NVIDIA/framework-reproducibility

Issues

• Changing class name/structure changes program functionality (using TensorFlow)

  #40 opened a year ago by edwardyehuang
  2

• Determinism across GPU architectures?

  #28 opened 4 years ago by isarandi
  5

• Non-reproducible model training results with TensorFlow 2.5

  #38 opened 2 years ago by brandhsu
  7

• `torch_status.md` determinism tracker similiar to `tensorflow_status.md`

  #41 opened 2 years ago by h-grieve
  1

• tf.data.experimental.sample_from_datasets non-deterministic in multi-gpu.

  #39 opened 2 years ago by yanniskar
  4

• Model nondeterminism when using `K.layers.DepthwiseConv2D` / `tf.nn.depthwise_conv2d` (known issue)

  #26 opened 4 years ago by kimzt
  18

• Reproducibility issue with transformers (BERT) and tf2.2

  #19 opened 4 years ago by MFreidank
  21

• Message passing neural network determinism thwarted by tf.math.segment_sum and tf.gather

  #25 opened 4 years ago by dangthatsright
  13

• Early versions of TensorFlow (e.g. 1.9) do not have a version attribute

  #37 opened 3 years ago by SageAgastya
  3

• Usage of numpy.random.Generator to deal with Data-Loader Parallelism

  #36 opened 3 years ago by zakajd
  7

• patch tf.image.resize with bilinear by casting image to tf.float64

  #32 opened 3 years ago by llan-ml
  2

• Don't get deterministic results with NGC 20.09

  #29 opened 4 years ago by ornellamarciano
  8

• garder14/byol-tensorflow2 (batch-norm & softmax/cross-entropy)

  #35 opened 3 years ago by evolu8
  1

• Nondeterminism from tf.image.crop_and_resize

  #18 opened 4 years ago by xthyax
  2

• More segment ops need to be patched

  #31 opened 3 years ago by llan-ml
  5

• Model not deterministic/reproducible with all seeds set and os.environ['TF_DETERMINISTIC_OPS'] = '1'

  #30 opened 4 years ago by stevenwong951028
  0

• Exception thrown with "No algorithm worked!" message on NGC 20.09

  #27 opened 4 years ago by disembarrasing
  13

• Why tensorflow-determinism and not simply determinism?

  #4 opened 4 years ago by kirk86
  4

• non-determinism on tf.keras.layers.UpSampling2D(..., interpolation='nearest')

  #24 opened 4 years ago by emiliocoutinho
  3

• How can I get deterministic, back-propagatable upsampling?

  #12 opened 4 years ago by louislbc
  5

• Running on stock TensorFlow version >= 2.1

  #22 opened 4 years ago by emiliocoutinho
  5

• Make tf_determinism.patch() handle tf>=2.1 as well

  #20 opened 4 years ago by MFreidank
  4

• Lack of reproducibility when using Huggingface transformers library (TensorFlow version)

  #14 opened 4 years ago by dmitriydligach
  11

• Getting OpenNMT-tf to train reproducibly

  #9 opened 4 years ago by atebbifakhr
  13

• Will this package work on tf.keras model?

  #13 opened 4 years ago by leocnj
  1

• Raise more specific exceptions

  #7 opened 4 years ago by bersbersbers
  3

• Unidentified model not training reproducibly

  #3 opened 4 years ago by invoxiaehu
  8

• When will the debug tool be available?

  #1 opened 4 years ago by daddydrac
  2

• How can I confirm that the deterministic environment variable is working?

  #2 opened 4 years ago by cklsoft
  5

• Tensorflow-determinism not working with conda tensorflow-gpu

  #5 opened 4 years ago by Eugen2525
  6

• How can I check whether I'm in a NGC TensorFlow container?

  #6 opened 4 years ago by AndreaPi
  2

• Are there disadvantages to enabling deterministic ops?

  #10 opened 4 years ago by lminer
  2

• Support for TF 2.1.0

  #8 opened 4 years ago by jmwoloso
  3