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 |
|---|---|---|---|---|---|---|---|---|---|
260,411 | 16.06.2021 13:25:56 | -10,800 | e405a73663adaa1abf739ba789feb5943f72a54a | [call_tf] Fixed the handling of x64 values in JAX_ENABLE_X64=False
TensorFlow interprets Python scalars in x64 mode, or it may contain
explicit x64 computation. When used with call_tf, we canonicalize
inputs and outputs using the JAX rules. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/call_tf.py",
"new_path": "jax/experimental/jax2tf/call_tf.py",
"diff": "@@ -28,6 +28,7 @@ from typing import Callable, Sequence\nimport jax\nfrom jax import core\nfrom jax import dlpack\n+from jax import dtypes\nfrom jax import numpy as... | Python | Apache License 2.0 | google/jax | [call_tf] Fixed the handling of x64 values in JAX_ENABLE_X64=False
TensorFlow interprets Python scalars in x64 mode, or it may contain
explicit x64 computation. When used with call_tf, we canonicalize
inputs and outputs using the JAX rules. |
260,411 | 15.06.2021 08:37:01 | -10,800 | d394f6917b2c3b938b7f9809e46e1930710d33f7 | Added clipping of OOB indices to match XLA behavior | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/g3doc/primitives_with_limited_support.md",
"new_path": "jax/experimental/jax2tf/g3doc/primitives_with_limited_support.md",
"diff": "# Primitives with limited support for jax2tf\n-*Last generated on (YYYY-MM-DD): 2021-06-14*\n+*Last gene... | Python | Apache License 2.0 | google/jax | Added clipping of OOB indices to match XLA behavior |
260,374 | 02.06.2021 17:05:26 | 0 | 0e9f7de6c839c31ed1a9e5574841865921bb522f | Add LRU eviction policy to persisent compilation cache | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/compilation_cache/file_system_cache.py",
"new_path": "jax/experimental/compilation_cache/file_system_cache.py",
"diff": "# See the License for the specific language governing permissions and\n# limitations under the License.\n-from typing impo... | Python | Apache License 2.0 | google/jax | Add LRU eviction policy to persisent compilation cache |
260,411 | 17.06.2021 14:25:01 | -10,800 | a6c16119c91e61d5e9fbfd8d3bd1dcfe14cb838f | Renaming master to main: step 1.
Here we just enable the GitHub actions to run on both master and main. | [
{
"change_type": "MODIFY",
"old_path": ".github/workflows/ci-build.yaml",
"new_path": ".github/workflows/ci-build.yaml",
"diff": "@@ -2,13 +2,15 @@ name: CI\non:\n# Trigger the workflow on push or pull request,\n- # but only for the master branch\n+ # but only for the main branch\npush:\nbranches:\n... | Python | Apache License 2.0 | google/jax | Renaming master to main: step 1.
Here we just enable the GitHub actions to run on both master and main. |
260,411 | 17.06.2021 14:48:21 | -10,800 | 0ec53eec72b855dbd053b2464ea9b3a0422b6c09 | Test PR merged to main, before making main default
Want to see the GitHub actions working, and copybara import | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -912,6 +912,7 @@ to generate tables of limitations, e.g.,\ne.g., due to unimplemented cases in TensorFlow. This list is incremental\non top of the unsupported JAX ... | Python | Apache License 2.0 | google/jax | Test PR merged to main, before making main default
Want to see the GitHub actions working, and copybara import |
260,411 | 17.06.2021 15:58:51 | -10,800 | 268a242c65f634d5d380fa9357fe9b271d5be9e6 | Test 5 for rename master to main | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -892,6 +892,7 @@ As of today, the tests are run using `tf_nightly==2.6.0-dev20210611`.\n## Running on GPU\n+\nTo run jax2tf on GPU, both jaxlib and TensorFlow must... | Python | Apache License 2.0 | google/jax | Test 5 for rename master to main |
260,411 | 17.06.2021 16:08:54 | -10,800 | 4fa1506825337c5fff343e27299ffbd94e02d980 | One more change, to tickle copybara | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -890,6 +890,7 @@ in [call_tf_test.py](https://github.com/google/jax/blob/master/jax/experimental/\nThe ``jax2tf.convert`` and `call_tf` require very recent version... | Python | Apache License 2.0 | google/jax | One more change, to tickle copybara |
260,563 | 17.06.2021 19:25:46 | -7,200 | ccc8bb7f19321402caa9c8d6a02999f74c795ea8 | Add auxiliary data support to lax.custom_root | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/control_flow.py",
"new_path": "jax/_src/lax/control_flow.py",
"diff": "@@ -2018,7 +2018,7 @@ def _split_root_args(args, const_lengths):\n@api_boundary\n-def custom_root(f, initial_guess, solve, tangent_solve):\n+def custom_root(f, initial_guess, s... | Python | Apache License 2.0 | google/jax | Add auxiliary data support to lax.custom_root |
260,563 | 17.06.2021 20:26:16 | -7,200 | cba5b13a39028d4389e063d6284cf68ce67afbb2 | Improve concretization errors for jnp indexing routines | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/numpy/lax_numpy.py",
"new_path": "jax/_src/numpy/lax_numpy.py",
"diff": "@@ -3331,6 +3331,21 @@ def meshgrid(*args, **kwargs):\nreturn output\n+def _make_1d_grid_from_slice(s: slice, op_name: str):\n+ start =core.concrete_or_error(None, s.start,\n+ f\... | Python | Apache License 2.0 | google/jax | Improve concretization errors for jnp indexing routines |
260,372 | 11.06.2021 16:40:39 | 21,600 | c33388b1367dd0206c1fef9c1c22ca6a0e0e04ee | Support complex numbers in jax.scipy.signal.convolve/correlate | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/scipy/signal.py",
"new_path": "jax/_src/scipy/signal.py",
"diff": "@@ -66,8 +66,6 @@ def convolve(in1, in2, mode='full', method='auto',\nprecision=None):\nif method != 'auto':\nwarnings.warn(\"convolve() ignores method argument\")\n- if jnp.issubdtype... | Python | Apache License 2.0 | google/jax | Support complex numbers in jax.scipy.signal.convolve/correlate |
260,374 | 04.06.2021 19:22:57 | 0 | a87a291afd35daffd0154e64951e772842998559 | Implemented get_cache_key funtion with unit tests | [
{
"change_type": "ADD",
"old_path": null,
"new_path": "jax/experimental/compilation_cache/compilation_cache.py",
"diff": "+# Copyright 2021 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+# Yo... | Python | Apache License 2.0 | google/jax | Implemented get_cache_key funtion with unit tests |
260,411 | 23.06.2021 10:52:03 | -7,200 | 7e335e0e2e31057843c5c0352ebf5bc1a88faa8d | [jax2tf] Fix conversion of gradients for shape polymorphic functions.
This fixes the case when the primal shape polymorphic function has
output shapes that are polynomials of the input shapes (not just
dimension variables). | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -1080,7 +1080,7 @@ class ShapedArray(UnshapedArray):\nself.weak_type, self.named_shape)\ndef join(self, other):\n- if self.shape == other.shape and self.dtype == other.dtype:\n+ if symbolic_equal_shape(self.... | Python | Apache License 2.0 | google/jax | [jax2tf] Fix conversion of gradients for shape polymorphic functions.
