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 |
|---|---|---|---|---|---|---|---|---|---|
260,447 | 03.11.2022 21:39:10 | 25,200 | 46368e4e734938bffe7978cb4077923343095abc | [sparse] Update the guard of cusparse SpMM and SpMv algorithms to cusparse version 11.7.1 onwards. | [
{
"change_type": "MODIFY",
"old_path": "jaxlib/gpu/vendor.h",
"new_path": "jaxlib/gpu/vendor.h",
"diff": "@@ -224,9 +224,9 @@ typedef cusparseDnVecDescr_t gpusparseDnVecDescr_t;\n// Use CUSPARSE_SPMV_COO_ALG2 and CUSPARSE_SPMV_CSR_ALG2 for SPMV and\n// use CUSPARSE_SPMM_COO_ALG2 and CUSPARSE_SPMM_CS... | Python | Apache License 2.0 | google/jax | [sparse] Update the guard of cusparse SpMM and SpMv algorithms to cusparse version 11.7.1 onwards.
PiperOrigin-RevId: 486051658 |
260,720 | 27.10.2022 15:33:03 | 25,200 | e3cd9671b93b0235b81ace9d84c3665480107169 | Add x86_64 dependency note to pip installation
Currently non x86_64 linux architectures are not supported, see for request to change this. This can lead to installation confusion, as jax will install, but jaxlib will not. For example see This adds a note to the install sections for the relevant pip wheels. | [
{
"change_type": "MODIFY",
"old_path": "README.md",
"new_path": "README.md",
"diff": "@@ -413,7 +413,7 @@ pip install --upgrade \"jax[cpu]\"\n```\nOn Linux, it is often necessary to first update `pip` to a version that supports\n-`manylinux2014` wheels.\n+`manylinux2014` wheels. Also note that for L... | Python | Apache License 2.0 | google/jax | Add x86_64 dependency note to pip installation
Currently non x86_64 linux architectures are not supported, see #7097 for request to change this. This can lead to installation confusion, as jax will install, but jaxlib will not. For example see #12307. This adds a note to the install sections for the relevant pip wheels. |
260,283 | 04.11.2022 20:53:52 | 0 | e6c88f2c58c4eda7192db60646d917a85ac66273 | update pytest.ini to print warning message for compilation_cache_test | [
{
"change_type": "MODIFY",
"old_path": "pytest.ini",
"new_path": "pytest.ini",
"diff": "@@ -19,5 +19,8 @@ filterwarnings =\n# numpy uses distutils which is deprecated\nignore:The distutils.* is deprecated.*:DeprecationWarning\nignore:`sharded_jit` is deprecated. Please use `pjit` instead.*:Deprecati... | Python | Apache License 2.0 | google/jax | update pytest.ini to print warning message for compilation_cache_test |
260,335 | 04.11.2022 19:49:07 | 25,200 | 190204ff7d186efd9e4f78d4d49bb3387e3759ab | fix jax.random.logits shape argument
fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/random.py",
"new_path": "jax/_src/random.py",
"diff": "@@ -1327,10 +1327,12 @@ def categorical(key: KeyArray,\nshape = tuple(shape)\n_check_shape(\"categorical\", shape, batch_shape)\n- sample_shape = shape[:len(shape)-len(batch_shape)]\n+ shape_prefi... | Python | Apache License 2.0 | google/jax | fix jax.random.logits shape argument
fixes #13124 |
260,335 | 17.10.2022 11:15:14 | 25,200 | f2f2faa4fa166f40a4a93bc966379cf1ebb720d1 | add a basic prototype of piles, behind jax_dynamic_shapes | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/lax.py",
"new_path": "jax/_src/lax/lax.py",
"diff": "@@ -45,6 +45,7 @@ from jax.interpreters import xla\nfrom jax.interpreters import pxla\nfrom jax.interpreters import ad\nfrom jax.interpreters import batching\n+from jax.interpreters.batching imp... | Python | Apache License 2.0 | google/jax | add a basic prototype of piles, behind jax_dynamic_shapes
Co-authored-by: Adam Paszke <apaszke@google.com>
Co-authored-by: Dougal Maclaurin <dougalm@google.com> |
260,411 | 07.11.2022 11:18:32 | -7,200 | d9b1dc336d4759c3ddc08f8c1837b5dd386857bc | [jax2tf] Fixed jax2tf Limitations
Improved the documentation, and fixed a dot_general limitations for preferred_element_type on GPU | [
{
