Search results

Large CPU/GPU numerical discrepancy in tf.linalg.matmul for float32 linear layer equivalent

type:bugBugBug

Status: Open.

#127346 In tensorflow/tensorflow;

· dingjiawei65 opened on Sep 14, 2026

[Feature Request] Fused Linear Cross-Entropy API in tf.nn for memory-efficient LLM training

comp:apisHighlevel API related issuesHighlevel API related issues

stat:contribution welcomeStatus - Contributions welcomeStatus - Contributions welcome

type:featureFeature requestsFeature requests

Status: Open.

#127290 In tensorflow/tensorflow;

· powerofaisinstudy-debug opened on Sep 12, 2026

tf.keras.metrics.categorical_crossentropy uses wrong class count for label smoothing with axis zero

2.21.0tensorflow 2.21.0tensorflow 2.21.0

comp:kerasKeras related issuesKeras related issues

stat:awaiting responseStatus - Awaiting response from authorStatus - Awaiting response from author

type:bugBugBug

Status: Open.

#127249 In tensorflow/tensorflow;

· ALinrunrun opened on Sep 11, 2026

tf.keras.ops.numpy.rot90 returns incorrect values for two rotations of a rectangular plane

2.21.0tensorflow 2.21.0tensorflow 2.21.0

comp:kerasKeras related issuesKeras related issues

comp:opsOPs related issuesOPs related issues

stat:awaiting responseStatus - Awaiting response from authorStatus - Awaiting response from author

type:bugBugBug

Status: Open.

#127248 In tensorflow/tensorflow;

· ALinrunrun opened on Sep 11, 2026

tf.keras.ops.image.gaussian_blur appears to swap height and width sigma parameters

2.21.0tensorflow 2.21.0tensorflow 2.21.0

comp:opsOPs related issuesOPs related issues

type:bugBugBug

Status: Open.

#127247 In tensorflow/tensorflow;

· ALinrunrun opened on Sep 11, 2026

tf.keras.activations.sparsemax computes incorrect output for a simple three-element input

2.21.0tensorflow 2.21.0tensorflow 2.21.0

comp:kerasKeras related issuesKeras related issues

stat:awaiting responseStatus - Awaiting response from authorStatus - Awaiting response from author

type:bugBugBug

Status: Open.

#127246 In tensorflow/tensorflow;

· ALinrunrun opened on Sep 11, 2026

tf.signal.kaiser_window loses the second derivative at beta zero

2.21.0tensorflow 2.21.0tensorflow 2.21.0

comp:coreissues related to core part of tensorflowissues related to core part of tensorflow

type:bugBugBug

Status: Open.

#127245 In tensorflow/tensorflow;

· ALinrunrun opened on Sep 11, 2026

tf.keras.ops.numpy.ldexp returns incorrect gradient at zero

2.21.0tensorflow 2.21.0tensorflow 2.21.0

awaiting PR mergeawaiting PR mergeawaiting PR merge

comp:opsOPs related issuesOPs related issues

type:bugBugBug

Status: Open.

#127244 In tensorflow/tensorflow;

· ALinrunrun opened on Sep 11, 2026

tf.math.cumulative_logsumexp produces NaN second derivative under nested forward-mode autodiff

2.21.0tensorflow 2.21.0tensorflow 2.21.0

comp:coreissues related to core part of tensorflowissues related to core part of tensorflow

type:bugBugBug

Status: Open.

#127243 In tensorflow/tensorflow;

· ALinrunrun opened on Sep 11, 2026

tf.keras.ops.image.affine_transform returns incorrect image gradient with nearest-neighbor sampling

2.21.0tensorflow 2.21.0tensorflow 2.21.0

comp:opsOPs related issuesOPs related issues

type:bugBugBug

Status: Open.

#127241 In tensorflow/tensorflow;

· ALinrunrun opened on Sep 11, 2026

tf.nn.isotonic_regression returns incorrect gradient for a pooled block

2.21.0tensorflow 2.21.0tensorflow 2.21.0

comp:coreissues related to core part of tensorflowissues related to core part of tensorflow

type:bugBugBug

Status: Open.

#127234 In tensorflow/tensorflow;

· ALinrunrun opened on Sep 11, 2026

tf.math.segment_mean produces NaN JVP for a valid empty segment

2.21.0tensorflow 2.21.0tensorflow 2.21.0

comp:opsOPs related issuesOPs related issues

type:bugBugBug

Status: Open.

#127233 In tensorflow/tensorflow;

· ALinrunrun opened on Sep 11, 2026