NVIDIA/framework-reproducibility
Providing reproducibility in deep learning frameworks
PythonApache-2.0
Issues
- 2
Changing class name/structure changes program functionality (using TensorFlow)
#40 opened by edwardyehuang - 5
Determinism across GPU architectures?
#28 opened by isarandi - 7
- 1
- 4
- 18
Model nondeterminism when using `K.layers.DepthwiseConv2D` / `tf.nn.depthwise_conv2d` (known issue)
#26 opened by kimzt - 21
- 13
Message passing neural network determinism thwarted by tf.math.segment_sum and tf.gather
#25 opened by dangthatsright - 3
- 7
- 2
- 8
- 1
- 2
Nondeterminism from tf.image.crop_and_resize
#18 opened by xthyax - 5
More segment ops need to be patched
#31 opened by llan-ml - 0
Model not deterministic/reproducible with all seeds set and os.environ['TF_DETERMINISTIC_OPS'] = '1'
#30 opened by stevenwong951028 - 13
- 4
- 3
non-determinism on tf.keras.layers.UpSampling2D(..., interpolation='nearest')
#24 opened by emiliocoutinho - 5
- 5
Running on stock TensorFlow version >= 2.1
#22 opened by emiliocoutinho - 4
- 11
Lack of reproducibility when using Huggingface transformers library (TensorFlow version)
#14 opened by dmitriydligach - 13
Getting OpenNMT-tf to train reproducibly
#9 opened by atebbifakhr - 1
Will this package work on tf.keras model?
#13 opened by leocnj - 3
Raise more specific exceptions
#7 opened by bersbersbers - 8
- 2
When will the debug tool be available?
#1 opened by daddydrac - 5
- 6
- 2
- 2
- 3
Support for TF 2.1.0
#8 opened by jmwoloso