author int64 658 755k | date stringdate 2012-06-12 08:34:29 2024-07-22 14:51:21 | timezone int64 -46,800 43.2k | hash stringlengths 40 40 | message stringlengths 5 490 | mods listlengths 1 16 | language stringclasses 20
values | license stringclasses 3
values | repo stringlengths 5 68 | original_message stringlengths 12 491 |
|---|---|---|---|---|---|---|---|---|---|
260,354 | 11.01.2022 10:43:26 | 18,000 | d3ed24f910eb5496875bea296c7ee916ce1c94eb | fix id_tap jit example | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/host_callback.py",
"new_path": "jax/experimental/host_callback.py",
"diff": "@@ -139,8 +139,8 @@ If your Python callbacks have side-effects you may need to wait until the\ncomputation has finished to ensure that the side-effects have been obse... | Python | Apache License 2.0 | google/jax | fix id_tap jit example |
260,335 | 11.12.2021 14:07:30 | 28,800 | 08aec823fda6d8b43eaa666cc8b51dbb219c7d20 | fix a custom_vjp post_process bug, related cleanups
related to doesn't completely fix it | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/custom_derivatives.py",
"new_path": "jax/_src/custom_derivatives.py",
"diff": "@@ -274,10 +274,10 @@ class CustomJVPCallPrimitive(core.CallPrimitive):\ndef bind(self, fun, jvp, *args):\nargs = map(core.full_lower, args)\ntop_trace = core.find_top_trac... | Python | Apache License 2.0 | google/jax | fix a custom_vjp post_process bug, related cleanups
related to #8783, doesn't completely fix it |
260,515 | 07.12.2021 01:26:23 | 0 | 2e5ab11652be377e0381eb6a2448d07e8c8652f0 | Resolves issue 8744 | [
{
"change_type": "MODIFY",
"old_path": "docs/notebooks/Custom_derivative_rules_for_Python_code.ipynb",
"new_path": "docs/notebooks/Custom_derivative_rules_for_Python_code.ipynb",
"diff": "\"id\": \"HowvqayEuy-H\"\n},\n\"source\": [\n- \"A limitation to this approach is that the argument `f` can't cl... | Python | Apache License 2.0 | google/jax | Resolves issue 8744 |
260,411 | 13.01.2022 12:28:25 | -7,200 | 5bfe1852a4626680f3612beee7d1a8b7a66fc79c | [jax2tf] Add jax2tf_associative_scan_reductions flag
This flag allows users to match the JAX performance for
associative reductions in CPU.
See README.md for details. | [
{
"change_type": "MODIFY",
"old_path": "CHANGELOG.md",
"new_path": "CHANGELOG.md",
"diff": "@@ -33,6 +33,10 @@ PLEASE REMEMBER TO CHANGE THE '..main' WITH AN ACTUAL TAG in GITHUB LINK.\nIf set, JAX dumps the MHLO/HLO IR it generates for each computation to a\nfile under the given path.\n* Added `jax... | Python | Apache License 2.0 | google/jax | [jax2tf] Add jax2tf_associative_scan_reductions flag
This flag allows users to match the JAX performance for
associative reductions in CPU.
See README.md for details. |
260,424 | 14.01.2022 11:18:40 | 0 | f591d0b2e946826fea57b4ced8d9928e4114748c | Add ensure_compile_time_eval docstring to docs | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.rst",
"new_path": "docs/jax.rst",
"diff": "@@ -95,6 +95,7 @@ Parallelization (:code:`pmap`)\n.. autofunction:: jit\n.. autofunction:: disable_jit\n+.. autofunction:: ensure_compile_time_eval\n.. autofunction:: xla_computation\n.. autofunction:: make_j... | Python | Apache License 2.0 | google/jax | Add ensure_compile_time_eval docstring to docs |
260,424 | 10.01.2022 18:21:41 | 0 | 8ea85769eaf35f591a0b3a9710ba4a49c31a7bc9 | Checkify: add way to disable categories of errors.
By default only user_asserts are lifted into the checked function. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/checkify.py",
"new_path": "jax/experimental/checkify.py",
"diff": "# See the License for the specific language governing permissions and\n# limitations under the License.\n+import enum\nfrom dataclasses import dataclass\nfrom functools import ... | Python | Apache License 2.0 | google/jax | Checkify: add way to disable categories of errors.
By default only user_asserts are lifted into the checked function. |
260,335 | 18.01.2022 11:17:48 | 28,800 | 5884c1f20eb63e0f07ca7ce41c3a9cc67950bdf6 | add caching to partial_eval_jaxpr
follow-up to | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/partial_eval.py",
"new_path": "jax/interpreters/partial_eval.py",
"diff": "@@ -723,6 +723,11 @@ def partial_eval_jaxpr(jaxpr: ClosedJaxpr, unknowns: Sequence[bool],\nto obtain the full outputs once `jaxpr_unknown` is ran. Outputs known ahead o... | Python | Apache License 2.0 | google/jax | add caching to partial_eval_jaxpr
follow-up to #9196 |
260,390 | 18.01.2022 14:17:17 | 28,800 | 9d43ccafa3598827063cabe49c697c6747335e0d | converting to use switch | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/scipy/linalg.py",
"new_path": "jax/_src/scipy/linalg.py",
"diff": "@@ -289,28 +289,20 @@ def _calc_P_Q(A):\nA_L1 = np_linalg.norm(A,1)\nn_squarings = 0\nif A.dtype == 'float64' or A.dtype == 'complex128':\n- U3, V3 = _pade3(A)\n- U5, V5 = _pade5(A)\n-... | Python | Apache License 2.0 | google/jax | converting to use switch |
260,390 | 18.01.2022 16:28:36 | 28,800 | cbb2f7baabaab1084b4ed5b6617ebd7ec1181d74 | changed to use argwhere | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/scipy/linalg.py",
"new_path": "jax/_src/scipy/linalg.py",
"diff": "@@ -294,15 +294,15 @@ def _calc_P_Q(A):\nA = A / 2**n_squarings\nconds = jnp.array([1.495585217958292e-002, 2.539398330063230e-001,\n9.504178996162932e-001, 2.097847961257068e+000, jnp... | Python | Apache License 2.0 | google/jax | changed to use argwhere |
260,390 | 19.01.2022 10:02:06 | 28,800 | 2d2ac12aa074a137cdf9d3d427b36f14b1fb3f1f | switching to use jnp.digitize for index identification | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/scipy/linalg.py",
"new_path": "jax/_src/scipy/linalg.py",
"diff": "@@ -293,15 +293,15 @@ def _calc_P_Q(A):\nn_squarings = jnp.maximum(0, jnp.floor(jnp.log2(A_L1 / maxnorm)))\nA = A / 2**n_squarings\nconds = jnp.array([1.495585217958292e-002, 2.5393983... | Python | Apache License 2.0 | google/jax | switching to use jnp.digitize for index identification |
260,335 | 19.01.2022 10:24:09 | 28,800 | 726f60f6bc2ba99128d815f9f8e2f22ce028b0ba | remove cpu platform setting in quickstart
fixes | [
{
"change_type": "MODIFY",
"old_path": "docs/notebooks/quickstart.ipynb",
"new_path": "docs/notebooks/quickstart.ipynb",
"diff": "\"from jax import random\"\n]\n},\n- {\n- \"cell_type\": \"code\",\n- \"execution_count\": null,\n- \"metadata\": {\n- \"tags\": [\n- \"remove-cell\"\n- ]\n- },\n- \"outp... | Python | Apache License 2.0 | google/jax | remove cpu platform setting in quickstart
fixes #9244 |
260,335 | 19.01.2022 11:43:02 | 28,800 | 13ede5b2eb8b1ca7e07642791c19bbf5df1c6284 | add origin info to leaked tracer error
add origin info to vmap tracers (BatchTracer) | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -747,7 +747,8 @@ the following:\nthreading.current_thread().pydev_do_not_trace = True\n\"\"\"\n-def maybe_find_leaked_tracers(x: Optional[Union[MainTrace, Sublevel]]):\n+def maybe_find_leaked_tracers(x: Opti... | Python | Apache License 2.0 | google/jax | add origin info to leaked tracer error
add origin info to vmap tracers (BatchTracer) |
260,335 | 18.01.2022 22:22:57 | 28,800 | 9488c5ae72432d82c1422dec91239c3aff5cf2c0 | checkify: fix and test post_process_call/map | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/checkify.py",
"new_path": "jax/experimental/checkify.py",