"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: 2022-10-26* (YYYY-MM-DD)\n+*Last generated on: 2022-11-07* (... | Python | Apache License 2.0 | google/jax | [jax2tf] Fixed jax2tf Limitations
Improved the documentation, and fixed a dot_general limitations for preferred_element_type on GPU |
260,681 | 07.11.2022 11:54:30 | 28,800 | 218305964fcd28690274525f73a7ebdcaf9b695f | Fix GDA error message formatting | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/global_device_array.py",
"new_path": "jax/experimental/global_device_array.py",
"diff": "@@ -288,7 +288,7 @@ class GlobalDeviceArray:\nss = get_shard_shape(self._global_shape, self._global_mesh, self.mesh_axes)\nassert all(db.shape == ss for d... | Python | Apache License 2.0 | google/jax | Fix GDA error message formatting
PiperOrigin-RevId: 486724647 |
260,424 | 07.11.2022 15:46:53 | 0 | e80c34d6246e6628544fe22746541e87b765a757 | Don't donate arguments in jit/pmap/pjit when debug_nans=True. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/api.py",
"new_path": "jax/_src/api.py",
"diff": "@@ -413,7 +413,9 @@ def _prepare_jit(fun, static_argnums, static_argnames, donate_argnums,\nf, args = argnums_partial_except(f, static_argnums, args, allow_invalid=True)\nf, kwargs = argnames_partial_ex... | Python | Apache License 2.0 | google/jax | Don't donate arguments in jit/pmap/pjit when debug_nans=True. |
260,335 | 06.11.2022 22:56:51 | 28,800 | 0b463efb70f6f87feecb38465fb1c977fed2806a | tighten up vmap w/ piles: require pile_axis in_axes/out_axes | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/batching.py",
"new_path": "jax/interpreters/batching.py",
"diff": "@@ -70,6 +70,15 @@ class Pile:\naval: PileTy\ndata: Array\n+# To vmap over a pile, one must specify the axis as PileAxis.\n+class PileAxis: pass\n+pile_axis = PileAxis()\n+\n+#... | Python | Apache License 2.0 | google/jax | tighten up vmap w/ piles: require pile_axis in_axes/out_axes |
260,455 | 08.11.2022 10:49:22 | 28,800 | 96f6c1c9d414e6ebc54ff7f08115a9a9a6d6a8f8 | Let is_user_frame ignore frames from stdlib.
When using decorators, we found contextlib.py from stdlib sometimes become the most recent non-jax frame. But it's not a user frame. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/source_info_util.py",
"new_path": "jax/_src/source_info_util.py",
"diff": "@@ -17,6 +17,7 @@ import dataclasses\nimport functools\nimport itertools\nimport os.path\n+import sysconfig\nimport threading\nimport types\nfrom typing import Optional, Iterat... | Python | Apache License 2.0 | google/jax | Let is_user_frame ignore frames from stdlib.
When using decorators, we found contextlib.py from stdlib sometimes become the most recent non-jax frame. But it's not a user frame.
PiperOrigin-RevId: 486993924 |
260,447 | 08.11.2022 23:02:39 | 28,800 | 3b1ddf2881cff41ad06276b41e62644bddf9d0a6 | [linalg] Add jax.scipy.special.bessel_jn (Bessel function of the first kind). | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/scipy/special.py",
"new_path": "jax/_src/scipy/special.py",
"diff": "# limitations under the License.\nfrom functools import partial\n+import operator\nfrom typing import cast, Any, List, Optional, Tuple\nimport numpy as np\nimport scipy.special as os... | Python | Apache License 2.0 | google/jax | [linalg] Add jax.scipy.special.bessel_jn (Bessel function of the first kind).
PiperOrigin-RevId: 487146250 |
260,424 | 09.11.2022 12:08:57 | 0 | 053b8b5bcd912ad85e23cb9e052f1d903b506ae5 | Checkify: fix nan_checks+PRNGKeys - a PRNGKey is never NaN!
Add a guard to the nan_error_rule to not call jnp.isnan on keys. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/checkify.py",
"new_path": "jax/_src/checkify.py",
"diff": "@@ -36,6 +36,7 @@ from jax.tree_util import tree_flatten, tree_unflatten, register_pytree_node\nfrom jax._src import source_info_util, traceback_util\nfrom jax._src.lax import control_flow as ... | Python | Apache License 2.0 | google/jax | Checkify: fix nan_checks+PRNGKeys - a PRNGKey is never NaN!