This fixes the case when the primal shape polymorphic function has
output shapes that are polynomials of the input shapes (not just
dimension variables). |
260,374 | 23.06.2021 17:40:23 | 0 | d460a53a5664d17a977df124d0025d7ed163d24e | changed 1 to 0 | [
{
"change_type": "MODIFY",
"old_path": "tests/compilation_cache_test.py",
"new_path": "tests/compilation_cache_test.py",
"diff": "@@ -124,7 +124,7 @@ class CompilationCacheTest(jtu.JaxTestCase):\ndebug_options_obj.xla_cpu_fast_math_honor_division = False\ndebug_options_obj.xla_cpu_fast_math_honor_fu... | Python | Apache License 2.0 | google/jax | changed 1 to 0 |
260,411 | 24.06.2021 08:40:41 | -7,200 | 174a4779a1fb624bd6abb727b48a0871b5ebf56f | [jax2tf] Documented recipe for ensuring that model parameters are not embedded | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -94,7 +94,8 @@ is trivial:\n# You can save the model just like you would with any other TensorFlow function:\nmy_model = tf.Module()\n# Save a function that can ta... | Python | Apache License 2.0 | google/jax | [jax2tf] Documented recipe for ensuring that model parameters are not embedded |
260,411 | 24.06.2021 09:33:29 | -7,200 | 76a3c88c451cfbe796f00c382899e271e70a447f | Improved handling of tf.TensorShape | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -585,7 +585,6 @@ def _args_to_avals_and_env(\nfor i, d in enumerate(aval_shape):\nif not shape_poly.is_poly_dim(d):\n- assert isinstance(d, int)\nassert d == arg_s... | Python | Apache License 2.0 | google/jax | Improved handling of tf.TensorShape |
260,411 | 28.06.2021 08:40:43 | -7,200 | 44b95426d1760c9a8cb3f69e816192eb979a1c87 | [jax2tf] Fix the conversion of reduce_sum and reduce_prod for booleans
Also update the documentation | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/g3doc/jax_primitives_coverage.md",
"new_path": "jax/experimental/jax2tf/g3doc/jax_primitives_coverage.md",
"diff": "# Primitives with limited JAX support\n-*Last generated on: 2021-06-14* (YYYY-MM-DD)\n+*Last generated on: 2021-06-28* (... | Python | Apache License 2.0 | google/jax | [jax2tf] Fix the conversion of reduce_sum and reduce_prod for booleans
Also update the documentation |
260,374 | 17.06.2021 17:40:32 | 0 | 0ea0525740c647747a767f1daf0c04add6b53844 | implementing thread saftey check | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/compilation_cache/file_system_cache.py",
"new_path": "jax/experimental/compilation_cache/file_system_cache.py",
"diff": "# limitations under the License.\nimport os\n+import tempfile\nfrom typing import Optional\nimport warnings\n@@ -41,8 +42,... | Python | Apache License 2.0 | google/jax | implementing thread saftey check |
260,411 | 29.06.2021 14:57:42 | -10,800 | ffd8fb84b48ff658713f5b0aeea84b2255bd554c | [jax2tf] A few fixed for handling of float0 in jax2tf and call_tf
TF returns None or 0 for the gradients of functions with integer
arguments. JAX expects float0. We must convert to and from float0
at the JAX-TF boundary. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -589,6 +589,47 @@ jax2tf.convert(jax.grad(f_jax, allow_int=True))(2))\n# returns a `bfloat16` zero: tf.Tensor(0, shape=(), dtype=bfloat16)\n```\n+### Different beh... | Python | Apache License 2.0 | google/jax | [jax2tf] A few fixed for handling of float0 in jax2tf and call_tf
TF returns None or 0 for the gradients of functions with integer
arguments. JAX expects float0. We must convert to and from float0
at the JAX-TF boundary. |
260,335 | 24.06.2021 15:00:19 | 25,200 | a0eb1126e4d3ce74135200468956ecc3bb26ebb8 | remat: don't apply cse-foiling widget to primal | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/api.py",
"new_path": "jax/_src/api.py",
"diff": "@@ -2408,7 +2408,8 @@ def eval_shape(fun: Callable, *args, **kwargs):\nreturn tree_unflatten(out_tree(), out)\n-def checkpoint(fun: Callable, concrete: bool = False) -> Callable:\n+def checkpoint(fun: C... | Python | Apache License 2.0 | google/jax | remat: don't apply cse-foiling widget to primal |
260,335 | 02.07.2021 10:31:28 | 25,200 | 5e92faccbb4497755c8cee901b57b5e2788fabb0 | tweak xla_bridge.py flags
* add environment variables for jax_disable_most_optimizations and
jax_cpu_backend_variant
* comment on the default values in help strings | [
{
"change_type": "MODIFY",
"old_path": "jax/lib/xla_bridge.py",
"new_path": "jax/lib/xla_bridge.py",
"diff": "@@ -30,7 +30,7 @@ from absl import logging\nlogging._warn_preinit_stderr = 0\nimport jax.lib\n-from .._src.config import flags\n+from .._src.config import flags, bool_env\nfrom . import tpu_... | Python | Apache License 2.0 | google/jax | tweak xla_bridge.py flags
* add environment variables for jax_disable_most_optimizations and
jax_cpu_backend_variant
* comment on the default values in help strings |
260,447 | 02.07.2021 10:42:29 | 25,200 | d97b393694570a35fb0a092be63a30adf845c823 | Adds spherical harmonics. | [
{
"change_type": "MODIFY",
"old_path": "CHANGELOG.md",
"new_path": "CHANGELOG.md",
"diff": "@@ -10,6 +10,8 @@ PLEASE REMEMBER TO CHANGE THE '..main' WITH AN ACTUAL TAG in GITHUB LINK.\n## jax 0.2.17 (unreleased)\n* [GitHub commits](https://github.com/google/jax/compare/jax-v0.2.16...main).\n+* New f... | Python | Apache License 2.0 | google/jax | Adds spherical harmonics.
Co-authored-by: Jake VanderPlas <jakevdp@google.com> |
260,635 | 03.07.2021 18:09:58 | -7,200 | 7b10965794be4f592d2ad1975aa45d7dde32359c | Implement complex logaddexp | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/numpy/lax_numpy.py",
"new_path": "jax/_src/numpy/lax_numpy.py",
"diff": "@@ -656,45 +656,65 @@ def power(x1, x2):\n@custom_jvp\n@_wraps(np.logaddexp)\ndef logaddexp(x1, x2):\n- x1, x2 = _promote_shapes(\"logaddexp\", *_promote_dtypes_inexact(x1, x2))\... | Python | Apache License 2.0 | google/jax | Implement complex logaddexp |
260,286 | 05.07.2021 11:59:27 | -3,600 | 4bae17dd703fafbc07ccb5b096af06ce6fac3748 | Add a helper decorator to disable implicit rank promotion in unit tests. | [
{
"change_type": "MODIFY",
"old_path": "jax/test_util.py",
"new_path": "jax/test_util.py",
"diff": "@@ -855,6 +855,20 @@ class JaxTestLoader(absltest.TestLoader):\nreturn names\n+def disable_implicit_rank_promotion(test_method):\n+ \"\"\"A decorator for test methods to raise error on implicit rank p... | Python | Apache License 2.0 | google/jax | Add a helper decorator to disable implicit rank promotion in unit tests. |
260,411 | 05.07.2021 18:17:59 | -7,200 | 2fe02268b245f50cf9c656783da5c34ec21437b0 | [jax2tf] Ensure that the conversion of Round uses the same algorithm as JAX | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/g3doc/jax_primitives_coverage.md",
"new_path": "jax/experimental/jax2tf/g3doc/jax_primitives_coverage.md",
"diff": "# Primitives with limited JAX support\n-*Last generated on: 2021-06-28* (YYYY-MM-DD)\n+*Last generated on: 2021-07-05* (... | Python | Apache License 2.0 | google/jax | [jax2tf] Ensure that the conversion of Round uses the same algorithm as JAX |
260,286 | 30.06.2021 13:52:32 | -3,600 | f945982b8ce9b95d47edd26afb413fa1c9baca99 | Broadcast arrays manually in categorical sampling. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/random.py",
"new_path": "jax/_src/random.py",
"diff": "@@ -1306,7 +1306,10 @@ def categorical(key: jnp.ndarray,\n_check_shape(\"categorical\", shape, batch_shape)\nsample_shape = shape[:len(shape)-len(batch_shape)]\n- return jnp.argmax(gumbel(key, sam... | Python | Apache License 2.0 | google/jax | Broadcast arrays manually in categorical sampling. |
260,671 | 09.07.2021 02:23:28 | -19,080 | c60115ab9d428c0e3dfa966e8fc734e6bb2619a8 | adding s_ and index_exp | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/numpy/lax_numpy.py",
"new_path": "jax/_src/numpy/lax_numpy.py",
"diff": "@@ -3633,9 +3633,11 @@ class CClass(_AxisConcat):\ntrans1d = 0\nop_name = \"c_\"\n-\nc_ = CClass()\n+s_ = np.s_\n+\n+index_exp = np.index_exp\n@_wraps(np.i0)\ndef i0(x):\n"
},
... | Python | Apache License 2.0 | google/jax | adding s_ and index_exp |
260,411 | 10.07.2021 19:08:15 | -10,800 | 5520fcb59f4cbd1a73428f2c2ea13ced2404f761 | Improve error message when vjp is called with cotangent of wrong shape.
