33 Pull requests merged by 7 people

• navi4 support and rm hard coding on gfx

  #2897 merged Mar 20, 2025

• Fix distributed save test 2.16

  #2880 merged Mar 17, 2025

• Increase timeout for multi_process_runner_test subtest

  #2893 merged Mar 17, 2025

• Fix failure_handler_test

  #2892 merged Mar 17, 2025

• Fix nccl distributed test by allowing multiple ncclmanager instances 2.19

  #2881 merged Mar 17, 2025

• skip test if is tpu strategy

  #2891 merged Mar 17, 2025

• Develop upstream sync 250310

  #2888 merged Mar 16, 2025

• Exclude //tensorflow/dtensor/python/tests:multi_client_test_nccl_local_2gpus from single gpu test suite

  #2890 merged Mar 13, 2025

• Fix distributed_save_test 2.18

  #2879 merged Mar 13, 2025

• Fix distributed save test 2.19

  #2886 merged Mar 13, 2025

• Enable previously disabled tests

  #2887 merged Mar 12, 2025

• Add release branch creation process document

  #2860 merged Mar 12, 2025

• Fix nccl distributed test by allowing multiple ncclmanager instances 2.17

  #2884 merged Mar 12, 2025

• [ROCm] Use kernel cache per stream executor for redzone checker

  #2877 merged Mar 11, 2025

• R2.18 redzone kernel init blas fix

  #2875 merged Mar 11, 2025

• Dev upst qa rocm64 redzone init blas fix

  #2876 merged Mar 11, 2025

• Fix nccl distributed test by allowing multiple ncclmanager instances 2.18

  #2883 merged Mar 11, 2025

• Remove irrelevant checks

  #2869 merged Mar 11, 2025

• [ROCm] Enable unsafe fp atomics and cleanup gpu_device_functions.h

  #2854 merged Mar 10, 2025

• Develop upstream sync 250221

  #2867 merged Mar 10, 2025

• [ROCM] Fix vector bias add fusion into BLAS call

  #2866 merged Mar 7, 2025

• [ROCm] Enable unsafe fp atomics and cleanup gpu_device_functions.h

  #2853 merged Mar 7, 2025

• [ROCm] Fix multi process runner test

  #2874 merged Mar 6, 2025

• revert spack related changes

  #2871 merged Mar 5, 2025

• fix distributed save test upstream

  #2872 merged Mar 5, 2025

• Restore Dockerfile.rocm symlink

  #2868 merged Mar 4, 2025

• Fix distributed_save_test

  #2862 merged Mar 4, 2025

• Remove gfx940 and gfx941

  #2865 merged Feb 28, 2025

• Remove gfx940 and gfx941

  #2864 merged Feb 28, 2025

• Remove gfx940 and gfx941

  #2863 merged Feb 28, 2025

• Remove gfx940 and gfx941

  #2861 merged Feb 28, 2025

• [ROCm] Enable unsafe fp atomics and cleanup gpu_device_functions.h

  #2852 merged Feb 25, 2025

• [ROCm] Enable unsafe fp atomics and cleanup gpu_device_functions.h

  #2851 merged Feb 25, 2025

3 Pull requests opened by 2 people

• Enable nccl_ops_test

  #2889 opened Mar 13, 2025

• Develop upstream sync 250318

  #2896 opened Mar 19, 2025

• Add dummy file

  #2898 opened Mar 20, 2025

12 Issues closed by 2 people

• 8x Rx600 multiple gpu in a BTC T37 motherboard

  #2126 closed Mar 11, 2025

• Undeclared inclusions building from source

  #1467 closed Mar 11, 2025

• Only first prediction of model returns the expected

  #1331 closed Mar 11, 2025

• model.fit hangs after a few hundred/thousand epochs

  #1123 closed Mar 11, 2025

• compilation error of tensorflow-rocm

  #849 closed Mar 11, 2025

• Tensorflow Probability bayesian_neural_network hits NaNs

  #303 closed Mar 11, 2025

• should I expect performance increase using mixed_precision.Policy('mixed_float16') with radeon7?

  #827 closed Mar 11, 2025

• Inconsistent results with ROCm/Radeon VII

  #622 closed Mar 11, 2025

• Problem with different calculation results on RadeonVII(gfx906)+ROCm2.4

  #449 closed Mar 11, 2025

• Memory Access Faults during training

  #414 closed Mar 11, 2025

• Inport Error librocm_smi64.so.5 with ROCm 5.0.2

  #1600 closed Mar 4, 2025

• tensorflow.python.framework.errors_impl.InternalError: Failed copying input tensor from /job:localhost/replica:0/task:0/device:CPU:0 to /job:localhost/replica:0/task:0/device:GPU:0 in order to run _EagerConst: Dst tensor is not initialized.

  #1469 closed Feb 25, 2025

1 Issue opened by 1 person

• Observing ValueError for multiple TF_CPU tests with r2.18_3.10 docker

  #2894 opened Mar 18, 2025

6 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Build instruction is outdated

  #2858 commented on Mar 10, 2025 • 0 new comments

• NotFoundError: /usr/local/lib/python3.6/dist-packages/tensorflow_text/python/metrics/_text_similarity_metric_ops.so: undefined symbol: _ZNK10tensorflow8OpKernel11TraceStringERKNS_15OpKernelContextEb

  #1455 commented on Mar 11, 2025 • 0 new comments

• ROCm Ops Lists

  #1634 commented on Mar 11, 2025 • 0 new comments

• RNN padding support in tensorflow as MIopen already support RNN padding in rocm6.0

  #2582 commented on Mar 18, 2025 • 0 new comments

• [ROCm] Enable unsafe fp atomics and cleanup gpu_device_functions.h

  #2855 commented on Mar 6, 2025 • 0 new comments

• amd bf16 gpu float support

  #2859 commented on Feb 26, 2025 • 0 new comments