Add a guard to the nan_error_rule to not call jnp.isnan on keys. |
260,368 | 09.11.2022 11:32:12 | 28,800 | 8ac7422e265d2d80491d26611dd6eb6d463eab32 | [JAX] Disables large k test cases in ann_test.
Will investigate probability properties for the corner cases in the future. | [
{
"change_type": "MODIFY",
"old_path": "tests/ann_test.py",
"new_path": "tests/ann_test.py",
"diff": "@@ -60,12 +60,14 @@ def compute_recall(result_neighbors, ground_truth_neighbors) -> float:\nclass AnnTest(jtu.JaxTestCase):\n+ # TODO(b/258315194) Investigate probability property when input is arou... | Python | Apache License 2.0 | google/jax | [JAX] Disables large k test cases in ann_test.
Will investigate probability properties for the corner cases in the future.
PiperOrigin-RevId: 487302143 |
260,356 | 08.11.2022 14:45:58 | 28,800 | cb3762e5dd18a87317eacb833dc2171853a8b947 | Consolidate links in JAX documentation
Move notes down | [
{
"change_type": "MODIFY",
"old_path": "docs/index.rst",
"new_path": "docs/index.rst",
"diff": "@@ -15,7 +15,7 @@ parallelize, Just-In-Time compile to GPU/TPU, and more.\nnotebooks/Common_Gotchas_in_JAX\n.. toctree::\n- :maxdepth: 2\n+ :maxdepth: 1\njax-101/index\n@@ -54,20 +54,9 @@ parallelize, Jus... | Python | Apache License 2.0 | google/jax | Consolidate links in JAX documentation
Move notes down |
260,510 | 09.11.2022 17:18:19 | 28,800 | 3731e446c0fde13b1d1863073ba39db00d92b813 | Set default layout for Python callback | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/mlir.py",
"new_path": "jax/interpreters/mlir.py",
"diff": "@@ -1544,12 +1544,25 @@ def _emit_tpu_python_callback(\nctx.module_context.add_host_callback(opaque)\nreturn outputs, token, opaque\n+def _layout_to_mlir_layout(minor_to_major: Optiona... | Python | Apache License 2.0 | google/jax | Set default layout for Python callback
PiperOrigin-RevId: 487388682 |
260,510 | 10.11.2022 11:59:16 | 28,800 | 74b136e62c7ee8f6d29773e470094be129e1559c | Delete `jax_experimental_name_stack` flag | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/api.py",
"new_path": "jax/_src/api.py",
"diff": "@@ -3290,20 +3290,8 @@ def named_call(\nif name is None:\nname = fun.__name__\n- _, in_tree = tree_flatten(())\n-\n- if config.jax_experimental_name_stack:\nreturn source_info_util.extend_name_stack(nam... | Python | Apache License 2.0 | google/jax | Delete `jax_experimental_name_stack` flag
PiperOrigin-RevId: 487601864 |
260,447 | 09.11.2022 23:01:24 | 28,800 | 311fb24ff96c6be56e3199929be7043cfeb2673d | [sparse] Add BCSR from_scipy_sparse. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/sparse/bcsr.py",
"new_path": "jax/experimental/sparse/bcsr.py",
"diff": "@@ -348,6 +348,20 @@ class BCSR(JAXSparse):\n\"\"\"Create a dense version of the array.\"\"\"\nreturn bcsr_todense(self)\n+ @classmethod\n+ def from_scipy_sparse(cls, mat... | Python | Apache License 2.0 | google/jax | [sparse] Add BCSR from_scipy_sparse.
Co-authored-by: Jake Vanderplas <vanderplas@google.com> |
260,447 | 10.11.2022 13:55:35 | 28,800 | 332fced0ccc76d4a3d74abf3e6ec833e099ece51 | sparse] BCSR batching rule.
[Co-authored-by: Jake Vanderplas: | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/sparse/bcsr.py",
"new_path": "jax/experimental/sparse/bcsr.py",
"diff": "@@ -179,6 +179,21 @@ def _bcoo_fromdense_abstract_eval(mat, *, nse, n_batch, n_dense, index_dtype):\ncore.ShapedArray(indptr_shape, index_dtype))\n+def _bcsr_fromdense_ba... | Python | Apache License 2.0 | google/jax | sparse] BCSR batching rule.
