TensorFlow Determinism

This repository serves three purposes:

1. Provide up-to-date information (in this file) about non-determinism sources and solutions in TensorFlow and beyond, with a focus on determinism when running on GPUs.
2. Provide a patch to attain various levels of GPU-specific determinism in stock TensorFlow, via the installation of the tensorflow-determinism pip package.
3. Be the location where a TensorFlow determinism debug tool will be released as part of the tensorflow-determinism pip package.

For more information, please watch the video of the GTC 2019 talk Determinism in Deep Learning. The desciption under that video also includes links to the slides from the talk and to a poster presentation on this topic.

Installation

Use pip to install:

pip install tensorflow-determinism

This will install a package that can be imported as tfdeterminism. The installation of tensorflow-determinism will not automatically install TensorFlow. The intention of this is to allow you to install your chosen version of TensorFlow. You will need to install your chosen version of TensorFlow before you can import and use tfdeterminism.

Deterministic TensorFlow Solutions

There are currently two main ways to access GPU-deterministic functionality in TensorFlow for most deep learning applications. The first way is to use an NVIDIA NGC TensorFlow container. The second way is to use version 1.14 or 1.15 of stock TensorFlow with GPU support, plus the application of a patch supplied in this repo. Patch support for version 2.0 of stock TensorFlow is currently in development.

The longer-term intention and plan is to upstream all solutions into stock TensorFlow.

NVIDIA NGC TensorFlow Containers

NGC TensorFlow containers, starting with version 19.06, implement GPU-deterministic TensorFlow functionality. In Python code running inside the container, this can be enabled as follows:

import tensorflow as tf
import os
os.environ['TF_DETERMINISTIC_OPS'] = '1'
# Now build your graph and train it

The following table shows which version of TensorFlow each NGC container version is based on:

NGC Container Version	TensorFlow Version
19.06	1.13
19.07 - 19.09	1.14

For information about pulling and running the NVIDIA NGC containers, see these instructions.

Stock TensorFlow

Versions 1.14 and 1.15 of stock TensorFlow implement a reduced form of GPU determinism, which must be supplemented with a patch provided in this repo. The following Python code is running on a machine in which pip package tensorflow-gpu=1.14.0 has been installed correctly and on which tensorflow-determinism has also been installed (as shown in the installation section above).

import tensorflow as tf
from tfdeterminism import patch
patch()
# build your graph and train it

Stock TensorFlow with GPU support can be installed as follows:

pip install tensorflow-gpu=1.14.0

The TensorFlow project includes detailed instructions for installing TensorFlow with GPU support.

Additional Ingredients in the Determinism Recipe

You'll also need to set any and all appropriate random seeds:

os.environ['PYTHONHASHSEED']=str(SEED)
random.seed(SEED)
np.random.seed(SEED)
tf.set_random_seed(SEED)

If you're using Horovod for multi-GPU training, you may need to disable Tensor Fusion (assuming that the non-determinism associated with Tensor Fusion has not yet been resolved):

os.environ['HOROVOD_FUSION_THRESHOLD']='0'

Detailed Status of Determinism in TensorFlow and Beyond

Confirmed and likely sources of non-determinism, along with any existing solutions, are being tracked here.

GPU-Specific Sources of Non-Determinism

Historic GPU-Specific Sources of Non-Determinism

In the past, tf.math.reduce_sum and tf.math.reduce_mean operated non-deterministically when running on a GPU. This was resolved before TensorFlow version 1.12. These ops now function deterministically by default when running on a GPU.

Confirmed Current GPU-Specific Sources of Non-Determinism (With Solutions)

Source	NGC 19.06+	TF 1.14+	TF 2.0
TF auto-tuning of cuDNN convolution algorithms	TCD or TDO	TCD or TDP	TCD
cuDNN convolution backprop to weight gradients	TCD or TDO	TCD or TDP	TCD
cuDNN convolution backprop to data gradients	TCD or TDO	TCD or TDP	TCD
cuDNN max-pooling backprop	TCD or TDO	TCD or TDP	TCD
tf.nn.bias_add backprop	TDO	TDP	NS2
tf.image.resize_bilinear fwd and bwd	NS1	NS1	NS1

Key to the solutions refenced above:

Solution	Description
TCD	Set environment variable TF_CUDNN_DETERMINISTIC to '1' or 'true'. Also do not set environment variable TF_USE_CUDNN_AUTOTUNE at all (and particularly do not set it to '0' or 'false').
TDO	Set environment variable TF_DETERMINISTIC_OPS to '1' or 'true'. Also do not set environment variable TF_USE_CUDNN_AUTOTUNE at all (and particularly do not set it to '0' or 'false').
TDP	Apply tfdeterminism.patch. Note that we are currently working on getting solution TDO into stock TensorFlow (see PR 31465).
NS1	There is currently no solution available for this, but one is under development.
NS2	The patch, TDP (see above), is currently being updated so that it can be applied to TF version 2.0.