Previously the error was an internal assertion error. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/api.py",
"new_path": "jax/_src/api.py",
"diff": "@@ -1888,19 +1888,24 @@ def _lift_linearized(jaxpr, primal_avals, io_tree, out_pvals, consts, *py_args):\nreturn apply_flat_fun(fun, io_tree, *py_args)\n-def _vjp_pullback_wrapper(cotangent_dtypes, io_t... | Python | Apache License 2.0 | google/jax | Improve error message when vjp is called with cotangent of wrong shape.
Previously the error was an internal assertion error. |
260,411 | 11.07.2021 10:09:19 | -10,800 | 1f946ad51ee5506b9f7bf6bbbeeb1f6e59074b27 | Fix grad of conv 0D.
This bug was introduced in and was not caught by existing tests.
Add a reproducing test. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/lax.py",
"new_path": "jax/_src/lax/lax.py",
"diff": "@@ -6921,6 +6921,9 @@ def _conv_general_vjp_lhs_padding(\ndef _conv_general_vjp_rhs_padding(\nin_shape, window_dimensions, window_strides, out_shape, padding,\nlhs_dilation, rhs_dilation):\n+\n+... | Python | Apache License 2.0 | google/jax | Fix grad of conv 0D.
This bug was introduced in #6345, and was not caught by existing tests.
Add a reproducing test. |
260,411 | 13.07.2021 10:39:38 | -10,800 | caba2ed9b80f910d41f92f7c49f9ade5f198492b | [jax2tf] Support tf.compat.v1.Dimension when parsing polymorphic_shapes | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/shape_poly.py",
"new_path": "jax/experimental/jax2tf/shape_poly.py",
"diff": "@@ -31,6 +31,7 @@ import opt_einsum\nfrom jax import config\nfrom jax import core\nimport numpy as np\n+import tensorflow as tf # type: ignore[import]\nDimSiz... | Python | Apache License 2.0 | google/jax | [jax2tf] Support tf.compat.v1.Dimension when parsing polymorphic_shapes |
260,374 | 07.07.2021 21:42:34 | 0 | 052062f1638c071dba3f3f7c26989af69e349d10 | initialize_cache, get_executable, and put_executable functions | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/compilation_cache/compilation_cache.py",
"new_path": "jax/experimental/compilation_cache/compilation_cache.py",
"diff": "import hashlib\nimport jax\n+from jax.experimental.compilation_cache.file_system_cache import FileSystemCache\n+from jax.l... | Python | Apache License 2.0 | google/jax | initialize_cache, get_executable, and put_executable functions |
260,287 | 14.07.2021 11:39:52 | 0 | 1c1ec79edd74e5ceb6e51c4a8c7d42ff7811803b | Clarify the error message for out-of-bounds in_axes in pmap and vmap
Fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/api.py",
"new_path": "jax/_src/api.py",
"diff": "@@ -1320,16 +1320,17 @@ def vmap(fun: F, in_axes=0, out_axes=0, axis_name=None) -> F:\nreturn batched_fun\ndef _mapped_axis_size(tree, vals, dims, name, *, kws=False):\n- def _get_axis_size(name: str, i... | Python | Apache License 2.0 | google/jax | Clarify the error message for out-of-bounds in_axes in pmap and vmap
Fixes #5201. |
260,287 | 14.07.2021 06:24:48 | 25,200 | 1049e7205f2cc8a0a3969f8f44ee77eabe19c3d1 | Implement vmap rules for with_sharding_constraint
pjit already supports batching, so there's no need to hold off on that. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/pjit.py",
"new_path": "jax/experimental/pjit.py",
"diff": "@@ -265,6 +265,7 @@ class SpecSync(IntEnum):\nWe use this to make sure we don't show garbage modified values while claiming\nthat the users have specified them like that.\n\"\"\"\n+ OU... | Python | Apache License 2.0 | google/jax | Implement vmap rules for with_sharding_constraint
pjit already supports batching, so there's no need to hold off on that.
PiperOrigin-RevId: 384684263 |
260,631 | 14.07.2021 11:25:37 | 25,200 | b9aaff48702bb2e2e6910f5750b771b4eb1212dd | Parse absl flags in compilation_cache_test.py | [
{
"change_type": "MODIFY",
"old_path": "tests/compilation_cache_test.py",
"new_path": "tests/compilation_cache_test.py",
"diff": "@@ -23,6 +23,10 @@ import tempfile\nimport unittest\nfrom unittest import SkipTest\n+from jax.config import config\n+config.parse_flags_with_absl()\n+FLAGS = config.FLAGS... | Python | Apache License 2.0 | google/jax | Parse absl flags in compilation_cache_test.py
PiperOrigin-RevId: 384744462 |
260,287 | 15.07.2021 04:22:08 | 25,200 | 64510bd5b69933cb6bec62f63b9b7ba2f1eff3fc | Add `axis` and `tiled` options to `lax.all_gather`.
This is especially convenient when using JAX as an HLO generator, because the
HLO AllGather defaults to the tiling behavior. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/parallel.py",
"new_path": "jax/_src/lax/parallel.py",
"diff": "@@ -954,13 +954,7 @@ batching.collective_rules[all_to_all_p] = _all_to_all_batched_collective\ncore.axis_substitution_rules[all_to_all_p] = partial(_subst_all_names_in_param, 'axis_nam... | Python | Apache License 2.0 | google/jax | Add `axis` and `tiled` options to `lax.all_gather`.
This is especially convenient when using JAX as an HLO generator, because the
HLO AllGather defaults to the tiling behavior.
PiperOrigin-RevId: 384897270 |
260,287 | 15.07.2021 13:10:05 | 0 | 478828a21a0f1a6cce200ebb3f2d1970f916dc37 | Add support for reverse-mode AD of xmap | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/maps.py",
"new_path": "jax/experimental/maps.py",
"diff": "@@ -43,7 +43,7 @@ from ..lib import xla_bridge as xb\nfrom ..lib import xla_client as xc\nfrom .._src.util import (safe_map, safe_zip, HashableFunction,\nas_hashable_function, unzip2, ... | Python | Apache License 2.0 | google/jax | Add support for reverse-mode AD of xmap |
260,687 | 16.07.2021 02:36:45 | -19,080 | 61949b6f8b78687a1960c17593876cef4d6034b1 | fix small typo in docs/developer.md | [
{
"change_type": "MODIFY",
"old_path": "docs/developer.md",
"new_path": "docs/developer.md",
"diff": "@@ -131,7 +131,7 @@ sets up symbolic links from site-packages into the repository.\nTo run all the JAX tests, we recommend using `pytest-xdist`, which can run tests in\nparallel. First, install `pyt... | Python | Apache License 2.0 | google/jax | fix small typo in docs/developer.md |
260,299 | 16.07.2021 12:24:19 | -7,200 | 6ca775b10aa85ebfe31af1fec3db693f57f48c93 | Add TPU precision details to README gotchas | [
{
"change_type": "MODIFY",
"old_path": "README.md",
"new_path": "README.md",
"diff": "@@ -368,6 +368,10 @@ Some standouts:\ndouble-precision](https://jax.readthedocs.io/en/latest/notebooks/Common_Gotchas_in_JAX.html#Double-(64bit)-precision)\n(64-bit, e.g. `float64`) one needs to set the `jax_enable... | Python | Apache License 2.0 | google/jax | Add TPU precision details to README gotchas |
260,411 | 16.07.2021 17:23:34 | -10,800 | bd77a61d3170e72c5b8d3279fffacb058457c1dd | [jax2tf] Fix the shape polymorphism for batched while, and random_gamma. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -2634,16 +2634,15 @@ def _batched_cond_while(*args: TfVal, cond_nconsts: int,\nnew_carry: Sequence[TfVal] = _interpret_jaxpr(body_jaxpr, *body_consts,\n*carry,\nex... | Python | Apache License 2.0 | google/jax | [jax2tf] Fix the shape polymorphism for batched while, and random_gamma. |
260,374 | 08.07.2021 16:11:39 | 0 | 262b10ee5967fc2ae22d90359adcbfa74ab58098 | pmap integration | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/compilation_cache/compilation_cache.py",
"new_path": "jax/experimental/compilation_cache/compilation_cache.py",
"diff": "@@ -28,7 +28,7 @@ def initialize_cache(path):\ndef get_executable(xla_computation, compile_options) -> Optional[xla_client... | Python | Apache License 2.0 | google/jax | pmap integration |
260,411 | 16.07.2021 16:52:15 | -10,800 | 41d46b2a9143a2331802ea5055b63948fe97b451 | [jax2tf] Expand the handling of shape-polymorphic einsum.