[Co-authored-by: Jake Vanderplas: <vanderplas@google.com> |
260,368 | 11.11.2022 07:08:27 | 28,800 | 10e6fe8cde3a26a9c37793ecb4a26f577ff0c0b5 | Correct norm in ann.py doc. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/ann.py",
"new_path": "jax/_src/lax/ann.py",
"diff": "@@ -188,10 +188,10 @@ def approx_min_k(operand: Array,\n>>>\n>>> qy = jax.numpy.array(np.random.rand(50, 64))\n>>> db = jax.numpy.array(np.random.rand(1024, 64))\n- >>> half_db_norms = jax.numpy... | Python | Apache License 2.0 | google/jax | Correct norm in ann.py doc.
PiperOrigin-RevId: 487814084 |
260,516 | 11.11.2022 08:04:38 | 28,800 | d2afa84a6ef14dc6ffba31bba2e658b808acc6b2 | PRNGKeyArray is now a virtual subclass of ndarray | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/prng.py",
"new_path": "jax/_src/prng.py",
"diff": "@@ -31,6 +31,7 @@ from jax.interpreters import batching\nfrom jax.interpreters import mlir\nfrom jax.interpreters import pxla\nfrom jax.interpreters import xla\n+from jax._src import basearray\nfrom j... | Python | Apache License 2.0 | google/jax | PRNGKeyArray is now a virtual subclass of ndarray |
260,510 | 11.11.2022 14:00:33 | 28,800 | 4bdfdd736343f17a774188ec390a3675e6a66cd0 | Update changelog w/ info about deleting `jax_experimental_name_stack` | [
{
"change_type": "MODIFY",
"old_path": "CHANGELOG.md",
"new_path": "CHANGELOG.md",
"diff": "@@ -15,6 +15,8 @@ Remember to align the itemized text with the first line of an item within a list\n{func}`jax.lax.linalg.tridiagonal`, and\n{func}`jax.lax.linalg.householder_product` were added. Householder ... | Python | Apache License 2.0 | google/jax | Update changelog w/ info about deleting `jax_experimental_name_stack` |
260,278 | 10.11.2022 19:03:04 | 0 | 09f62dec3cacf8cb21fb87196a8c62a71486fcbe | Moved abs to inputs of lcm and added specific test | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/numpy/lax_numpy.py",
"new_path": "jax/_src/numpy/lax_numpy.py",
"diff": "@@ -4367,11 +4367,12 @@ def gcd(x1: ArrayLike, x2: ArrayLike) -> Array:\ndef lcm(x1: ArrayLike, x2: ArrayLike) -> Array:\n_check_arraylike(\"lcm\", x1, x2)\nx1, x2 = _promote_dty... | Python | Apache License 2.0 | google/jax | Moved abs to inputs of lcm and added specific test |
260,294 | 11.11.2022 22:41:51 | 0 | 31d8f62826d019b41f182b56d6d9213da420a93f | Sytrd solver and SytrdDescriptor should NOT be CUDA only | [
{
"change_type": "MODIFY",
"old_path": "jaxlib/gpu/solver_kernels.h",
"new_path": "jaxlib/gpu/solver_kernels.h",
"diff": "@@ -160,6 +160,7 @@ struct GesvdjDescriptor {\nvoid Gesvdj(gpuStream_t stream, void** buffers, const char* opaque,\nsize_t opaque_len, XlaCustomCallStatus* status);\n+#endif // J... | Python | Apache License 2.0 | google/jax | Sytrd solver and SytrdDescriptor should NOT be CUDA only |
260,510 | 11.11.2022 15:23:44 | 28,800 | e15619ceabfe04a3b9e0d315011240bb3facaf55 | Convert string axis name into tuple of strings in Mesh constructor | [
{
"change_type": "MODIFY",
"old_path": "CHANGELOG.md",
"new_path": "CHANGELOG.md",
"diff": "@@ -17,6 +17,9 @@ Remember to align the itemized text with the first line of an item within a list\ntridiagonal reductions are supported on CPU only.\n* Breaking Changes\n* Deleted the `jax_experimental_name_... | Python | Apache License 2.0 | google/jax | Convert string axis name into tuple of strings in Mesh constructor
PiperOrigin-RevId: 487930412 |
260,496 | 14.11.2022 14:28:08 | -3,600 | 9fd99edca060ed0aa101e817a1f05246c473403d | [jax2tf] Update link in README.md | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -963,8 +963,8 @@ of shape polymorphism.\n### TensorFlow XLA ops\nFor most JAX primitives there is a natural TensorFlow op that fits the needed semantics.\n-There a... | Python | Apache License 2.0 | google/jax | [jax2tf] Update link in README.md |
260,310 | 14.11.2022 10:11:29 | 28,800 | 7c6a65bee2e7eb41cc078d6814a75f3e37f8e9cd | Add tf.get_current_name_scope() as prefix of name_stack during tf execution. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -1039,9 +1039,14 @@ class TensorFlowTrace(core.Trace):\n# We don't use `str(name_stack)` because it uses parentheses for\n# transformations, which aren't allowed i... | Python | Apache License 2.0 | google/jax | Add tf.get_current_name_scope() as prefix of name_stack during tf execution.