Other Possible GPU-Specific Sources of Non-Determinism

Going beyond the above-mentioned sources, in version 1.12 of TensorFlow (and also in the master branch on 2019-03-03, afer release 1.31.1), the following files call CUDA atomicAdd either directly or indirectly. This makes them candidates for the injection of non-determinism.

• crop_and_resize_op_gpu.cu.cc
• scatter_functor_gpu.cu.h
• scatter_nd_op_gpu.cu.cc
• sparse_tensor_dense_matmul_op_gpu.cu.cc
• resize_nearest_neighbor_op_gpu.cu.cc
• segment_reduction_ops.h
• segment_reduction_ops_gpu.cu.cc
• dilation_ops_gpu.cu.cc
• maxpooling_op_gpu.cu.cc
• svd_op_gpu.cu.cc
• cuda_kernel_helper_test.cu.cc
• depthwise_conv_op_gpu.h
• resampler_ops_gpu.cu.cc
• histogram_op_gpu.cu.cc
• stateful_random_ops_gpu.cu.cc

Unless you are using TensorFlow ops that depend on these files (i.e. ops with similar names), then your model will not be affected by these potential sources of non-determinism.

Beyond atomicAdd, there are ten other CUDA atomic functions whose use could lead to the injection of non-determinism, such as atomicCAS (the most generic, atomic compare and swap). Note also that the word 'atomic' was present in 167 files in the TensorFlow repo and some of these may be related to the use of CUDA atomic operations. It's important to remember that it's possible to use CUDA atomic operations without injecting non-determinism, and that, therefore, when CUDA atomic operations are present in op code, it doesn't guarantee that the op injects non-determinism into the computation.

Sources of Non-Determinism in TensorFlow Unrelated to GPU

• Issue 29101: Random seed not set in graph context of Dataset#map. This may have been resolved in version 1.14 of TensorFlow.
• tf.data.Dataset with more than one worker. The work-around is to use only one worker.

Sources of Non-Determinism Beyond TensorFlow

• TensorRT timing-based kernel schedule. Each time an inference engine is generated, it could be slightly different, particularly if there is varying load on the machine used to run TensorRT. There is a solution planned for this.
• Horovod Tensor Fusion. Work-around: disable Tensor Fusion by setting the environment variable HOROVOD_FUSION_THRESHOLD to '0'. This issue may have been resolved by Horovod pull-request 1130 (not yet confirmed).

Relevant Links

This section catalogs relevant links.

TensorFlow Issues

Number	Title	Updated
2652	Backward pass of broadcasting on GPU is non-deterministic	2019-10-08
2732	Mention that GPU reductions are nondeterministic in docs	2019-10-08
13932	Non-determinism from tf.data.Dataset.map with random ops
16889	Problems Getting TensorFlow to behave Deterministically	2019-10-08
18096	Feature Request: Support for configuring deterministic options of cuDNN conv routines	2019-10-08
29101	Random seed not set in graph context of Dataset#map

TensorFlow Pull Requests

Number	Title	Status	Updated
10636	Non-determinism Docs	closed (not merged)	2019-10-08
24273	Enable dataset.map to respect seeds from the outer context	closed (not merged)
24747	Add cuDNN deterministic env variable (only for convolution).	merged into master
25269	Add deterministic cuDNN max-pooling	merged into master
25796	Added tests for TF_CUDNN_DETERMINISTIC	merged into master
29667	Add release note about TF_CUDNN_DETERMINISTIC	merged into r1.14
31389	Enhance release notes related to TF_CUDNN_DETERMINISTIC	merged into r1.14
31465	Add GPU-deterministic tf.nn.bias_add	ready to pull
32979	Fix typo in release note	awaiting review

Miscellaneous

• Two Sigma: A Workaround for Non-Determinism in TensorFlow
• Keras issue 12800: Unable to get reproducible results using Keras with TF backend on GPU (updated on 2019-10-08)
• PyTorch Reproducibility (from the official documentation)
• Chainer PR 2710: cuDNN Deterministic mode
• Stack Overflow: Tensorflow: Different results with the same random seed
• Stack Overflow: Are tensorflow random values guaranteed to be the same inside a single run? (comment) (updated 2019-10-10).