einsum supports an API where the arrays are interleaved with list
of indices. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/shape_poly.py",
"new_path": "jax/experimental/jax2tf/shape_poly.py",
"diff": "@@ -435,7 +435,8 @@ def _einsum_contract_path(*operands, **kwargs):\n# sizes of operands and intermediate results.\nfake_ops = []\nfor operand in operands:\n-... | Python | Apache License 2.0 | google/jax | [jax2tf] Expand the handling of shape-polymorphic einsum.
einsum supports an API where the arrays are interleaved with list
of indices. |
260,456 | 17.07.2021 08:04:56 | 0 | ee08acd046e6065aacf5a0e94bbdf76000f63400 | update rocblas because of | [
{
"change_type": "MODIFY",
"old_path": "jaxlib/rocblas.cc",
"new_path": "jaxlib/rocblas.cc",
"diff": "@@ -67,7 +67,7 @@ template <>\nif (stream) {\nThrowIfErrorStatus(rocblas_set_stream(handle, stream));\n}\n- return rocBlasHandlePool::Handle(pool, handle);\n+ return rocBlasHandlePool::Handle(pool, ... | Python | Apache License 2.0 | google/jax | update rocblas because of PR-7306 |
260,411 | 18.07.2021 19:35:14 | -10,800 | 117d0d23ab26a5e1585f2027bc90f75393e23fb7 | Attempt to fix RTD build
It seems that the failure is for transformations.md | [
{
"change_type": "MODIFY",
"old_path": "docs/conf.py",
"new_path": "docs/conf.py",
"diff": "@@ -117,6 +117,8 @@ exclude_patterns = [\n'notebooks/*.md',\n'jax-101/*.md',\n'autodidax.md',\n+ # Attempt to fix RTD build failure\n+ 'transformations.md',\n]\n# The name of the Pygments (syntax highlighting... | Python | Apache License 2.0 | google/jax | Attempt to fix RTD build
It seems that the failure is for transformations.md |
260,411 | 18.07.2021 19:03:59 | -10,800 | 0693c316e4044a6bb425da69e7e79364162b3e59 | [jax2tf] Improved error checking for inconsistent use of a dimension variable
Previously the check was done only for multiple occurrences of a shape
variable in one argument. Now we check across all arguments. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -711,7 +711,7 @@ self.assertEqual(0, f_jax(x0)) # JAX sees that the x.shape[0] == 0\n# jax2tf catches the broken assumption b >= 1 if the converted function is exe... | Python | Apache License 2.0 | google/jax | [jax2tf] Improved error checking for inconsistent use of a dimension variable
Previously the check was done only for multiple occurrences of a shape
variable in one argument. Now we check across all arguments. |
260,374 | 19.07.2021 21:53:27 | 0 | 711c3130261f26fcdf570c28dd152f7f54d406c6 | added tearDown() method to test class | [
{
"change_type": "MODIFY",
"old_path": "tests/compilation_cache_test.py",
"new_path": "tests/compilation_cache_test.py",
"diff": "@@ -38,6 +38,10 @@ class CompilationCacheTest(jtu.JaxTestCase):\nif jax.lib.xla_bridge.get_backend().runtime_type == \"tfrt\":\nraise SkipTest(\"the new TFRT runtime does... | Python | Apache License 2.0 | google/jax | added tearDown() method to test class |
260,374 | 19.07.2021 19:10:16 | 0 | 2eaadd86b6c7249e5ea09b6a23ce7e79fa9d8b7d | jit integration with persistent compilation cache | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/pxla.py",
"new_path": "jax/interpreters/pxla.py",
"diff": "@@ -59,20 +59,6 @@ from . import partial_eval as pe\nfrom . import xla\nfrom . import ad\n-def compile_or_get_cached(backend, computation, compile_options):\n- # Avoid import cycle bet... | Python | Apache License 2.0 | google/jax | jit integration with persistent compilation cache |
260,374 | 20.07.2021 16:16:21 | 0 | 65c6a6a6fb0b771cf93512fe06c0b2c62f8395b6 | xmap and pjit integration | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/pxla.py",
"new_path": "jax/interpreters/pxla.py",
"diff": "@@ -1569,7 +1569,7 @@ def compile_and_wrap_mesh_hlo(computation: xc.XlaComputation, backend,\nreturn backend.compile_replicated(computation, compile_options,\ninput_indices, local_inpu... | Python | Apache License 2.0 | google/jax | xmap and pjit integration |
260,424 | 21.07.2021 10:29:46 | -3,600 | 24c9a933d6b00c8690b52c9ee03dc5a5dc435011 | Add shape and dtype of leaked tracer to UnexpectedTracerError. | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -445,7 +445,8 @@ def escaped_tracer_error(tracer, detail=None):\npass\nelse:\nmsg += ('\\nThe tracer that caused this error was created on line '\n- f'{source_info_util.summarize(line_info)}.\\n')\n+ f'{sour... | Python | Apache License 2.0 | google/jax | Add shape and dtype of leaked tracer to UnexpectedTracerError. |
260,333 | 21.07.2021 14:50:10 | 14,400 | 215e7d618fcec2f0a9d31adb1c13dae4c73c7cf8 | Make jax arrays not pass isinstance checks for hashable. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/xla.py",
"new_path": "jax/interpreters/xla.py",
"diff": "@@ -1311,11 +1311,6 @@ for device_array in [DeviceArray]:\n# clobbered when jax.numpy is imported, but useful in tests\nsetattr(device_array, \"__eq__\", lambda self, other: self._value ... | Python | Apache License 2.0 | google/jax | Make jax arrays not pass isinstance checks for hashable. |
260,333 | 21.07.2021 16:12:52 | 14,400 | 00a273d1fe46f2b8fd643377bac72b6c6188eabc | Add assert raises check. | [
{
"change_type": "MODIFY",
"old_path": "tests/api_test.py",
"new_path": "tests/api_test.py",
"diff": "@@ -2077,6 +2077,8 @@ class APITest(jtu.JaxTestCase):\nrep = jnp.ones(()) + 1.\nself.assertIsInstance(rep, jax.interpreters.xla.DeviceArray)\nself.assertNotIsInstance(rep, collections.Hashable)\n+ w... | Python | Apache License 2.0 | google/jax | Add assert raises check. |
260,333 | 21.07.2021 17:13:40 | 14,400 | 231b8cbc497f99e138766fc2e953b02209bb5b9d | Explicitly set array to be unhashable. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/xla.py",
"new_path": "jax/interpreters/xla.py",
"diff": "@@ -1308,6 +1308,9 @@ for device_array in [DeviceArray]:\nsetattr(device_array, \"__reduce__\",\npartialmethod(_forward_to_value, op.methodcaller(\"__reduce__\")))\n+ # explicitly set to... | Python | Apache License 2.0 | google/jax | Explicitly set array to be unhashable. |
260,374 | 21.07.2021 22:22:40 | 0 | 86985fd4c18c0e712620d283540ea51f809db69d | add parameter to initalize_cache function and added __init__ file | [
{
"change_type": "ADD",
"old_path": null,
"new_path": "jax/experimental/compilation_cache/__init__.py",
"diff": "+# Copyright 2021 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 obt... | Python | Apache License 2.0 | google/jax | add parameter to initalize_cache function and added __init__ file |
260,335 | 21.07.2021 21:14:40 | 25,200 | fd5dec7b685afae4ce83ffb663cb1f55ddad9070 | fix xla.lower_fun axis env issue | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/xla.py",
"new_path": "jax/interpreters/xla.py",
"diff": "@@ -1021,6 +1021,7 @@ def lower_fun(fun, multiple_results, parallel=False, with_avals=False, backend=N\nwrapped_fun = lu.wrap_init(fun, params)\nif not multiple_results:\nwrapped_fun = _... | Python | Apache License 2.0 | google/jax | fix xla.lower_fun axis env issue |
260,335 | 22.07.2021 21:09:58 | 25,200 | fa274f3acec948bd7dc8be36c626b0e804a45867 | small tweaks, type annotations | [
{
"change_type": "MODIFY",
"old_path": "docs/autodidax.ipynb",
"new_path": "docs/autodidax.ipynb",
"diff": "\" dynamic_trace = prev_dynamic_trace\\n\",\n\"\\n\",\n\"@lru_cache()\\n\",\n- \"def make_jaxpr(f, *avals_in):\\n\",\n+ \"def make_jaxpr(f: Callable, *avals_in: ShapedArray,\\n\",\n+ \" ) -> T... | Python | Apache License 2.0 | google/jax | small tweaks, type annotations |
260,411 | 23.07.2021 10:59:44 | -10,800 | 24b00e856b20cd74aae8ff499e7cd244373dbff0 | [jax2tf] Adjustments for the polymorphic_shapes specification.