PiperOrigin-RevId: 488399560 |
260,424 | 15.11.2022 07:35:15 | 28,800 | 3116ed52a94eb996db63a67d16109c03b5d94845 | Checkify: fix nan check when primitive has multiple results. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/checkify.py",
"new_path": "jax/_src/checkify.py",
"diff": "@@ -568,10 +568,10 @@ def nan_error_check(prim, error, enabled_errors, *in_vals, **params):\ndef isnan(x):\nif isinstance(x, prng.PRNGKeyArray):\nreturn False\n- return jnp.isnan(x)\n+ return ... | Python | Apache License 2.0 | google/jax | Checkify: fix nan check when primitive has multiple results.
PiperOrigin-RevId: 488653856 |
260,631 | 15.11.2022 07:36:21 | 28,800 | 20f092c916050baab72d7d1da9758aa3c9ea9628 | As explained in the JAX FAQ, jax.numpy.where has suprising behavior when
its gradient is taken and one of the inputs is NaN. This CL adds
a short description of the behavior to the jax.numpy.where docs,
which is the logical place that users would look for it. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/numpy/lax_numpy.py",
"new_path": "jax/_src/numpy/lax_numpy.py",
"diff": "@@ -1059,7 +1059,16 @@ def where(condition: ArrayLike, x: Optional[ArrayLike] = None,\nof :py:func:`jax.numpy.where` because its output shape is data-dependent. The\nthree-argume... | Python | Apache License 2.0 | google/jax | As explained in the JAX FAQ, jax.numpy.where has suprising behavior when
its gradient is taken and one of the inputs is NaN. This CL adds
a short description of the behavior to the jax.numpy.where docs,
which is the logical place that users would look for it.
PiperOrigin-RevId: 488654036 |
260,287 | 15.11.2022 11:02:58 | 28,800 | d742e6a41083a98d13edd3c1fa695ccc609181c9 | Transpose all_gather to reduce_scatter
Also, add support for AD and batching of reduce_scatter (with its transpose being all_gather again). | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/parallel.py",
"new_path": "jax/_src/lax/parallel.py",
"diff": "@@ -1192,16 +1192,11 @@ def _all_gather_abstract_eval(x, *, all_gather_dimension, axis_name, axis_index_\nreturn x_aval.update(shape=new_shape, named_shape=new_named_shape)\ndef _all_g... | Python | Apache License 2.0 | google/jax | Transpose all_gather to reduce_scatter
Also, add support for AD and batching of reduce_scatter (with its transpose being all_gather again).
PiperOrigin-RevId: 488706478 |
260,631 | 15.11.2022 12:41:08 | 28,800 | 726b2bc2ee1bbc5690ad2b6ab0e2b77aa88eb868 | Add JAX monitoring library that instruments code via events. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/dispatch.py",
"new_path": "jax/_src/dispatch.py",
"diff": "@@ -36,6 +36,7 @@ import jax\nfrom jax import core\nfrom jax import linear_util as lu\nfrom jax.errors import UnexpectedTracerError\n+from jax.monitoring import record_event_duration_secs\nimp... | Python | Apache License 2.0 | google/jax | Add JAX monitoring library that instruments code via events.
