Search results

XLA tf.random.stateless_uniform produces different values from eager and graph for the same seed

type:bugBugBug

Status: Open.

#121243 In tensorflow/tensorflow;

· laolvfan opened on Jun 15, 2026

XLA tf.cast from finite out-of-range float32 to uint8 differs from eager and graph execution

type:bugBugBug

Status: Open.

#121242 In tensorflow/tensorflow;

· laolvfan opened on Jun 15, 2026

XLA fails to compile tf.random.shuffle on rank-1 floating tensors due to invalid TOTALORDER comparison on s32 keys

2.21.0tensorflow 2.21.0tensorflow 2.21.0

comp:xlaXLAXLA

type:bugBugBug

Status: Open.

#121232 In tensorflow/tensorflow;

· laolvfan opened on Jun 15, 2026

Misaligned flatbuffer Vector access in TFLite FlatBufferModelBase via the unverified build path

comp:liteTF Lite related issuesTF Lite related issues

type:bugBugBug

Status: Open.

#121217 In tensorflow/tensorflow;

· professor-moody opened on Jun 14, 2026

TFLite ParseExample crashes for empty serialized batch with sparse outputs

2.21.0tensorflow 2.21.0tensorflow 2.21.0

awaiting PR mergeawaiting PR mergeawaiting PR merge

comp:liteTF Lite related issuesTF Lite related issues

type:bugBugBug

Status: Open.

#121166 In tensorflow/tensorflow;

· ChiragSW opened on Jun 13, 2026

TensorFlow PyPI aarch64 wheels aren't built with CUDA

subtype: ubuntu/linuxUbuntu/Linux Build/Installation IssuesUbuntu/Linux Build/Installation Issues

type:build/installBuild and install issuesBuild and install issues

Status: Open.

#121157 In tensorflow/tensorflow;

· VRehnberg opened on Jun 13, 2026

Aether Bridge API - Autonomous Agent Team Management for TensorFlow Apps

Status: Open.

#121032 In tensorflow/tensorflow;

· atomeam opened on Jun 12, 2026

Non-extreme BF16 AdamW lr=0.05 still reproduces execution-path drift under AUTO all-reduce

comp:dist-stratDistribution Strategy related issuesDistribution Strategy related issues

TF 2.18

type:bugBugBug

Status: Open.

#121003 In tensorflow/tensorflow;

· whitephilomel opened on Jun 12, 2026

DATA-sharded RING SGD at lr=0.5 produces different loss and gradient endpoints from the same synthetic batches

comp:dist-stratDistribution Strategy related issuesDistribution Strategy related issues

TF 2.18

type:bugBugBug

Status: Open.

#121002 In tensorflow/tensorflow;

· whitephilomel opened on Jun 12, 2026

AdamW cosine schedule at lr=1.0 takes a separate RING loss trajectory without changing the input tensors

comp:dist-stratDistribution Strategy related issuesDistribution Strategy related issues

TF 2.18

type:bugBugBug

Status: Open.

#121001 In tensorflow/tensorflow;

· whitephilomel opened on Jun 12, 2026

AutoShard disabled in float32 SGD: accumulated gradients disagree after mirrored GPU execution

comp:dist-stratDistribution Strategy related issuesDistribution Strategy related issues

TF 2.18

type:bugBugBug

Status: Open.

#121000 In tensorflow/tensorflow;

· whitephilomel opened on Jun 12, 2026

RING plus DATA sharding plus BF16 AdamW surfaces a reduced-gradient discrepancy during accumulated updates

comp:dist-stratDistribution Strategy related issuesDistribution Strategy related issues

TF 2.18

type:bugBugBug

Status: Open.

#120999 In tensorflow/tensorflow;

· whitephilomel opened on Jun 12, 2026