Expanded the polymorphic_shapes specification to detect when
it is a string or PolyShape, and then replicate it for all
arguments.
Some improvements for the error messages for incorrect
polymorphic shapes specification. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -176,7 +176,7 @@ def inside_call_tf():\n@partial(api_util.api_hook, tag=\"jax2tf_convert\")\ndef convert(fun: Callable,\n*,\n- polymorphic_shapes: Optional[Sequenc... | Python | Apache License 2.0 | google/jax | [jax2tf] Adjustments for the polymorphic_shapes specification.
Expanded the polymorphic_shapes specification to detect when
it is a string or PolyShape, and then replicate it for all
arguments.
Some improvements for the error messages for incorrect
polymorphic shapes specification. |
260,411 | 25.07.2021 09:44:40 | -10,800 | f29f5214ab5ff0d6c3e1e34b2b87025ca9f93961 | [jax2tf] Add support for tridiagonal_solve | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -1017,7 +1017,6 @@ tf_not_yet_impl = [\n\"lu_pivots_to_permutation\",\n\"rng_bit_generator\",\n\"xla_pmap\",\n- \"tridiagonal_solve\",\n]\ntf_impl[ad_util.stop_gra... | Python | Apache License 2.0 | google/jax | [jax2tf] Add support for tridiagonal_solve |
260,411 | 25.07.2021 10:13:13 | -10,800 | 00872d58058f600bb64b3f35f6fed1f36db15682 | [jax2tf] Added test for round trip for custom gradients
Fixes: 7123
The fix is not here, it is in TF. Here we only add the test. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/tests/savedmodel_test.py",
"new_path": "jax/experimental/jax2tf/tests/savedmodel_test.py",
"diff": "# limitations under the License.\nfrom absl.testing import absltest\n+import os\nimport jax\nfrom jax import lax\n@@ -153,6 +154,61 @@ c... | Python | Apache License 2.0 | google/jax | [jax2tf] Added test for round trip for custom gradients
Fixes: 7123
The fix is not here, it is in TF. Here we only add the test. |
260,411 | 16.07.2021 08:16:09 | -10,800 | ab050059d15ffb3780e42ac1ff2b13450fe6113d | Ensure that we run the test on the proper TF device | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/call_tf.py",
"new_path": "jax/experimental/jax2tf/call_tf.py",
"diff": "@@ -280,7 +280,7 @@ def _code_generator_and_avals(\n# TODO(necula): For unoptimized HLO, does it make a difference which device we use?\ntf_device_name = \"/device:... | Python | Apache License 2.0 | google/jax | Ensure that we run the test on the proper TF device |
260,411 | 16.07.2021 13:16:21 | -10,800 | 458d66fca34d9fcb31505e1748853559d96cb4eb | Added a suggestion for safer check for captured inputs | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/call_tf.py",
"new_path": "jax/experimental/jax2tf/call_tf.py",
"diff": "@@ -295,9 +295,9 @@ def _code_generator_and_avals(\nfor inp in concrete_function_flat_tf.captured_inputs:\nif inp.dtype == tf.resource: # A variable; assume the nex... | Python | Apache License 2.0 | google/jax | Added a suggestion for safer check for captured inputs |
260,411 | 25.07.2021 18:09:08 | -10,800 | ea5959cd5ba191b21f51610b468b975737f64d47 | Force the TF compilation to use JAX's default platform | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/call_tf.py",
"new_path": "jax/experimental/jax2tf/call_tf.py",
"diff": "@@ -277,8 +277,15 @@ def _code_generator_and_avals(\nshape=a.shape,\ndtype=a.dtype) for a in args_flat_sig_tf]\n- # TODO(necula): For unoptimized HLO, does it make ... | Python | Apache License 2.0 | google/jax | Force the TF compilation to use JAX's default platform |
260,411 | 25.07.2021 14:50:54 | -10,800 | 0b697ce23229c050b6c7bfd328a372ce99468b67 | [host_callback] Fix the JVP rule for id_tap(result=...)
The previous rule was leaking ad.Zero. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/host_callback.py",
"new_path": "jax/experimental/host_callback.py",
"diff": "@@ -798,17 +798,24 @@ id_tap_dep_p.multiple_results = False\nid_tap_dep_p.def_impl(lambda r, _: r)\nxla.translations[id_tap_dep_p] = lambda comp, a_res, a_tap: a_res\... | Python | Apache License 2.0 | google/jax | [host_callback] Fix the JVP rule for id_tap(result=...)
The previous rule was leaking ad.Zero. |
260,411 | 25.07.2021 20:06:36 | -10,800 | 7638ad04e4d45cada989b97c2d90e9080add9fb2 | Fixed the safety check for call_tf
The previous check was failing on TPUs due to different layout | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/call_tf.py",
"new_path": "jax/experimental/jax2tf/call_tf.py",
"diff": "@@ -342,17 +342,21 @@ def _code_generator_and_avals(\n# Check that the function does not have compile-time constant inputs that\n# have been inlined in the compiled... | Python | Apache License 2.0 | google/jax | Fixed the safety check for call_tf
The previous check was failing on TPUs due to different layout |
260,411 | 25.06.2021 08:43:04 | -7,200 | 2749b6352420e2c063ff0900c4e0c9d11909d5da | Ensure zeros from AD are generated on device.
Fixes:
Also fixes type checking in jax2tf, because now we have to be careful
about computations that have result float0 (the broadcast_in_dim used
to compute the zeros). | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/abstract_arrays.py",
"new_path": "jax/_src/abstract_arrays.py",
"diff": "@@ -38,7 +38,8 @@ def make_shaped_array(x):\ndef zeros_like_array(x):\ndtype = dtypes.canonicalize_dtype(dtypes.result_type(x))\n- return zeros_like_shaped_array(ShapedArray(np.s... | Python | Apache License 2.0 | google/jax | Ensure zeros from AD are generated on device.
Fixes: #7093
Also fixes type checking in jax2tf, because now we have to be careful
about computations that have result float0 (the broadcast_in_dim used
to compute the zeros). |
260,411 | 25.07.2021 16:07:33 | -10,800 | 4a3e8e99e6b952ec53266ba7d18e25b6918d6c0b | Fix for numpy 1.17.5 | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/lax.py",
"new_path": "jax/_src/lax/lax.py",
"diff": "@@ -1612,8 +1612,12 @@ def _device_put_raw(x, weak_type=None):\ndef zeros_like_shaped_array(aval):\nassert isinstance(aval, ShapedArray)\n- # The .astype is useful for float0\n- return broadcast... | Python | Apache License 2.0 | google/jax | Fix for numpy 1.17.5 |
260,411 | 26.07.2021 17:48:21 | -10,800 | fd1b13cc04ba50b3b87aea79a22da85405def3e0 | [jax2tf] Improve the error message when forgetting polymorphic_shapes
Also replaced "shape variable" with "dimension variable" in error
message and documentation.