"diff": "@@ -29,11 +29,15 @@ from jax.interpreters import partial_eval as pe\nfrom jax.tree_util import tree_flatten, tree_unflatten, register_pytree_node\nfrom jax._src import source_i... | Python | Apache License 2.0 | google/jax | checkify: fix and test post_process_call/map |
260,335 | 19.01.2022 18:44:31 | 28,800 | bbaee9c54eeb0303f1b2ae45fc852993007ae5aa | pjit: add test for basic static_argnums | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/pjit.py",
"new_path": "jax/experimental/pjit.py",
"diff": "@@ -40,7 +40,7 @@ from jax.interpreters import batching\nfrom jax.interpreters import partial_eval as pe\nfrom jax.interpreters.sharded_jit import PartitionSpec\nfrom jax._src.lib impo... | Python | Apache License 2.0 | google/jax | pjit: add test for basic static_argnums |
260,335 | 19.01.2022 20:40:38 | 28,800 | ced2ae0104057295169fdc3b1ee9ec64d8c41da3 | NotImplementedError for new_checkpoint-of-xmap
Previously this was getting a default behavior which caused surprising
errors downstream. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/control_flow.py",
"new_path": "jax/_src/lax/control_flow.py",
"diff": "@@ -617,7 +617,8 @@ xla.register_translation(while_p, _while_loop_translation_rule,\ninitial_style=True)\nad.primitive_transposes[while_p] = _while_transpose_error\nbatching.ax... | Python | Apache License 2.0 | google/jax | NotImplementedError for new_checkpoint-of-xmap
Previously this was getting a default behavior which caused surprising
errors downstream. |
260,536 | 01.11.2021 13:58:37 | -3,600 | d05431a1ffd27abc3951d723c82fa63db88bdd69 | has_aux for jvp and forward-mode AD value_and_grad
Changes:
revert value_and_grad_fwd
add has_aux to jacfwd and jacrev
tests
fix mypy error | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/api.py",
"new_path": "jax/_src/api.py",
"diff": "@@ -964,7 +964,8 @@ def value_and_grad(fun: Callable, argnums: Union[int, Sequence[int]] = 0,\ntuple). If ``argnums`` is an integer then the gradient has the same shape\nand type as the positional argum... | Python | Apache License 2.0 | google/jax | has_aux for jvp and forward-mode AD value_and_grad
Changes:
- revert value_and_grad_fwd
- add has_aux to jacfwd and jacrev
- tests
fix mypy error |
260,578 | 21.01.2022 14:37:41 | 28,800 | c0212d4079b39468f57a0c06c412e310df9c3b14 | Add GDA to the API pages | [
{
"change_type": "ADD",
"old_path": null,
"new_path": "docs/jax.experimental.global_device_array.rst",
"diff": "+jax.experimental.global_device_array module\n+============================\n+\n+.. automodule:: jax.experimental.global_device_array\n+\n+API\n+---\n+\n+.. autoclass:: GlobalDeviceArray\n... | Python | Apache License 2.0 | google/jax | Add GDA to the API pages |
260,411 | 20.01.2022 14:56:27 | -7,200 | 83b818d45c474515d833b1d2084a7b39f4f00d72 | Add more documentation for buffer donation
Fixes: | [
{
"change_type": "MODIFY",
"old_path": "docs/faq.rst",
"new_path": "docs/faq.rst",
"diff": "@@ -366,6 +366,73 @@ Interestingly, ``jax.grad(divide)(3., 2.)``, works because :func:`jax.grad`\nuses concrete tracers, and resolves the conditional using the concrete\nvalue of ``y``.\n+.. _faq-donation:\n+... | Python | Apache License 2.0 | google/jax | Add more documentation for buffer donation
Fixes: #9237 |
260,411 | 20.01.2022 13:25:35 | -7,200 | 34ba42b5daaf4e1ed5c2d4793ea5d866a342b957 | [jax2tf] Improve shape polymorphism for scatter with CLIP mode
Fixes: | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/lax.py",
"new_path": "jax/_src/lax/lax.py",
"diff": "@@ -701,7 +701,7 @@ def broadcast_in_dim(operand: Array, shape: Shape,\nArgs:\noperand: an array\nshape: the shape of the target array\n- broadcast_dimensions: which dimension in the target shap... | Python | Apache License 2.0 | google/jax | [jax2tf] Improve shape polymorphism for scatter with CLIP mode
Fixes: #9231 |
260,411 | 24.01.2022 10:06:22 | -3,600 | cb3fadce29fa7d3075547f1b5ab8b2249f7d8cd4 | [jax2tf] Move the no_xla_limitations documentation
Put it in the g3doc directory, along with the other pieces
of jax2tf documentation. This enables special documentation
features in the Google repo. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -851,29 +851,12 @@ There are several drawbacks of using XLA TF ops:\n* These ops will only be executable by a consumer that has XLA linked in.\nThis should not be ... | Python | Apache License 2.0 | google/jax | [jax2tf] Move the no_xla_limitations documentation
Put it in the g3doc directory, along with the other pieces
of jax2tf documentation. This enables special documentation
features in the Google repo. |
260,586 | 25.01.2022 09:54:23 | 0 | 70bf2812505505e68f528e7e4f446935673c22eb | Fix `max_squarings` in `expm` | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/scipy/linalg.py",
"new_path": "jax/_src/scipy/linalg.py",
"diff": "@@ -275,7 +275,7 @@ def expm(A, *, upper_triangular=False, max_squarings=16):\ndef _compute(args):\nA, P, Q = args\nR = _solve_P_Q(P, Q, upper_triangular)\n- R = _squaring(R, n_squarin... | Python | Apache License 2.0 | google/jax | Fix `max_squarings` in `expm` |
260,314 | 25.01.2022 12:20:55 | 18,000 | 3afe67367df36b73053371c0429923eb549bc793 | NamedShape matches tuple behavior | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -1639,13 +1639,11 @@ class NamedShape:\nf\"{', '.join(f'{k}={v}' for k, v in self.__named.items())})\")\ndef __eq__(self, other):\n- if other is None:\n- return False\nif isinstance(other, NamedShape):\nretu... | Python | Apache License 2.0 | google/jax | NamedShape matches tuple behavior |
260,335 | 25.01.2022 15:27:29 | 28,800 | 98816f3ffa70d85655636f417ad1df2f4b3aedae | jaxpr staging: only one tracer per jaxpr variable | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/partial_eval.py",
"new_path": "jax/interpreters/partial_eval.py",
"diff": "@@ -1193,20 +1193,29 @@ class DynamicJaxprTracer(core.Tracer):\nif not self._trace.main.jaxpr_stack: # type: ignore\nraise core.escaped_tracer_error(self, None)\n+Trace... | Python | Apache License 2.0 | google/jax | jaxpr staging: only one tracer per jaxpr variable |
260,447 | 31.01.2022 11:01:58 | 28,800 | 97d55eb13c1c03ba6e2714d52e37b8c45bee9162 | [JAX] Re-enables lowering bcoo dot general to cuSparse. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/config.py",
"new_path": "jax/_src/config.py",
"diff": "@@ -663,3 +663,10 @@ enable_mlir = config.define_bool_state(\ndefault=lib.mlir_api_version >= 1,\nhelp=('Enables an experimental code path that compiles JAX programs via '\n'emitting the MLIR MHLO... | Python | Apache License 2.0 | google/jax | [JAX] Re-enables lowering bcoo dot general to cuSparse.
PiperOrigin-RevId: 425410511 |
260,631 | 31.01.2022 13:45:26 | 28,800 | c12ca7f64cb4a0338939d84b7dc5c25ccc5b47be | [XLA:TPU] Add 'aggregate_to_topk' option to ann in jax
Also adds a pmap test for demonstrating multi-tpu ann. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/ann.py",
"new_path": "jax/experimental/ann.py",
"diff": "# See the License for the specific language governing permissions and\n# limitations under the License.\n-\"\"\"ANN (Approximate Nearest Neighbor) is an **experimental** module for fast ... | Python | Apache License 2.0 | google/jax | [XLA:TPU] Add 'aggregate_to_topk' option to ann in jax
Also adds a pmap test for demonstrating multi-tpu ann.
