncovered by with_mixed_precision_test.py.

thanks for the comprehensive list! Do you want to open issues to manually vectorize the remaining layers?

if you do please link a template PR (any of the previous ones with a benchmark test should suffice) If not I’m happy to do so!

sebastian-sz · 2023-02-19T09:59:23Z

@LukeWood

Do you want to open issues to manually vectorize the remaining layers?

I won't be able to contribute to this in the nearest time :(

LukeWood · 2023-02-19T20:53:52Z

@LukeWood

Do you want to open issues to manually vectorize the remaining layers?

I won't be able to contribute to this in the nearest time :(

Sounds good - I’ll open a list for the remaining layers. Thanks for your hard work on the is effort @sebastian-sz!

james77777778 · 2023-02-20T10:05:07Z

I'm willing to vectorize some of the remaining layers.

james77777778 · 2023-03-10T06:42:01Z

Some updates (based on @sebastian-sz #291 (comment))

ChannelShuffle: Vectorize ChannelShuffle #1433 (should be ready to merge?)
RandomCropAndResize: Vectorize CropAndResize #1439
RandomlyZoomedCrop: Vectorize RandomlyZoomedCrop #1499
RandomTranslation: Vectorize RandomTranslation #1480
RandomJpegQuality
I think this is a hard one...
RandomShear Vectorized Random Shear. #1537
Mosaic Vectorize Mosaic #1555
RandomZoom: Vectorize RandomZoom #1435

I will try to vectorize some of them when I'm available

LukeWood · 2023-04-27T06:06:20Z

Woop woop - @james77777778 you've absolutely crushed this issue! Thanks for your hard work on this!

…am#291) * Added testing * Readd traceback * Format * Format * Update layer.py

bhack mentioned this issue Apr 15, 2022

tf.vectorized_maps resolve fallbacks tensorflow/tensorflow#55639

Closed

bhack mentioned this issue Apr 18, 2022

LookupError: gradient registry has no entry for RandomShuffle #259

Closed

copybara-service bot pushed a commit to tensorflow/tensorflow that referenced this issue Apr 23, 2022

Expose a warn argument in vectorized_map to mute fallback to whil…

88a263e

…e conversions. Reference issues: keras-team/keras-cv#264, keras-team/keras-cv#291 PiperOrigin-RevId: 443809170

This was referenced Apr 27, 2022

Repeated augmentation layer #372

Closed

Update the benchmarking script to accept an arbitrary BaseImageAugmentationLayer subclass #361

Closed

bhack mentioned this issue Apr 30, 2022

[AutoAugment] Add Rotate operation. #402

Closed

bhack mentioned this issue May 16, 2022

Fix ChannelShuffle #418

Closed

bhack mentioned this issue May 24, 2022

Model trains extremely slow after upgrade to tf 2.9 tensorflow/tensorflow#56242

Closed

bhack mentioned this issue May 25, 2022

Update performance benchmarking script to include a layer with ops that rely on fallbacks for autovectorization #455

Closed

qlzh727 assigned qlzh727 and LukeWood May 25, 2022

bhack mentioned this issue May 27, 2022

jit_compile control in BaseImageAugmentationLayer keras-team/keras-io#1541

Open

bhack mentioned this issue Jul 13, 2022

BUG: RandomContrast slows down training 6-fold #581

Closed

bhack mentioned this issue Oct 25, 2022

WARNING:tensorflow:Using a while_loop tensorflow/tensorflow#58296

Closed

tanzhenyu added the stat:awaiting keras-team label Oct 27, 2022

bhack mentioned this issue Dec 24, 2022

investigate the impact of changing arguments in map_fn/vmap for preprocessing layers #595

Closed

bhack mentioned this issue Feb 7, 2023

Implement vectorized base image augmentation layer w/ Grayscale() example #1331

Closed

LukeWood changed the title ~~tf.vectorized_map fallback cause~~ vectorize KPLS: tf.vectorized_map fallback cause Feb 8, 2023

bhack mentioned this issue Feb 14, 2023

Added vectorized Auto Contrast implementation. #1394

Merged

5 tasks

james77777778 mentioned this issue Mar 10, 2023

Vectorize RandomTranslation #1480

Merged

5 tasks

This was referenced Mar 14, 2023

Vectorize RandomlyZoomedCrop #1499

Merged

Vectorize Mosaic #1555

Merged

freedomtan pushed a commit to freedomtan/keras-cv that referenced this issue Jul 20, 2023

Allow training in any layer call instead of erroring out. (keras-te…

1bd0fba

…am#291) * Added testing * Readd traceback * Format * Format * Update layer.py

ianstenbit closed this as completed Aug 4, 2023

vectorize KPLS: tf.vectorized_map fallback cause #291

vectorize KPLS: tf.vectorized_map fallback cause #291

Comments

bhack commented Apr 10, 2022 • edited Loading

bhack commented Apr 10, 2022

bhack commented Apr 11, 2022 • edited Loading

bhack commented Apr 11, 2022 • edited Loading

bhack commented Apr 13, 2022 • edited Loading

bhack commented Apr 15, 2022

wangpengmit commented Apr 15, 2022

bhack commented Apr 15, 2022 • edited Loading

wangpengmit commented Apr 15, 2022

bhack commented Apr 15, 2022

bhack commented Apr 28, 2022

bhack commented May 24, 2022

qlzh727 commented May 25, 2022

LukeWood commented May 25, 2022

bhack commented May 25, 2022 • edited Loading

wangpengmit commented Jul 25, 2022

Mirandatz commented Aug 29, 2022

bhack commented Aug 29, 2022

Mirandatz commented Feb 2, 2023

LukeWood commented Feb 2, 2023

Mirandatz commented Feb 4, 2023 • edited Loading

LukeWood commented Feb 8, 2023

LukeWood commented Feb 9, 2023

bhack commented Feb 9, 2023 • edited Loading

bhack commented Feb 9, 2023

LukeWood commented Feb 9, 2023

bhack commented Feb 9, 2023

bhack commented Feb 9, 2023

bhack commented Feb 9, 2023 • edited Loading

sebastian-sz commented Feb 18, 2023

bhack commented Feb 18, 2023 • edited Loading

LukeWood commented Feb 18, 2023

sebastian-sz commented Feb 19, 2023

LukeWood commented Feb 19, 2023 • edited Loading

james77777778 commented Feb 20, 2023

james77777778 commented Mar 10, 2023 • edited Loading

LukeWood commented Apr 27, 2023

vectorize KPLS: `tf.vectorized_map` fallback cause #291

vectorize KPLS: `tf.vectorized_map` fallback cause #291

bhack commented Apr 10, 2022 •

edited

Loading

bhack commented Apr 11, 2022 •

edited

Loading

bhack commented Apr 11, 2022 •

edited

Loading

bhack commented Apr 13, 2022 •

edited

Loading

bhack commented Apr 15, 2022 •

edited

Loading

bhack commented May 25, 2022 •

edited

Loading

Mirandatz commented Feb 4, 2023 •

edited

Loading

bhack commented Feb 9, 2023 •

edited

Loading

bhack commented Feb 9, 2023 •

edited

Loading

bhack commented Feb 18, 2023 •

edited

Loading

LukeWood commented Feb 19, 2023 •

edited

Loading

james77777778 commented Mar 10, 2023 •

edited

Loading