Fixes: | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -228,10 +228,10 @@ f_tf.get_concrete_function(tf.TensorSpec([None, 28, 28], tf.float32))\nThe `polymorphic_shapes` parameter, in the form of a sequence of strings ... | Python | Apache License 2.0 | google/jax | [jax2tf] Improve the error message when forgetting polymorphic_shapes
Also replaced "shape variable" with "dimension variable" in error
message and documentation.
Fixes: #7213 |
260,439 | 27.07.2021 11:12:51 | 25,200 | e7f03073dd070bb7f619b36c7bae4df7bcb67866 | ReduceScatter translation and abstract eval. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/parallel.py",
"new_path": "jax/_src/lax/parallel.py",
"diff": "@@ -1129,6 +1129,168 @@ batching.primitive_batchers[all_gather_p] = _all_gather_batcher\nbatching.collective_rules[all_gather_p] = _all_gather_batched_collective\ncore.axis_substitutio... | Python | Apache License 2.0 | google/jax | ReduceScatter translation and abstract eval.
PiperOrigin-RevId: 387152857 |
260,424 | 30.06.2021 10:46:37 | -3,600 | 19ee7b22e1d69c9a63211b1c1fc496a974257bdc | Expose UnexpectedTracerError and add docs. | [
{
"change_type": "MODIFY",
"old_path": "docs/errors.rst",
"new_path": "docs/errors.rst",
"diff": "@@ -10,3 +10,4 @@ along with representative examples of how one might fix them.\n.. autoclass:: NonConcreteBooleanIndexError\n.. autoclass:: TracerArrayConversionError\n.. autoclass:: TracerIntegerConve... | Python | Apache License 2.0 | google/jax | Expose UnexpectedTracerError and add docs. |
260,424 | 28.07.2021 13:02:06 | -3,600 | f966e7ef5ce28fa9b6ce95494d6c2f52f9c66387 | Fix some of the formatting and reword some of the sections. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/errors.py",
"new_path": "jax/_src/errors.py",
"diff": "@@ -440,18 +440,20 @@ class UnexpectedTracerError(JAXTypeError):\nWhat does it mean to leak a value? If you use a JAX transformation on a\nfunction ``f`` that stores, in some scope outside of ``f`... | Python | Apache License 2.0 | google/jax | Fix some of the formatting and reword some of the sections. |
260,510 | 07.07.2021 11:03:59 | 25,200 | 49f7ac22cce1f421e85d039a075f40b9c861ca09 | change while loop batching fixed point condition | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/control_flow.py",
"new_path": "jax/_src/lax/control_flow.py",
"diff": "@@ -373,7 +373,7 @@ def _pred_bcast_select(c, pred, x, y, x_y_aval: core.AbstractValue):\nelif x_y_aval is core.abstract_token:\nreturn xops.AfterAll(c, [x, y])\nelse:\n- asser... | Python | Apache License 2.0 | google/jax | change while loop batching fixed point condition
Co-authored-by: Sharad Vikram <sharad.vikram@gmail.com>
Co-authored-by: Adam Paszke <apaszke@google.com> |
260,578 | 28.07.2021 15:22:42 | 25,200 | bf28b884b2c54c877d6ae00f29979692e74e21da | Log a warning after 60 secs to remind the user to run code on all hosts for Cloud TPU 1VM | [
{
"change_type": "MODIFY",
"old_path": "jax/lib/xla_bridge.py",
"new_path": "jax/lib/xla_bridge.py",
"diff": "@@ -145,6 +145,22 @@ def _make_tpu_driver_client():\nreturn tpu_driver_client.TpuBackend.create(worker=FLAGS.jax_backend_target)\n+def tpu_client_timer_callback(timer: float):\n+ def _log_wa... | Python | Apache License 2.0 | google/jax | Log a warning after 60 secs to remind the user to run code on all hosts for Cloud TPU 1VM |
260,578 | 28.07.2021 18:28:37 | 25,200 | cdfc2dc8fe5a5351abff18d0d449af3f4fc5abe0 | Use warnings instead of absl.logging.warning | [
{
"change_type": "MODIFY",
"old_path": "jax/lib/xla_bridge.py",
"new_path": "jax/lib/xla_bridge.py",
"diff": "@@ -147,7 +147,7 @@ def _make_tpu_driver_client():\ndef tpu_client_timer_callback(timer: float):\ndef _log_warning():\n- logging.warning('Did you run your code on all the hosts?')\n+ warning... | Python | Apache License 2.0 | google/jax | Use warnings instead of absl.logging.warning |
260,578 | 28.07.2021 18:56:22 | 25,200 | ee7759f9939a10663c559ef0c5363f0b60d8f102 | Fix xla_bridge_test.py | [
{
"change_type": "MODIFY",
"old_path": "tests/xla_bridge_test.py",
"new_path": "tests/xla_bridge_test.py",
"diff": "# See the License for the specific language governing permissions and\n# limitations under the License.\n+import warnings\nimport time\nfrom absl.testing import absltest\n@@ -65,9 +66,... | Python | Apache License 2.0 | google/jax | Fix xla_bridge_test.py |
260,578 | 28.07.2021 19:38:58 | 25,200 | 1a994297f1aef7e0bf52d5a1dbafae08f6a44530 | Check for length of warning too | [
{
"change_type": "MODIFY",
"old_path": "tests/xla_bridge_test.py",
"new_path": "tests/xla_bridge_test.py",
"diff": "@@ -70,6 +70,7 @@ class XlaBridgeTest(absltest.TestCase):\nwarnings.simplefilter('always')\nxb.tpu_client_timer_callback(0.01)\nmsg = str(w[-1].message)\n+ self.assertLen(w, 1)\nself.a... | Python | Apache License 2.0 | google/jax | Check for length of warning too |
260,411 | 28.07.2021 19:30:44 | -10,800 | 3d85709e307fec58ec587608ff8525a9f517d63c | [jax2tf] Test code refactoring | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/call_tf.py",
"new_path": "jax/experimental/jax2tf/call_tf.py",
"diff": "@@ -100,7 +100,7 @@ def call_tf(callable_tf: Callable) -> Callable:\ndef make_tensorspec(a_jax):\na_tf_dtype = jax2tf_internal._to_tf_dtype(a_jax.dtype)\nif any(not... | Python | Apache License 2.0 | google/jax | [jax2tf] Test code refactoring |
260,411 | 29.07.2021 13:16:06 | -10,800 | 881daa5c901c1bbb9600ebd0af5ebe88ccd6e812 | [jax2tf] Handle truediv div for shape polymorphism | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/shape_poly.py",
"new_path": "jax/experimental/jax2tf/shape_poly.py",
"diff": "@@ -320,6 +320,19 @@ class _DimPolynomial(dict):\ndef __rfloordiv__(self, other):\nreturn _ensure_poly(other).__floordiv__(self)\n+ def __truediv__(self, divi... | Python | Apache License 2.0 | google/jax | [jax2tf] Handle truediv div for shape polymorphism |
260,411 | 29.07.2021 16:07:13 | -10,800 | 0b766b27a4fbb7b3f66504c975bdf0ab8f913e67 | [jax2tf] Improved testing and shape polymorphism support for lax.random. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/random.py",
"new_path": "jax/_src/random.py",
"diff": "@@ -236,7 +236,13 @@ def threefry_2x32(keypair, count):\nmsg = \"threefry_2x32 requires uint32 arguments, got {}\"\nraise TypeError(msg.format([lax.dtype(x) for x in [key1, key2, count]]))\n+ try:... | Python | Apache License 2.0 | google/jax | [jax2tf] Improved testing and shape polymorphism support for lax.random. |
260,424 | 21.07.2021 13:27:48 | -3,600 | 21907346370320209a26fbf7a4f619bb62978352 | Add tracers to LeakChecker error, and filter out false positives this way.