PiperOrigin-RevId: 425451716 |
260,411 | 01.02.2022 16:06:31 | -7,200 | 0d4fdccbe6354f600532c2dfa25c9723c66f7fac | [jax2tf] minor changes to the documentation for TF Lite | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/examples/tflite/mnist/README.md",
"new_path": "jax/experimental/jax2tf/examples/tflite/mnist/README.md",
"diff": "@@ -59,6 +59,15 @@ tf_predict = tf.function(\nautograph=False)\n```\n+Note the `enable_xla=False` parameter to `jax2tf.con... | Python | Apache License 2.0 | google/jax | [jax2tf] minor changes to the documentation for TF Lite |
260,442 | 02.02.2022 14:01:54 | 28,800 | 08fd83b68c1dea0b35b13f661a9e1a50db9c2e61 | [mhlo] jax infeed/outfeed lowering changes in response to detupling of the ops. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/lax.py",
"new_path": "jax/_src/lax/lax.py",
"diff": "@@ -4058,24 +4058,45 @@ infeed_p.def_abstract_eval(_infeed_abstract_eval)\nxla.register_translation(infeed_p, _infeed_translation_rule)\n+if jax._src.lib.mlir_api_version >= 2:\n+\n+ def _infeed... | Python | Apache License 2.0 | google/jax | [mhlo] jax infeed/outfeed lowering changes in response to detupling of the ops.
PiperOrigin-RevId: 425972002 |
260,631 | 02.02.2022 15:49:28 | 28,800 | a0abe8e4aca186b4746887ad69f571802bb83563 | [JAX] Move the ann recall computation to ann.py.
This function is very useful for our users to evaluate the ann results
against the standard ann datasets that provides the ground truth. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/ann.py",
"new_path": "jax/experimental/ann.py",
"diff": "@@ -163,6 +163,45 @@ def approx_min_k(operand: Array,\naggregate_to_topk=aggregate_to_topk)\n+def ann_recall(result_neighbors: Array, ground_truth_neighbors: Array) -> float:\n+ \"\"\"Co... | Python | Apache License 2.0 | google/jax | [JAX] Move the ann recall computation to ann.py.
This function is very useful for our users to evaluate the ann results
against the standard ann datasets that provides the ground truth.
PiperOrigin-RevId: 425997236 |
260,411 | 03.02.2022 11:09:12 | -7,200 | 53be1b9c2c6752a8700ed3b07d316bf92108f8f5 | [jax2tf] Add a TF version check for support of conv:batch_group_count
The batch_group_count != 1 is only supported with TF version 2.8.0 and
above.
Fixes: | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -1086,8 +1086,8 @@ cannot be sure that the shape-polymorphic conversion is safe.\n## TensorFlow versions supported\n-The ``jax2tf.convert`` and `call_tf` require v... | Python | Apache License 2.0 | google/jax | [jax2tf] Add a TF version check for support of conv:batch_group_count
The batch_group_count != 1 is only supported with TF version 2.8.0 and
above.
Fixes: #9384 |
260,447 | 03.02.2022 11:16:18 | 28,800 | 5a012d5e7b056047208aa4c21ca226faa68fd8bb | [JAX] Added jit-able singular value decomposition. | [
{
"change_type": "ADD",
"old_path": null,
"new_path": "tests/svd_test.py",
"diff": "+# Copyright 2022 Google LLC\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may obtain a copy of the License at\... | Python | Apache License 2.0 | google/jax | [JAX] Added jit-able singular value decomposition.
PiperOrigin-RevId: 426193395 |
260,335 | 28.01.2022 15:54:19 | 28,800 | e186aa3f1e428a2716f6dff7b37119e88aadf0e5 | add and test pytree utils for better errors | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/api.py",
"new_path": "jax/_src/api.py",
"diff": "@@ -1404,11 +1404,12 @@ def vmap(fun: F, in_axes=0, out_axes=0, axis_name=None, axis_size=None) -> F:\nrange ``[-ndim, ndim)`` for each array, where ``ndim`` is the number of\ndimensions (axes) of the c... | Python | Apache License 2.0 | google/jax | add and test pytree utils for better errors |
260,631 | 04.02.2022 20:02:45 | 28,800 | 7a6986c4c8fd8469bae36306efb0417b0a2f6d8c | [JAX] Update the ANN document.
Deletes the documentation that explains the algorithm.
I don't think it is the necessary detail for users.
We'll write a paper to explain it in detail very soon. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/ann.py",
"new_path": "jax/experimental/ann.py",
"diff": "# See the License for the specific language governing permissions and\n# limitations under the License.\n-\"\"\"ANN (Approximate Nearest Neighbor) computes top-k with a configurable reca... | Python | Apache License 2.0 | google/jax | [JAX] Update the ANN document.
Deletes the documentation that explains the algorithm.
I don't think it is the necessary detail for users.
We'll write a paper to explain it in detail very soon.
PiperOrigin-RevId: 426546480 |
260,287 | 07.02.2022 10:34:36 | 28,800 | 42cd7ed1d0653c5a9d1b25c3f701ff3ec2b6c74b | Allow nesting MANUAL-style xmaps in pjits | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/maps.py",
"new_path": "jax/experimental/maps.py",
"diff": "@@ -1312,7 +1312,7 @@ pxla.SPMDBatchTrace.process_xmap = partialmethod(_batch_trace_process_xmap, True\ndef _xmap_lowering_rule(*args, **kwargs):\nif config.experimental_xmap_spmd_lowe... | Python | Apache License 2.0 | google/jax | Allow nesting MANUAL-style xmaps in pjits
PiperOrigin-RevId: 426955137 |
260,287 | 07.02.2022 10:37:01 | 28,800 | 296832e891def967b64ad0d24e3df1ad0cb5cf8f | Use aval_out to construct a sharding spec in shard to full
The shard's dimensions might be too small and might trigger asserts, even though
the shape has no influence on sharding specs. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/pxla.py",
"new_path": "jax/interpreters/pxla.py",
"diff": "@@ -1977,7 +1977,7 @@ def _shard_to_full_lowering(ctx, x, *, axes: ArrayMapping, mesh: Mesh):\nmanual_proto = _manual_proto(aval_in, axes, mesh)\nresult_type, = mlir.aval_to_ir_types(a... | Python | Apache License 2.0 | google/jax | Use aval_out to construct a sharding spec in shard to full
The shard's dimensions might be too small and might trigger asserts, even though
the shape has no influence on sharding specs.
PiperOrigin-RevId: 426955706 |
260,303 | 07.02.2022 17:11:10 | 28,800 | 085e65863a5137cb1e25bfca8f494f22410bffb4 | Update compilation cache logging to log individual hashes as well as cumulative hash | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/compilation_cache/compilation_cache.py",
"new_path": "jax/experimental/compilation_cache/compilation_cache.py",
"diff": "@@ -58,8 +58,14 @@ def put_executable(module_name, xla_computation, compile_options,\nserialized_executable = backend.seri... | Python | Apache License 2.0 | google/jax | Update compilation cache logging to log individual hashes as well as cumulative hash |
260,330 | 08.02.2022 13:28:41 | -3,600 | c07a0f11395b07caa3660d0186e58f8a1752eab7 | Add test and jet-primitive for dynamic_slice | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jet.py",
"new_path": "jax/experimental/jet.py",
"diff": "@@ -238,6 +238,7 @@ deflinear(lax.reshape_p)\ndeflinear(lax.rev_p)\ndeflinear(lax.transpose_p)\ndeflinear(lax.slice_p)\n+deflinear(lax.dynamic_slice_p)\ndeflinear(lax.reduce_sum_p)\ndefl... | Python | Apache License 2.0 | google/jax | Add test and jet-primitive for dynamic_slice |
260,424 | 10.01.2022 12:25:47 | 0 | 0042edb5f444462cdf4d00d3c83c956087fa710a | Checkify: rename some symbols and add some docstrings. | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.experimental.rst",
"new_path": "docs/jax.experimental.rst",
"diff": "@@ -28,3 +28,7 @@ Experimental APIs\nenable_x64\ndisable_x64\n+\n+ jax.experimental.checkify.checkify\n+ jax.experimental.checkify.check\n+ jax.experimental.checkify.check_error\n"
... | Python | Apache License 2.0 | google/jax | Checkify: rename some symbols and add some docstrings. |
260,424 | 08.02.2022 19:33:55 | 0 | 010eb82ad01a08f8b9a17e6b2f39676bc7066eb6 | Rename wrapper functions to always refer to the JAX api function.
