Peter Hawkins 0b470361da Change the default jnp.take mode to "fill". ...

Previously, `jnp.take` defaulted to clamping out-of-bounds indices into range. Now, `jnp.take` returns invalid values (e.g., NaN) for out-of-bounds indices. This change attempts to prevent latent bugs caused by inadvertent out-of-bounds indices.

The previous behavior can be approximated using the "clip" or "wrap" fill modes.

PiperOrigin-RevId: 445130143

2022-04-28 06:01:56 -07:00

..

__init__.py

register the lax module for stack trace filtering

2021-02-22 21:11:41 -08:00

ann.py

[MHLO] Remove most XLA translation rules.

2022-04-18 08:28:35 -07:00

control_flow.py

add scan dce rule

2022-04-27 20:47:43 -07:00

convolution.py

[JAX] Delete last references to conv/dot translation rules.

2022-04-18 17:42:47 -07:00

fft.py

[MHLO] Remove most XLA translation rules.

2022-04-18 08:28:35 -07:00

lax.py

Change the default jnp.take mode to "fill".

2022-04-28 06:01:56 -07:00

linalg.py

lax.linalg.qr: allow jvp when m == n and full_matrices=True

2022-04-26 10:34:50 -07:00

other.py

Use real Precision type for lax.PrecisionType

2022-03-04 04:21:25 -08:00

parallel.py

Add unary xeinsum and allow named axis reductions for unary and binary xeinsums

2022-04-26 09:55:42 +00:00

polar.py

Document jax.lax.Precision

2021-11-08 14:15:31 -08:00

qdwh.py

[linalg] Fix type promotion in QDWH.

2022-03-03 20:59:28 -08:00

slicing.py

Change the default jnp.take mode to "fill".

2022-04-28 06:01:56 -07:00

svd.py

[linalg] Adds full_matrices option to TPU SVD.

2022-04-20 12:32:00 -07:00

utils.py

[MHLO] Remove most XLA translation rules.

2022-04-18 08:28:35 -07:00

windowed_reductions.py

add in mlir lowering for tokens

2022-04-21 11:28:58 -07:00