If we can't find any hanging tracers in the gc.get_referrers chain, is it
really a leak? Probably not! | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -18,6 +18,7 @@ from operator import attrgetter\nfrom contextlib import contextmanager\nfrom collections import namedtuple\nfrom functools import total_ordering\n+import gc\nimport itertools as it\nfrom weakr... | Python | Apache License 2.0 | google/jax | Add tracers to LeakChecker error, and filter out false positives this way.
If we can't find any hanging tracers in the gc.get_referrers chain, is it
really a leak? Probably not! |
260,335 | 29.07.2021 10:34:43 | 25,200 | c31688d2d1632f01491011d7150a527efc2007ff | fix cond-of-pmap bug | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -307,11 +307,13 @@ def extract_call_jaxpr(\nreturn (params[\"call_jaxpr\"], new_params)\n+# TODO(mattjj): replace this approach with a primitive-keyed table of rules\ndef traverse_jaxpr_params(f, params):\n\... | Python | Apache License 2.0 | google/jax | fix cond-of-pmap bug |
260,411 | 30.07.2021 10:52:34 | -10,800 | 64d2e5c107a28d1d4a81c28e9bee3e9371fad9ae | [jax2tf] More cleanup for shape polymorphism testing | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/shape_poly.py",
"new_path": "jax/experimental/jax2tf/shape_poly.py",
"diff": "@@ -632,8 +632,8 @@ def parse_spec(spec: Optional[Union[str, PolyShape]],\ndim_poly = _parse_dim(dim_spec)\nif not is_poly_dim(dim_poly):\nif dim_poly != dim_... | Python | Apache License 2.0 | google/jax | [jax2tf] More cleanup for shape polymorphism testing |
260,411 | 30.07.2021 13:35:21 | -10,800 | 6aecb3c3c6b52ee61f29e600b4ec4a570ba72555 | [jax2tf] enable multiple round-trips from JAX to TF | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/call_tf.py",
"new_path": "jax/experimental/jax2tf/call_tf.py",
"diff": "@@ -411,6 +411,7 @@ TfVal = jax2tf_internal.TfVal\ndef _jax2tf_call_tf(*args: TfVal,\ncallable_flat_tf: Callable,\n**_) -> TfVal:\n+ with jax2tf_internal.inside_cal... | Python | Apache License 2.0 | google/jax | [jax2tf] enable multiple round-trips from JAX to TF |
260,374 | 29.07.2021 17:12:15 | 0 | 5383e977d5c577771f9a1dc66aa9a663292647a2 | added debugging logs to compilation cache | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/compilation_cache/compilation_cache.py",
"new_path": "jax/experimental/compilation_cache/compilation_cache.py",
"diff": "@@ -16,6 +16,7 @@ import hashlib\nimport jax\nfrom jax.experimental.compilation_cache.file_system_cache import FileSystemC... | Python | Apache License 2.0 | google/jax | added debugging logs to compilation cache |
260,456 | 30.07.2021 19:17:21 | 0 | b44d35664c94a42f7145b73f27619e21dbc678cf | change skip_on_devices to handle device tags | [
{
"change_type": "MODIFY",
"old_path": "jax/test_util.py",
"new_path": "jax/test_util.py",
"diff": "@@ -423,16 +423,32 @@ def skip_if_unsupported_type(dtype):\nraise unittest.SkipTest(\nf\"Type {dtype.name} not supported on {device_under_test()}\")\n+def is_device_rocm():\n+ return xla_bridge.get_ba... | Python | Apache License 2.0 | google/jax | change skip_on_devices to handle device tags |
260,411 | 31.07.2021 10:24:36 | -10,800 | 022d2d6d6e592f60f8821cef47c43802c8d8cbe8 | [jax2tf] Update the limitations
To account for progress on XLA and TF | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/g3doc/jax_primitives_coverage.md",
"new_path": "jax/experimental/jax2tf/g3doc/jax_primitives_coverage.md",
"diff": "# Primitives with limited JAX support\n-*Last generated on: 2021-07-05* (YYYY-MM-DD)\n+*Last generated on: 2021-07-31* (... | Python | Apache License 2.0 | google/jax | [jax2tf] Update the limitations
To account for progress on XLA and TF |
260,578 | 01.08.2021 08:53:12 | 25,200 | 790b8e81871d589504885f9886c6020de1108d86 | Use separate variable names so that we don't get buildifier warnings | [
{
"change_type": "MODIFY",
"old_path": "WORKSPACE",
"new_path": "WORKSPACE",
"diff": "@@ -24,11 +24,14 @@ load(\"//third_party/pocketfft:workspace.bzl\", pocketfft = \"repo\")\npocketfft()\n# Initialize TensorFlow's external dependencies.\n-load(\"@org_tensorflow//tensorflow:workspace3.bzl\", \"work... | Python | Apache License 2.0 | google/jax | Use separate variable names so that we don't get buildifier warnings |
260,499 | 30.07.2021 09:33:34 | -3,600 | 69fcc0c20d9508596f506dbb4129d8ef58028e5a | Abstracts into a separate function to evaluation of a single jaxpr equation. | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -316,22 +316,8 @@ def traverse_jaxpr_params(f, params):\nif type(p) in (Jaxpr, ClosedJaxpr)}\n-def eval_jaxpr(jaxpr: Jaxpr, consts, *args):\n- def read(v):\n- if type(v) is Literal:\n- return v.val\n- else:\... | Python | Apache License 2.0 | google/jax | Abstracts into a separate function to evaluation of a single jaxpr equation. |
260,578 | 01.08.2021 20:13:02 | 25,200 | ca6465dded5e312d29bf7e7a0ffb855b4e31e2d5 | Add the changes suggested | [
{
"change_type": "MODIFY",
"old_path": "docs/async_dispatch.rst",
"new_path": "docs/async_dispatch.rst",
"diff": "@@ -10,8 +10,8 @@ program:\n>>> import jax.numpy as jnp\n>>> from jax import random\n>>> x = random.uniform(random.PRNGKey(0), (1000, 1000))\n->>> # Printing the result or repr(result) o... | Python | Apache License 2.0 | google/jax | Add the changes suggested |
260,578 | 02.08.2021 18:27:27 | 25,200 | 2f6f78891695be45476a2dd1295421083a3893c8 | Fix the timer->timer_secs typo | [
{
"change_type": "MODIFY",
"old_path": "jax/lib/xla_bridge.py",
"new_path": "jax/lib/xla_bridge.py",
"diff": "@@ -148,9 +148,10 @@ def _make_tpu_driver_client():\ndef tpu_client_timer_callback(timer_secs: float):\ndef _log_warning():\nwarnings.warn(\n- (f'TPU backend initialization is taking more th... | Python | Apache License 2.0 | google/jax | Fix the timer->timer_secs typo |
260,411 | 03.08.2021 09:12:04 | -10,800 | 0c1a37ce33514f24d79bcfdf6d8559bf38b16287 | [jax2tf] Add shape polymorphism support for jnp.eye | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/lax.py",
"new_path": "jax/_src/lax/lax.py",
"diff": "@@ -1640,16 +1640,14 @@ def broadcasted_iota(dtype: DType, shape: Shape, dimension: int) -> Array:\ndef _eye(dtype: DType, shape: Shape, offset: int) -> Array:\n\"\"\"Like numpy.eye, create a 2D... | Python | Apache License 2.0 | google/jax | [jax2tf] Add shape polymorphism support for jnp.eye |
260,631 | 03.08.2021 09:25:00 | 25,200 | 0d8ef03a939264c2fefeed6b353c979abba02b09 | Added file system cache interface | [
{
"change_type": "ADD",
"old_path": null,
"new_path": "jax/experimental/compilation_cache/cache_interface.py",
"diff": "+# Copyright 2021 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 ... | Python | Apache License 2.0 | google/jax | Added file system cache interface
PiperOrigin-RevId: 388473011 |
260,578 | 03.08.2021 16:36:27 | 25,200 | 70e9708e0abf3a156371e840637029919bf6a1e0 | Fix the tolerance so that the tests will pass on GPU | [
{
"change_type": "MODIFY",
"old_path": "tests/lax_control_flow_test.py",
"new_path": "tests/lax_control_flow_test.py",
"diff": "@@ -1440,7 +1440,7 @@ class LaxControlFlowTest(jtu.JaxTestCase):\nans = api.linearize(lambda c, as_: scan(f, c, as_), c, as_)[1](c, as_)\nexpected = api.linearize(lambda c,... | Python | Apache License 2.0 | google/jax | Fix the tolerance so that the tests will pass on GPU |
260,411 | 04.08.2021 07:30:12 | -10,800 | f126cabe77c79de79bdaca9c5d7ee39c25e71a4e | [jax2tf] Fix shape polymorphism for broadcast_in_dim_transpose | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/lax.py",
"new_path": "jax/_src/lax/lax.py",
"diff": "@@ -3626,7 +3626,7 @@ def _broadcast_in_dim_shape_rule(operand, *, shape, broadcast_dimensions):\ndef _broadcast_in_dim_transpose_rule(ct, operand, *, shape, broadcast_dimensions):\nshape_in = o... | Python | Apache License 2.0 | google/jax | [jax2tf] Fix shape polymorphism for broadcast_in_dim_transpose |
260,578 | 04.08.2021 08:56:55 | 25,200 | e54efe73edafa32335306d9ee7779b1f82b28757 | Update commit for release | [
{
"change_type": "MODIFY",
"old_path": "WORKSPACE",
"new_path": "WORKSPACE",
"diff": "@@ -7,10 +7,10 @@ load(\"@bazel_tools//tools/build_defs/repo:http.bzl\", \"http_archive\")\n# and update the sha256 with the result.\nhttp_archive(\nname = \"org_tensorflow\",\n- sha256 = \"77f1b73813553387c25e024c... | Python | Apache License 2.0 | google/jax | Update commit for release |
260,318 | 02.08.2021 19:01:13 | -10,800 | 0b1b812e3be92c1731ef84af2e5dbe8fe4ef03c4 | bugfix for sm3 optimizer
Ensure atleast 1d input. Optimizer should work now with scalar data.
fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/optimizers.py",
"new_path": "jax/experimental/optimizers.py",
"diff": "@@ -478,22 +478,24 @@ def sm3(step_size, momentum=0.9):\nreturn x[tuple(idx)]\ndef init(x0):\n+ x_shape = x0.shape\n+ x0 = jnp.atleast_1d(x0)\nvs = [jnp.zeros(sz, dtype=x0.... | Python | Apache License 2.0 | google/jax | bugfix for sm3 optimizer
Ensure atleast 1d input. Optimizer should work now with scalar data.
fixes #7421 |
260,411 | 04.08.2021 09:05:05 | -10,800 | b4e4acd88a57f038829604c9342326a46777ca20 | [jax2tf] Improve support for converting functions with kwargs
The previous conversion for kwargs did not work for AD. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -36,7 +36,6 @@ from jax._src.lax import fft as lax_fft\nfrom jax._src.lax import lax\nfrom jax._src.lax import linalg as lax_linalg\nimport jax._src.random\n-from ... | Python | Apache License 2.0 | google/jax | [jax2tf] Improve support for converting functions with kwargs
The previous conversion for kwargs did not work for AD.
Bug: #6791 |
260,578 | 05.08.2021 09:45:39 | 25,200 | 86aaf80dce5f7bec2a34d2d6a187e558ca63cbfa | Use bazel --version | [
{
"change_type": "MODIFY",
"old_path": "build/build.py",
"new_path": "build/build.py",
"diff": "@@ -209,7 +209,7 @@ def get_bazel_path(bazel_path_flag):\ndef check_bazel_version(bazel_path):\ntry:\n- version_output = shell([bazel_path, \"--bazelrc=/dev/null\", \"version\"])\n+ version_output = shell... | Python | Apache License 2.0 | google/jax | Use bazel --version |
260,688 | 06.08.2021 10:54:50 | 25,200 | d6df61c3051e4f5861b8a3d26100147ae17a1046 | Fix the move constructor for Handle | [
{
"change_type": "MODIFY",
"old_path": "jaxlib/handle_pool.h",
"new_path": "jaxlib/handle_pool.h",
"diff": "@@ -47,15 +47,19 @@ class HandlePool {\nHandle(Handle&& other) {\npool_ = other.pool_;\nhandle_ = other.handle_;\n+ stream_ = other.stream_;\nother.pool_ = nullptr;\nother.handle_ = nullptr;\n... | Python | Apache License 2.0 | google/jax | Fix the move constructor for Handle
PiperOrigin-RevId: 389212536 |
260,424 | 09.08.2021 17:23:05 | -3,600 | ddaef095bbef6571fd4f790e046db2b08bcfb964 | Reword UnexpectedTracerError and add visual dividers. | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -440,31 +440,32 @@ class Trace:\ndef escaped_tracer_error(tracer, detail=None):\nnum_frames = FLAGS.jax_tracer_error_num_traceback_frames\n- msg = (\"Encountered an unexpected tracer. Perhaps this tracer esc... | Python | Apache License 2.0 | google/jax | Reword UnexpectedTracerError and add visual dividers. |
260,411 | 10.08.2021 11:50:16 | -10,800 | d597a45499ac5d5694d42c6b1decb5f77c21e182 | [jax2tf] Small fixes for shape polymorphism.
* Fixed a few places where indexing required indices to be integer
constants. Use core._canonicalize_dimension in lieu of casting to integer.
Note that this may break code that passes floating point as indices.
* Relaxed einsum to work for multiple contractions.
* Added tests. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/lax.py",
"new_path": "jax/_src/lax/lax.py",
"diff": "@@ -1947,7 +1947,7 @@ def slice_in_dim(operand: Array, start_index: Optional[int],\ndef index_in_dim(operand: Array, index: int, axis: int = 0,\nkeepdims: bool = True) -> Array:\n\"\"\"Convenien... | Python | Apache License 2.0 | google/jax | [jax2tf] Small fixes for shape polymorphism.
* Fixed a few places where indexing required indices to be integer
constants. Use core._canonicalize_dimension in lieu of casting to integer.
Note that this may break code that passes floating point as indices.
* Relaxed einsum to work for multiple contractions.
* Added tests. |
260,424 | 10.08.2021 18:46:03 | -3,600 | 2f9caf3d642aa584571105a7516b47d9bf8d1789 | Ensure reduce_axes is a tuple. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/api.py",
"new_path": "jax/_src/api.py",
"diff": "@@ -852,6 +852,7 @@ def value_and_grad(fun: Callable, argnums: Union[int, Sequence[int]] = 0,\n_check_callable(fun)\nargnums = core.concrete_or_error(_ensure_index, argnums)\n+ reduce_axes = _ensure_str... | Python | Apache License 2.0 | google/jax | Ensure reduce_axes is a tuple. |
260,578 | 12.08.2021 11:12:35 | 25,200 | 1f13565b70cff1cf2c6eb81f5fcd2ee6d41286af | Remove opensource CS link since JAX is not indexed | [
{
"change_type": "MODIFY",
"old_path": "README.md",
"new_path": "README.md",
"diff": "| [**Neural net libraries**](#neural-network-libraries)\n| [**Change logs**](https://jax.readthedocs.io/en/latest/changelog.html)\n| [**Reference docs**](https://jax.readthedocs.io/en/latest/)\n-| [**Code search**]... | Python | Apache License 2.0 | google/jax | Remove opensource CS link since JAX is not indexed |
260,347 | 13.08.2021 14:52:03 | 0 | 6708cd31584f6bd14431ae1020df3de7a1635bb5 | Add dtype to string representation of ConcreteArray.
The string representation of ConcreteArray did not include the data type of the
wrapped value. This makes it harder to spot the reason for errors arising from
inconsistent values (issue This commit adds the data type to the string
representation of ConcreteArray. | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -1190,7 +1190,7 @@ class ConcreteArray(ShapedArray):\nraise TypeError(self, other)\ndef str_short(self) -> str:\n- return str(self.val)\n+ return f'{self.val}, dtype={self.dtype.name}'\n_bool = _nonzero = pa... | Python | Apache License 2.0 | google/jax | Add dtype to string representation of ConcreteArray.
The string representation of ConcreteArray did not include the data type of the
wrapped value. This makes it harder to spot the reason for errors arising from
inconsistent values (issue #5364). This commit adds the data type to the string
representation of ConcreteArray. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.