eg. batched_fun -> vmap_fun | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/api.py",
"new_path": "jax/_src/api.py",
"diff": "@@ -1544,7 +1544,7 @@ def vmap(fun: F, in_axes=0, out_axes=0, axis_name=None, axis_size=None) -> F:\n@wraps(fun, docstr=docstr)\n@api_boundary\n- def batched_fun(*args, **kwargs):\n+ def vmap_f(*args, *... | Python | Apache License 2.0 | google/jax | Rename wrapper functions to always refer to the JAX api function.
eg. batched_fun -> vmap_fun |
260,424 | 08.02.2022 20:41:19 | 0 | 4d0db5d9759b855a5751bf66ade04a73e5c71f66 | Fix build and address suggestion | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/checkify.py",
"new_path": "jax/experimental/checkify.py",
"diff": "@@ -292,16 +292,18 @@ def check_error(error: Error) -> None:\ncannot be staged out by `jit`, `pmap`, `scan`, etc. Both also can be\nfunctionalized by using `checkify`.\n- But u... | Python | Apache License 2.0 | google/jax | Fix build and address suggestion |
260,335 | 08.02.2022 12:45:38 | 28,800 | d57990ecf9dcc6b37c3c240d5abe021d20d2fef9 | improve pjit in/out_axis_resources pytree errors
This is an application of the utilities in | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/tree_util.py",
"new_path": "jax/_src/tree_util.py",
"diff": "# limitations under the License.\nimport collections\n+import difflib\nimport functools\nfrom functools import partial\nimport operator as op\nfrom typing import (Any, Callable, Hashable, It... | Python | Apache License 2.0 | google/jax | improve pjit in/out_axis_resources pytree errors
This is an application of the utilities in #9372. |
260,335 | 09.02.2022 11:18:40 | 28,800 | 4ce749e681d5cd9a9e302e7e6a6f27bdc9384daa | [checkify] handle custom_jvp | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/checkify.py",
"new_path": "jax/experimental/checkify.py",
"diff": "@@ -186,6 +186,34 @@ class CheckifyTrace(core.Trace):\nreturn (0, 0, *out_axes)\nreturn (err, code, *vals), (todo, out_axes_transform)\n+ def process_custom_jvp_call(self, prim... | Python | Apache License 2.0 | google/jax | [checkify] handle custom_jvp |
260,335 | 09.02.2022 12:04:34 | 28,800 | d9270b242d157ee7c578661636be2a0ed946560e | [checkify] add custom_vjp support | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/checkify.py",
"new_path": "jax/experimental/checkify.py",
"diff": "@@ -214,6 +214,21 @@ class CheckifyTrace(core.Trace):\nreturn [CheckifyTracer(trace, x) for x in vals]\nreturn (err, code, *vals), todo\n+ def process_custom_vjp_call(self, pri... | Python | Apache License 2.0 | google/jax | [checkify] add custom_vjp support |
260,424 | 10.02.2022 14:28:46 | 0 | 2ba8aec2746a01c211ae7a1a49220cc103161f37 | Checkify: Fix docstring formatting and polish enabled_errors sets. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/checkify.py",
"new_path": "jax/experimental/checkify.py",
"diff": "@@ -315,7 +315,7 @@ def checkify_fun_to_jaxpr(f, error, enabled_errors, in_avals):\n## assert primitive\ndef check(pred: Bool, msg: str) -> None:\n- \"\"\"Check a condition, ad... | Python | Apache License 2.0 | google/jax | Checkify: Fix docstring formatting and polish enabled_errors sets. |
260,330 | 12.02.2022 08:14:57 | -3,600 | faf57d813b74f4741823fca137dfac081986bb80 | Wrote docstring for jet | [
{
"change_type": "ADD",
"old_path": null,
"new_path": "docs/jax.experimental.jet.rst",
"diff": "+jax.experimental.jet module\n+===========================\n+\n+.. automodule:: jax.experimental.jet\n+\n+API\n+---\n+\n+.. autofunction:: jet\n"
},
{
"change_type": "MODIFY",
"old_path": "doc... | Python | Apache License 2.0 | google/jax | Wrote docstring for jet |
260,330 | 12.02.2022 08:19:54 | -3,600 | 4f744e9033e309d6f6973b8ca49f0715c4009829 | 'approximates' -> 'corresponding to' in jet-docstring. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jet.py",
"new_path": "jax/experimental/jet.py",
"diff": "@@ -26,7 +26,7 @@ r\"\"\"Jet is an experimental module for higher-order automatic differentiation\n.. math:: y(t) := y_0 + y_1 t + \\frac{y_2}{2!} t^2 + ... + \\frac{1}{d!} y_d t^d.\n- w... | Python | Apache License 2.0 | google/jax | 'approximates' -> 'corresponding to' in jet-docstring. |
260,330 | 12.02.2022 08:31:39 | -3,600 | 0210b563e832488a7796b7959c227d8abc45150b | Changed example output values to exact numbers | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jet.py",
"new_path": "jax/experimental/jet.py",
"diff": "@@ -104,13 +104,13 @@ def jet(fun, primals, series):\n:func:`jet` returns the Taylor coefficients of :math:`f(x(t)) = \\sin(t^3)`:\n>>> print(y0, np.sin(x0))\n- 0.12 0.12\n+ 0.12467473 0... | Python | Apache License 2.0 | google/jax | Changed example output values to exact numbers |
260,335 | 04.02.2022 19:12:57 | 28,800 | 004bb684ea630633782dc5071dd73813350b8975 | add flag for jaxpr colorful syntax highlighting
set it on by default | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/config.py",
"new_path": "jax/_src/config.py",
"diff": "@@ -440,6 +440,12 @@ flags.DEFINE_integer(\nhelp='Set the number of stack frames in JAX tracer error messages.'\n)\n+flags.DEFINE_bool(\n+ 'jax_pprint_use_color',\n+ bool_env('JAX_PPRINT_USE_COLOR... | Python | Apache License 2.0 | google/jax | add flag for jaxpr colorful syntax highlighting
set it on by default |
260,330 | 13.02.2022 08:19:24 | -3,600 | cbc9024b9db492ff2b865dfb97aa5f57f85104a6 | Replaced fraction-notation for derivatives, and link different notes | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jet.py",
"new_path": "jax/experimental/jet.py",
"diff": "@@ -18,19 +18,19 @@ r\"\"\"Jet is an experimental module for higher-order automatic differentiation\nHow? Through the propagation of truncated Taylor polynomials.\nLet :math:`x: \\mathbb... | Python | Apache License 2.0 | google/jax | Replaced fraction-notation for derivatives, and link different notes |
260,330 | 13.02.2022 08:52:34 | -3,600 | c2a20db9b68b22ddf853a744e7dbd37578caab01 | Changed Taylor-series notation in module docstring | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jet.py",
"new_path": "jax/experimental/jet.py",
"diff": "@@ -16,26 +16,31 @@ r\"\"\"Jet is an experimental module for higher-order automatic differentiation\nthat does not rely on repeated first-order automatic differentiation.\nHow? Through t... | Python | Apache License 2.0 | google/jax | Changed Taylor-series notation in module docstring |
260,330 | 13.02.2022 08:59:29 | -3,600 | 5446cee0e926fc3556f59e07061a63774668851b | fixed a typo in maths-part of jet-module-docstring | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jet.py",
"new_path": "jax/experimental/jet.py",
"diff": "@@ -27,7 +27,7 @@ r\"\"\"Jet is an experimental module for higher-order automatic differentiation\n.. math::\n(h_0, ... h_K) :=\n- (h(x), \\partial h(x)[v], \\partial^2 h(x)[v], ..., \\p... | Python | Apache License 2.0 | google/jax | fixed a typo in maths-part of jet-module-docstring |
260,580 | 13.02.2022 23:47:52 | -32,400 | fc4f47cdaec9160d054053e58924d59705224c39 | shape_poly_test.py : remove duplicate word | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/tests/shape_poly_test.py",
"new_path": "jax/experimental/jax2tf/tests/shape_poly_test.py",
"diff": "@@ -1681,7 +1681,7 @@ def _add_vmap_primitive_harnesses():\ndevice = jtu.device_under_test()\nfor h in all_h:\n- # Drop the the JAX limi... | Python | Apache License 2.0 | google/jax | shape_poly_test.py : remove duplicate word |
260,424 | 14.02.2022 15:02:05 | 0 | a4cacf5729466afb7efbec23b0f9dffeeb79faff | Checkify: handle named_call in process_call.
named_call does not specify donated_invars, this change handles this missing
param case.
