Issues · tensorflow/tensorflow · GitHub

Labels Milestones New issue

Bb

#118677

· troyasalvador5-beep opened

on May 16, 2026

Vh

#118676

· troyasalvador5-beep opened

on May 16, 2026

Grappler `ArgMax`/`ArgMin` monotonic stripping is wrong for strictly-monotonic ops with float32 saturation/overflow

#118675

· wuyii8941 opened

on May 16, 2026

XLA gradient of `tf.where` evaluates dead branch, leaking NaN/inf

#118674

· wuyii8941 opened

on May 16, 2026

XLA gradient of `reduce_max`/`reduce_min` silently returns 0 instead of NaN when input contains NaN

#118673

· wuyii8941 opened

on May 16, 2026

XLA `argmax`/`argmin` returns wrong index when NaN is present for non-power-of-2 sizes

#118647

· wuyii8941 opened

on May 16, 2026

GPU `x * tf.math.reciprocal(x)` returns 0 for normal float32 input `x = 1e38`

#118645

· wuyii8941 opened

on May 16, 2026

`tf.math.rsqrt` returns `inf` for subnormal float32 inputs on both CPU and GPU

#118644

· wuyii8941 opened

on May 16, 2026

`tf.math.atan2` returns NaN on CPU for subnormal float32 inputs

#118643

· wuyii8941 opened

on May 16, 2026

XLA `floordiv(x, y) * y + floormod(x, y)` violates division identity for float32

#118642

· wuyii8941 opened

on May 16, 2026

Support for resampling time-series data

#118610

· oliskir opened

on May 15, 2026

Support the use of the third-party KDNN library

#118587

· amaniyadunia opened

on May 15, 2026