For future reference: we might want to add a call_param_updater registry to define
how call params need to get updated wrt checkify, like eg. partial_eval/ad does. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/checkify.py",
"new_path": "jax/experimental/checkify.py",
"diff": "@@ -137,8 +137,10 @@ class CheckifyTrace(core.Trace):\nin_vals = [t.val for t in tracers]\ne = popattr(self.main, 'error')\nf, msgs = checkify_subtrace(f, self.main, tuple(e.ms... | Python | Apache License 2.0 | google/jax | Checkify: handle named_call in process_call.
named_call does not specify donated_invars, this change handles this missing
param case.
For future reference: we might want to add a call_param_updater registry to define
how call params need to get updated wrt checkify, like eg. partial_eval/ad does. |
260,447 | 14.02.2022 15:23:47 | 28,800 | 273ea6262404eaade9d50ec01c2c9bf3be133140 | [sparse] Updates `bcoo_dot_general` cuSparse lowering rule by adding sorted indices. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/sparse/bcoo.py",
"new_path": "jax/experimental/sparse/bcoo.py",
"diff": "@@ -142,6 +142,9 @@ def _bcoo_sort_indices(data, indices, shape):\nf = broadcasting_vmap(f)\nreturn f(data, indices)\n+_bcoo_sort_indices_rule = xla.lower_fun(\n+ _bcoo_s... | Python | Apache License 2.0 | google/jax | [sparse] Updates `bcoo_dot_general` cuSparse lowering rule by adding sorted indices.
PiperOrigin-RevId: 428621454 |
260,424 | 15.02.2022 02:42:30 | 28,800 | 0d9990e4f3305aa9b1180fcabc5b96234b6e40f1 | Run all tests with jax_traceback_filtering=off.
Context: If an AssertionError is thrown inside a test and traceback filtering
is enabled, most of the stack-trace is swallowed (due to
https://bugs.python.org/issue24959). | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/test_util.py",
"new_path": "jax/_src/test_util.py",
"diff": "@@ -917,7 +917,8 @@ class JaxTestCase(parameterized.TestCase):\n\"\"\"Base class for JAX tests including numerical checks and boilerplate.\"\"\"\n_default_config = {\n'jax_enable_checks': Tr... | Python | Apache License 2.0 | google/jax | Run all tests with jax_traceback_filtering=off.
Context: If an AssertionError is thrown inside a test and traceback filtering
is enabled, most of the stack-trace is swallowed (due to
https://bugs.python.org/issue24959).
PiperOrigin-RevId: 428729211 |
260,287 | 15.02.2022 03:43:40 | 28,800 | c551beda7af61fe0ba4d3a590d41333c913f5923 | Add a partial_eval_jaxpr_custom_rule for xmap
Additionaly fix a bug in partial_eval rule for xmap. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/ad_checkpoint.py",
"new_path": "jax/_src/ad_checkpoint.py",
"diff": "@@ -335,11 +335,8 @@ def remat_partial_eval(trace, *tracers, jaxpr, **params):\nreturn pe._zip_knowns(out_known_tracers, out_jaxpr_tracers, out_unknowns)\npe.custom_partial_eval_rule... | Python | Apache License 2.0 | google/jax | Add a partial_eval_jaxpr_custom_rule for xmap
Additionaly fix a bug in partial_eval rule for xmap.
PiperOrigin-RevId: 428738277 |
260,287 | 15.02.2022 04:00:19 | 28,800 | b75b0c04dee4a384864d4e9af1dad3783acf5336 | Limit the set of unspecified dims to those that are not explicitly converted to MANUAL. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/pxla.py",
"new_path": "jax/interpreters/pxla.py",
"diff": "@@ -1967,7 +1967,7 @@ def _full_to_shard_lowering(ctx, x, *, axes: ArrayMapping, mesh: Mesh):\nsx = mlir.wrap_with_sharding_op(x, sharding_proto, unspecified_dims=unspecified_dims)\nma... | Python | Apache License 2.0 | google/jax | Limit the set of unspecified dims to those that are not explicitly converted to MANUAL.
PiperOrigin-RevId: 428740792 |
260,287 | 15.02.2022 10:12:31 | 28,800 | a14ccb99d984ceb079267dd44d38901a2a9b057d | Try adding support for nesting with_sharding_constraint in MANUAL xmaps | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/dispatch.py",
"new_path": "jax/_src/dispatch.py",
"diff": "@@ -256,8 +256,8 @@ def lower_xla_callable(fun: lu.WrappedFun, device, backend, name,\nmodule_name = f\"jit_{fun.__name__}\"\nif config.jax_enable_mlir:\nmodule = mlir.lower_jaxpr_to_module(\n... | Python | Apache License 2.0 | google/jax | Try adding support for nesting with_sharding_constraint in MANUAL xmaps
PiperOrigin-RevId: 428812729 |
260,424 | 15.02.2022 13:12:19 | 28,800 | b15c7f609a01cc0a420141ef4ea30b37d0297666 | Checkify: fix check_error of nd-error. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/lax.py",
"new_path": "jax/_src/lax/lax.py",
"diff": "@@ -3202,7 +3202,7 @@ def _select_shape_rule(which, *cases):\nraise TypeError(\"select must have at least one case\")\nif any(case.shape != cases[0].shape for case in cases[1:]):\nmsg = \"select... | Python | Apache License 2.0 | google/jax | Checkify: fix check_error of nd-error.
PiperOrigin-RevId: 428857813 |
260,573 | 16.02.2022 09:02:55 | 0 | c1cbf786117654e51f8f465aaf036d0f2e213ada | fix: use boolean tf functions where possible instead of casting to int8 for jax2tf | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -1348,7 +1348,7 @@ def _not(x):\ntf_impl[lax.not_p] = _not\n-def bool_to_int8(f, argnums: Sequence[int]):\n+def handle_boolean_args(f, argnums: Sequence[int], bool... | Python | Apache License 2.0 | google/jax | fix: use boolean tf functions where possible instead of casting to int8 for jax2tf |
260,411 | 16.02.2022 15:41:37 | -3,600 | 461b37b2a80a6a3f46a35941c5b0f7401fae6253 | [jax2tf] Fixed stale documentation about XLA metadata.
jax2tf does not yet support passing source location information
through to TF. The mechanism is partially implemented but disabled.
Here we remove misleading documentation that suggests the mechanism
is enabled. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -606,13 +606,6 @@ operations. There is support for `sharded_jit` and `pjit`.\nIf you suspect that the SavedModel is larger than it should be, check first\nthat you... | Python | Apache License 2.0 | google/jax | [jax2tf] Fixed stale documentation about XLA metadata.
jax2tf does not yet support passing source location information
through to TF. The mechanism is partially implemented but disabled.
Here we remove misleading documentation that suggests the mechanism
is enabled. |
260,411 | 16.02.2022 16:16:08 | -3,600 | 1928f6e6b10ed59d09d6af65b7d162773fe5d577 | [jax2tf] Fixes shape polymorphism for jnp.take_along_axes
Fixes: | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/numpy/lax_numpy.py",
"new_path": "jax/_src/numpy/lax_numpy.py",
"diff": "@@ -5624,7 +5624,7 @@ def take_along_axis(arr, indices, axis: Optional[int]):\nidx_shape = indices.shape\nout_shape = lax.broadcast_shapes(idx_shape, arr_shape)\n- index_dims = [... | Python | Apache License 2.0 | google/jax | [jax2tf] Fixes shape polymorphism for jnp.take_along_axes
Fixes: #9552 |
260,424 | 16.02.2022 19:12:54 | 0 | 758c721605220ecc7de4fc208fc62dbaa8a26f30 | Checkify: fix nd-error case when array only has 1 element. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/checkify.py",
"new_path": "jax/experimental/checkify.py",
"diff": "@@ -410,7 +410,7 @@ def check_error(error: Error) -> None:\n>>> # can re-checkify\n>>> error, _ = checkify.checkify(with_inner_jit)(-1)\n\"\"\"\n- if np.size(error.err) > 1:\n+... | Python | Apache License 2.0 | google/jax | Checkify: fix nd-error case when array only has 1 element. |
260,397 | 12.02.2022 02:55:53 | -3,600 | 514d8883ce958a6e07f8fc106f4c27799d2ee9a4 | adds jax.scipy.schur | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/scipy/linalg.py",
"new_path": "jax/_src/scipy/linalg.py",
"diff": "@@ -103,7 +103,18 @@ def eigh(a, b=None, lower=True, eigvals_only=False, overwrite_a=False,\ndel overwrite_a, overwrite_b, turbo, check_finite\nreturn _eigh(a, b, lower, eigvals_only, ... | Python | Apache License 2.0 | google/jax | adds jax.scipy.schur |
260,397 | 12.02.2022 03:11:57 | -3,600 | cb732323f376a26cf88e177a96ebd955074acbfc | adds jax.scipy.linalg.sqrtm | [
{
"change_type": "MODIFY",
"old_path": "jax/scipy/linalg.py",
"new_path": "jax/scipy/linalg.py",
"diff": "@@ -21,7 +21,6 @@ from jax._src.scipy.linalg import (\ncho_solve as cho_solve,\ndet as det,\neigh as eigh,\n- schur as schur,\neigh_tridiagonal as eigh_tridiagonal,\nexpm as expm,\nexpm_frechet ... | Python | Apache License 2.0 | google/jax | adds jax.scipy.linalg.sqrtm |
260,710 | 03.02.2022 14:18:44 | -32,400 | e085370ec4137cf0f73c5163cb664bc4e1c46082 | Add some functions for spectral analysis.
This commit adds "stft", "csd", and "welch" functions in scipy.signal. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/scipy/signal.py",
"new_path": "jax/_src/scipy/signal.py",
"diff": "# limitations under the License.\nimport scipy.signal as osp_signal\n+import operator\nimport warnings\nimport numpy as np\n+import jax\n+import jax.numpy.fft\nfrom jax import lax\n+fr... | Python | Apache License 2.0 | google/jax | Add some functions for spectral analysis.
This commit adds "stft", "csd", and "welch" functions in scipy.signal. |
260,335 | 16.02.2022 23:11:22 | 28,800 | e0fb424d81d0898971a85587e8f60c0a1fc000d7 | use singleton dims in broadcasting binop batchers | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/batching.py",
"new_path": "jax/interpreters/batching.py",
"diff": "@@ -597,8 +597,10 @@ def broadcast_batcher(prim, args, dims, **params):\nout = prim.bind(*args, **params)\nreturn (out, (d,) * len(out)) if prim.multiple_results else (out, d)\... | Python | Apache License 2.0 | google/jax | use singleton dims in broadcasting binop batchers |
260,287 | 17.02.2022 03:27:38 | 28,800 | 57f423203d6313d2beba7fefbd700c25aa400372 | Fix uninitialized axis_env error when MLIR lowering is disabled | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/pxla.py",
"new_path": "jax/interpreters/pxla.py",
"diff": "@@ -2173,6 +2173,7 @@ def lower_mesh_computation(\nout_partitions_t = xla.tuple_sharding_proto(out_partitions)\npartitions_proto = True\naxis_ctx = mlir.SPMDAxisContext(mesh)\n+ axis_e... | Python | Apache License 2.0 | google/jax | Fix uninitialized axis_env error when MLIR lowering is disabled
PiperOrigin-RevId: 429267926 |
260,424 | 17.02.2022 04:30:27 | 28,800 | 73f23705d01b88d4b4aad13a279f6a71d9116d0b | Checkify: explicitly export public API, hide private symbols. | [
{
"change_type": "ADD",
"old_path": null,
"new_path": "jax/experimental/checkify/__init__.py",
"diff": "+# Copyright 2022 Google LLC\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may obtain a cop... | Python | Apache License 2.0 | google/jax | Checkify: explicitly export public API, hide private symbols.
PiperOrigin-RevId: 429277551 |
260,486 | 17.02.2022 14:31:23 | 28,800 | 50f4b5806e2891dd82e564da2a7d8fc34f6d9e25 | Fix Iree backend to for copy_to_device and executable results
Executable results can be a tuple, if so iterate over the entires.
Copy to device should just return the the IREE buffer as device buffer
management is still in progress. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/iree.py",
"new_path": "jax/_src/iree.py",
"diff": "@@ -59,6 +59,9 @@ class IreeBuffer(xla_client.DeviceArrayBase):\nself._device = device\nself._npy_value = np.asarray(npy_value)\n+ def copy_to_device(self, device):\n+ return self\n+\ndef to_py(self) ... | Python | Apache License 2.0 | google/jax | Fix Iree backend to for copy_to_device and executable results
Executable results can be a tuple, if so iterate over the entires.
Copy to device should just return the the IREE buffer as device buffer
management is still in progress. |
260,573 | 18.02.2022 08:20:48 | 0 | 20635e4ac51565a939411265856a6c44b9bc6d85 | fix: simplify reduce_min with reduce_all | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -1683,18 +1683,10 @@ tf_impl[lax.reduce_sum_p] = axes_to_axis(tf.reduce_sum)\ntf_impl[lax.reduce_prod_p] = axes_to_axis(tf.reduce_prod)\ntf_impl[lax.reduce_max_p] ... | Python | Apache License 2.0 | google/jax | fix: simplify reduce_min with reduce_all |
260,411 | 18.02.2022 09:38:10 | -3,600 | 3576388982571737290d4a428103a8e76c405e82 | [jax2tf] Added more links to the documentation | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -13,8 +13,12 @@ written in JAX, possibly including JAX transformations, and turn it into\na function that uses only TensorFlow operations. The converted function\n... | Python | Apache License 2.0 | google/jax | [jax2tf] Added more links to the documentation |
260,643 | 18.02.2022 09:07:29 | 0 | f926d089f803b2bc6440030eec7d8488412a80b5 | docs: spelling mistake fixed | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -1686,7 +1686,7 @@ tf_impl[lax.reduce_max_p] = handle_boolean_args(\nboolean_f=axes_to_axis(tf.reduce_any)) # Max is T if any one is T\ntf_impl[lax.reduce_min_p] =... | Python | Apache License 2.0 | google/jax | docs: spelling mistake fixed |
260,424 | 17.02.2022 12:31:59 | 0 | 2eeb683df0f264aba5c21f03b9d347e79c48fca8 | Checkify: address initial feedback.
- add a way to run checkify with no errors enabled
- clarify "can't be staged" error message
- export init_error: a way for users to set a default Error value | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/checkify/__init__.py",
"new_path": "jax/experimental/checkify/__init__.py",
"diff": "@@ -24,6 +24,7 @@ from jax.experimental.checkify.checkify_impl import (\ndiv_checks as div_checks,\nfloat_checks as float_checks,\nindex_checks as index_check... | Python | Apache License 2.0 | google/jax | Checkify: address initial feedback.
- add a way to run checkify with no errors enabled
- clarify "can't be staged" error message
- export init_error: a way for users to set a default Error value |
260,486 | 18.02.2022 22:58:51 | 28,800 | 23d5eb0896a92c227b3e72df1a368fc86cef933e | IREE's get_default_device_assignment should return List[Device]
Previously return List[List[Device]] which is not how the function is used.
Updated to use the alternative overload. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/iree.py",
"new_path": "jax/_src/iree.py",
"diff": "@@ -125,11 +125,11 @@ class IreeClient:\ndef get_default_device_assignment(\nself,\n- num_replicas: int,\n- num_partitions: int = 1) -> List[List[IreeDevice]]:\n- if num_replicas != 1 or num_partition... | Python | Apache License 2.0 | google/jax | IREE's get_default_device_assignment should return List[Device]
Previously return List[List[Device]] which is not how the function is used.
Updated to use the alternative overload. |
260,339 | 20.02.2022 09:46:49 | 21,600 | 4a8de96e88e7772a4a64ef96be0dd027e40376ee | Update README.md to fix link to prng.md | [
{
"change_type": "MODIFY",
"old_path": "README.md",
"new_path": "README.md",
"diff": "@@ -357,7 +357,7 @@ Some standouts:\n1. [In-place mutating updates of\narrays](https://jax.readthedocs.io/en/latest/notebooks/Common_Gotchas_in_JAX.html#in-place-updates), like `x[i] += y`, aren't supported, but [t... | Python | Apache License 2.0 | google/jax | Update README.md to fix link to prng.md |
260,527 | 21.02.2022 07:50:49 | 0 | add6c8254e03402d20c2e114616ea68cf789900a | Clarify tree_flatten docstring. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/tree_util.py",
"new_path": "jax/_src/tree_util.py",
"diff": "@@ -43,10 +43,9 @@ def tree_flatten(tree, is_leaf: Optional[Callable[[Any], bool]] = None):\nArgs:\ntree: a pytree to flatten.\nis_leaf: an optionally specified function that will be called ... | Python | Apache License 2.0 | google/jax | Clarify tree_flatten docstring. |
260,626 | 22.02.2022 13:49:48 | -14,400 | 6423741db02139d17d9c46e58350eb662b2ebd13 | Add copy button to code snippets in documentation | [
{
"change_type": "MODIFY",
"old_path": "docs/conf.py",
"new_path": "docs/conf.py",
"diff": "@@ -70,6 +70,7 @@ extensions = [\n'sphinx_autodoc_typehints',\n'myst_nb',\n\"sphinx_remove_toctrees\",\n+ 'sphinx_copybutton',\n'jax_extensions',\n]\n"
},
{
"change_type": "MODIFY",
"old_path": "d... | Python | Apache License 2.0 | google/jax | Add copy button to code snippets in documentation |
260,287 | 22.02.2022 06:01:36 | 28,800 | 2641f06152f3d730a1d09851a2d3b3f87e0c1408 | Always treat all mesh axes controlled by xmap as MANUAL | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/maps.py",
"new_path": "jax/experimental/maps.py",
"diff": "@@ -731,10 +731,12 @@ def make_xmap_callable(fun: lu.WrappedFun,\nfor ax, av, ips in safe_zip(mesh_in_axes, in_avals, in_positional_semantics)\n]\nin_is_gda = [ips == _PositionalSemant... | Python | Apache License 2.0 | google/jax | Always treat all mesh axes controlled by xmap as MANUAL
PiperOrigin-RevId: 430192736 |
260,335 | 22.02.2022 10:31:05 | 28,800 | 4b1d0a466b0b330ecb4285a0a80920497343b429 | fixing scan and other control flow | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/checkify/checkify_impl.py",
"new_path": "jax/experimental/checkify/checkify_impl.py",
"diff": "@@ -540,12 +540,13 @@ error_checks[lax.cond_p] = cond_error_check\ndef scan_error_check(error, enabled_errors, *in_flat, reverse, length, jaxpr,\nnu... | Python | Apache License 2.0 | google/jax | fixing scan and other control flow
Co-authored-by: Lena Martens <lenamartens@google.com> |
260,424 | 22.02.2022 23:08:22 | 0 | 45d3ddda3193c2006a1471d3a949f6d91683e1a7 | Fix tests and handle cond consts. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/checkify/checkify_impl.py",
"new_path": "jax/experimental/checkify/checkify_impl.py",
"diff": "@@ -553,12 +553,13 @@ def scan_error_check(error, enabled_errors, *in_flat, reverse, length, jaxpr,\nreturn outs, Error(err, code, new_msgs)\nerror_... | Python | Apache License 2.0 | google/jax | Fix tests and handle cond consts. |
260,631 | 24.02.2022 07:22:52 | 28,800 | 8372b98c4856b6b2363b7bb28abdb4579440a656 | [JAX] Move ann.ann_recall back to tests.
The function is simple enough for users to implement their own on the host. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/ann.py",
"new_path": "jax/experimental/ann.py",
"diff": "@@ -212,45 +212,6 @@ def approx_min_k(operand: Array,\naggregate_to_topk=aggregate_to_topk)\n-def ann_recall(result_neighbors: Array, ground_truth_neighbors: Array) -> float:\n- \"\"\"Co... | Python | Apache License 2.0 | google/jax | [JAX] Move ann.ann_recall back to tests.
The function is simple enough for users to implement their own on the host.
PiperOrigin-RevId: 430696789 |
260,314 | 24.02.2022 22:34:00 | 18,000 | e28ec78d7a0812dd8ec5753d03b52b156131a0fa | Make negative dimension return consistent results for image.scale_and_translate | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/image/scale.py",
"new_path": "jax/_src/image/scale.py",
"diff": "@@ -89,7 +89,10 @@ def _scale_and_translate(x, output_shape: core.Shape,\ncontractions = []\nin_indices = list(range(len(output_shape)))\nout_indices = list(range(len(output_shape)))\n+ ... | Python | Apache License 2.0 | google/jax | Make negative dimension return consistent results for image.scale_and_translate |
260,314 | 25.02.2022 17:39:47 | 18,000 | 0a859453b3a6ba4b71f33cc1eeb15bdd5e00d392 | Use canonicalize_axis | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/image/scale.py",
"new_path": "jax/_src/image/scale.py",
"diff": "@@ -20,6 +20,7 @@ from jax import core\nfrom jax import jit\nfrom jax import lax\nfrom jax import numpy as jnp\n+from jax._src.util import canonicalize_axis\nimport numpy as np\n@@ -91,8... | Python | Apache License 2.0 | google/jax | Use canonicalize_axis |
260,314 | 25.02.2022 21:48:39 | 18,000 | 0b022b71a5d1dd12bb53211c0bfd20fc06b892ff | Fix incorrect implementation to find canonicalize_axis | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/image/scale.py",
"new_path": "jax/_src/image/scale.py",
"diff": "@@ -90,9 +90,8 @@ def _scale_and_translate(x, output_shape: core.Shape,\ncontractions = []\nin_indices = list(range(len(output_shape)))\nout_indices = list(range(len(output_shape)))\n- s... | Python | Apache License 2.0 | google/jax | Fix incorrect implementation to find canonicalize_axis |
260,631 | 01.03.2022 14:24:21 | 28,800 | 1a1bf122d98852cf2803dfdf494da12e023d0c47 | [JAX] Fix github lint_and_typecheck for Primitives | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -265,9 +265,12 @@ Atom = Union[Var, Literal]\nclass Primitive:\nname: str\n- multiple_results = False # set for multi-output primitives\n- call_primitive = False # set for call primitives processed in final ... | Python | Apache License 2.0 | google/jax | [JAX] Fix github lint_and_typecheck for Primitives
PiperOrigin-RevId: 431776305 |
260,631 | 01.03.2022 14:46:04 | 28,800 | d9f82f7b9b51e0c27b2210f1e549518dcd27f498 | [JAX] Move `experimental.ann.approx_*_k` into `lax`.
Updated docs, tests and the example code snippets. | [
{
"change_type": "DELETE",
"old_path": "docs/jax.experimental.ann.rst",
"new_path": null,
"diff": "-jax.experimental.ann module\n-===========================\n-\n-.. automodule:: jax.experimental.ann\n-\n-API\n----\n-\n-.. autosummary::\n- :toctree: _autosummary\n-\n- approx_max_k\n- approx_min_k\n"... | Python | Apache License 2.0 | google/jax | [JAX] Move `experimental.ann.approx_*_k` into `lax`.
Updated docs, tests and the example code snippets.
PiperOrigin-RevId: 431781401 |
260,631 | 02.03.2022 08:58:22 | 28,800 | fb44d7c12f8d1c060d54cfe4a3aed23db5285845 | [JAX] Add release note for the graduration of the experimental.ann module. | [
{
"change_type": "MODIFY",
"old_path": "CHANGELOG.md",
"new_path": "CHANGELOG.md",
"diff": "@@ -16,6 +16,8 @@ PLEASE REMEMBER TO CHANGE THE '..main' WITH AN ACTUAL TAG in GITHUB LINK.\ndeprecated in 0.2.22, have been removed. Please use\n[the `.at` property on JAX arrays](https://jax.readthedocs.io/... | Python | Apache License 2.0 | google/jax | [JAX] Add release note for the graduration of the experimental.ann module.
PiperOrigin-RevId: 431951602 |
260,631 | 02.03.2022 10:11:09 | 28,800 | f1e71c11d729e0f16e84e76fe85051d223e9bb0b | [Jax] Format ann docstring. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/ann.py",
"new_path": "jax/_src/lax/ann.py",
"diff": "@@ -97,21 +97,21 @@ def approx_max_k(operand: Array,\nreduction_dimension : Integer dimension along which to search. Default: -1.\nrecall_target : Recall target for the approximation.\nreduction... | Python | Apache License 2.0 | google/jax | [Jax] Format ann docstring.
PiperOrigin-RevId: 431968329 |
260,578 | 03.03.2022 12:19:04 | 28,800 | f0df7b7348debbc6db314e6387756f4f7582f568 | Fix the display of GDA docs | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.experimental.global_device_array.rst",
"new_path": "docs/jax.experimental.global_device_array.rst",
"diff": "@@ -6,8 +6,6 @@ jax.experimental.global_device_array module\nAPI\n---\n-.. autosummary::\n- :toctree: _autosummary\n-\n- GlobalDeviceArray\n- ... | Python | Apache License 2.0 | google/jax | Fix the display of GDA docs |
260,578 | 03.03.2022 13:40:42 | 28,800 | 852e39e2edb6e82a21ac07e352a65599881467ce | convert to doctest | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/global_device_array.py",
"new_path": "jax/experimental/global_device_array.py",
"diff": "@@ -219,40 +219,51 @@ class GlobalDeviceArray:\nis_fully_replicated : True if the full array value is present on all devices\nof the global mesh.\n- Examp... | Python | Apache License 2.0 | google/jax | convert to doctest |
260,447 | 03.03.2022 20:58:46 | 28,800 | 2d17b4d637f3db7a707ae8d7ef167cfdfe1b0f68 | [linalg] Fix type promotion in QDWH. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/qdwh.py",
"new_path": "jax/_src/lax/qdwh.py",
"diff": "@@ -36,7 +36,7 @@ def _use_qr(u, params):\n\"\"\"Uses QR decomposition.\"\"\"\na, b, c = params\nm, n = u.shape\n- y = jnp.concatenate([jnp.sqrt(c) * u, jnp.eye(n)])\n+ y = jnp.concatenate([jn... | Python | Apache License 2.0 | google/jax | [linalg] Fix type promotion in QDWH.
PiperOrigin-RevId: 432352839 |
260,403 | 04.03.2022 04:20:57 | 28,800 | f6a5f0dca2d19d621e68d15e501f47a109ccd117 | Use real Precision type for lax.PrecisionType | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/lax.py",
"new_path": "jax/_src/lax/lax.py",
"diff": "@@ -19,7 +19,7 @@ from functools import partial\nimport itertools\nimport operator\nfrom typing import (Any, Callable, Optional, Sequence, Tuple, List, TypeVar,\n- Union)\n+ Union, cast as type_... | Python | Apache License 2.0 | google/jax | Use real Precision type for lax.PrecisionType
PiperOrigin-RevId: 432413742 |
260,531 | 04.03.2022 20:22:59 | -28,800 | ee2377b47d259a59cf7b530852a9414d993dbf33 | [Doc] Add jaxlib build instructions for TPU
Currently the documentation only contains build instructions for CUDA support, but does not mention TPU. This pull request add build instructions for TPU alongside CUDA. | [
{
"change_type": "MODIFY",
"old_path": "docs/developer.md",
"new_path": "docs/developer.md",
"diff": "@@ -53,26 +53,22 @@ You can install the necessary Python dependencies using `pip`:\npip install numpy six wheel\n```\n-To build `jaxlib` with CUDA support, you can run:\n+To build `jaxlib` without C... | Python | Apache License 2.0 | google/jax | [Doc] Add jaxlib build instructions for TPU
Currently the documentation only contains build instructions for CUDA support, but does not mention TPU. This pull request add build instructions for TPU alongside CUDA. |
260,578 | 04.03.2022 11:59:16 | 28,800 | 403eae0efee064e023881f69e38818d8830585fc | format nicely | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/global_device_array.py",
"new_path": "jax/experimental/global_device_array.py",
"diff": "@@ -411,12 +411,15 @@ class GlobalDeviceArray:\n>>> from jax.experimental.maps import Mesh\n>>> import numpy as np\n+ ...\n>>> global_input_shape = (8, 8)... | Python | Apache License 2.0 | google/jax | format nicely |
260,335 | 07.03.2022 12:44:50 | 28,800 | 24a7afdbf4655349950cd4a2575f3f8b6e960c14 | improve batch_jaxpr caching from
In I missed a related utilty function which needed memoization. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/batching.py",
"new_path": "jax/interpreters/batching.py",
"diff": "@@ -448,7 +448,6 @@ def batch_jaxpr(closed_jaxpr, axis_size, in_batched, instantiate, axis_name,\nreturn _batch_jaxpr(closed_jaxpr, axis_size, tuple(in_batched), inst,\naxis_na... | Python | Apache License 2.0 | google/jax | improve batch_jaxpr caching from #9196
In #9196 I missed a related utilty function which needed memoization.
Co-authored-by: Adam Paszke <apaszke@google.com> |
260,308 | 09.03.2022 17:05:28 | 28,800 | caf094d06be78fbe069eb627181cab91225ef472 | Support gamma distribution with PRNGKeys other than threefry2x32. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/random.py",
"new_path": "jax/_src/random.py",
"diff": "@@ -940,15 +940,14 @@ def _gamma_grad(sample, a):\ngrads = vmap(lax.random_gamma_grad)(alphas, samples)\nreturn grads.reshape(np.shape(a))\n-def _gamma_impl(key, a, use_vmap=False):\n+def _gamma_i... | Python | Apache License 2.0 | google/jax | Support gamma distribution with PRNGKeys other than threefry2x32.
PiperOrigin-RevId: 433614014 |
260,510 | 09.03.2022 12:20:28 | 28,800 | 2988901e6ccbd88dcac0720f36a77202162a350d | Refactor Jaxpr pretty-printing to use a `JaxprPpSettings` named tuple
and thread it into `pp_eqn_rules` so the settings are used recursively | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/lax.py",
"new_path": "jax/_src/lax/lax.py",
"diff": "@@ -2298,13 +2298,13 @@ def _convert_elt_type_fwd_rule(eqn):\nelse:\nreturn [None], eqn\n-def _convert_elt_type_pp_rule(eqn, context):\n+def _convert_elt_type_pp_rule(eqn, context, settings):\n#... | Python | Apache License 2.0 | google/jax | Refactor Jaxpr pretty-printing to use a `JaxprPpSettings` named tuple
and thread it into `pp_eqn_rules` so the settings are used recursively |
260,424 | 10.03.2022 13:31:24 | 0 | 76e1021cadc036ace21f076cf52b59820eec25b4 | Checkify: Fix empty enabled_errors case wrt user checks.
What looked like a quick win (short-cut on empty enabled_error, don't trace)
was actually a quick bug (user_checks throw error in eager). | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/checkify/checkify_impl.py",
"new_path": "jax/experimental/checkify/checkify_impl.py",
"diff": "@@ -750,11 +750,6 @@ def checkify(fun: Callable[..., Out],\nValueError: nan generated by primitive sin\n\"\"\"\n- if not errors:\n- def checked_fun_... | Python | Apache License 2.0 | google/jax | Checkify: Fix empty enabled_errors case wrt user checks.
What looked like a quick win (short-cut on empty enabled_error, don't trace)
was actually a quick bug (user_checks throw error in eager). |
260,392 | 09.03.2022 13:03:45 | 18,000 | c5d4aba2a939ae1a2171a33f2e723490d9197b71 | Fix fft dtype for norm='ortho' | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/numpy/fft.py",
"new_path": "jax/_src/numpy/fft.py",
"diff": "@@ -86,7 +86,8 @@ def _fft_core(func_name, fft_type, a, s, axes, norm):\ns += [max(0, 2 * (a.shape[axes[-1]] - 1))]\nelse:\ns = [a.shape[axis] for axis in axes]\n- transformed = lax.fft(a, f... | Python | Apache License 2.0 | google/jax | Fix fft dtype for norm='ortho' |
260,335 | 11.03.2022 11:36:13 | 28,800 | 39c2f8b051bd80cfd6de2a7d965d1b3cb2920456 | fixup from remove extraneous lines
also add test | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/windowed_reductions.py",
"new_path": "jax/_src/lax/windowed_reductions.py",
"diff": "@@ -278,8 +278,6 @@ def _generic_reduce_window_batch_rule(\noperands, init_values = util.split_list(batched_args, [num_operands])\noperand_bdims, init_value_bdims... | Python | Apache License 2.0 | google/jax | fixup from 5415306: remove extraneous lines
also add test |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.