PiperOrigin-RevId: 488731805 |
260,411 | 16.11.2022 11:53:43 | -3,600 | c7031ff369a634ccb5f73cb9cf02f394d4c290ea | [jax2tf] Disable experimental_native_lowering polymorphism tests when not jax_dynamic_shapes | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/examples/saved_model_main_test.py",
"new_path": "jax/experimental/jax2tf/examples/saved_model_main_test.py",
"diff": "@@ -47,6 +47,10 @@ class SavedModelMainTest(tf_test_util.JaxToTfTestCase):\ndef test_train_and_save_full(self,\nmodel=... | Python | Apache License 2.0 | google/jax | [jax2tf] Disable experimental_native_lowering polymorphism tests when not jax_dynamic_shapes |
260,631 | 16.11.2022 08:42:36 | 28,800 | f2bd1afb7e7e9a4bbf733f63077282d926f299d9 | Change repr on NamedSharding to match variable names. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/sharding.py",
"new_path": "jax/_src/sharding.py",
"diff": "@@ -265,7 +265,7 @@ class NamedSharding(XLACompatibleSharding):\n_check_mesh_resource_axis(self.mesh, self._parsed_pspec)\ndef __repr__(self):\n- return f'NamedSharding(mesh={dict(self.mesh.sh... | Python | Apache License 2.0 | google/jax | Change repr on NamedSharding to match variable names.
PiperOrigin-RevId: 488950019 |
260,503 | 16.11.2022 09:34:33 | 28,800 | ac2af539d310cc02d4e4e266810719b4ec5cd2e9 | Expand XlaExecutable.cost_analysis to call Executable.cost_analysis
in case the backend does not implement the "client" attribute on xla_executable. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/stages.py",
"new_path": "jax/_src/stages.py",
"diff": "@@ -208,17 +208,30 @@ class XlaExecutable(Executable):\nxla_ext_exe = self.xla_extension_executable()\nerr_msg = (\"cost analysis unsupported on current XLA backend: \"\nf\"{type(xla_ext_exe)}\")\... | Python | Apache License 2.0 | google/jax | Expand XlaExecutable.cost_analysis to call Executable.cost_analysis
in case the backend does not implement the "client" attribute on xla_executable.
PiperOrigin-RevId: 488962373 |
260,447 | 17.11.2022 17:52:54 | 28,800 | bf21480534296d1fbf5d7a1a4143609d7e8a4add | [sparse] Disable gpu bcoo_matmul test when type promotion is required. | [
{
"change_type": "MODIFY",
"old_path": "tests/sparse_test.py",
"new_path": "tests/sparse_test.py",
"diff": "@@ -2077,6 +2077,14 @@ class BCOOTest(sptu.SparseTestCase):\n)\n@jax.default_matmul_precision(\"float32\")\ndef test_bcoo_matmul(self, lhs_shape, lhs_dtype, rhs_shape, rhs_dtype):\n+ # TODO(b/... | Python | Apache License 2.0 | google/jax | [sparse] Disable gpu bcoo_matmul test when type promotion is required.
PiperOrigin-RevId: 489351869 |
260,631 | 18.11.2022 04:18:11 | 28,800 | 7a3dbcf94e7f7f14532dc8f6779302827e92b3e1 | Change params to _params to avoid clashes with downstream users. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/stages.py",
"new_path": "jax/_src/stages.py",
"diff": "@@ -458,22 +458,22 @@ class Compiled(Stage):\nreturn tree_util.tree_unflatten(self.out_tree, shardings_flat) # pytype: disable=attribute-error\n@staticmethod\n- def call(params, *args, **kwargs):\... | Python | Apache License 2.0 | google/jax | Change params to _params to avoid clashes with downstream users.
PiperOrigin-RevId: 489441483 |
260,631 | 21.11.2022 03:21:33 | 28,800 | 7890ec8164289df547a8ceeea4621beb139949e3 | Generalize TPU mesh computations. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/mesh_utils.py",
"new_path": "jax/experimental/mesh_utils.py",
"diff": "@@ -186,14 +186,14 @@ def _get_physical_tpu_mesh(jax_devices: Sequence[Any]) -> np.ndarray:\ndef sort_key(device):\nx, y, z = device.coords\ncore = device.core_on_chip\n- i... | Python | Apache License 2.0 | google/jax | Generalize TPU mesh computations.
PiperOrigin-RevId: 489936718 |
260,631 | 22.11.2022 08:30:38 | 28,800 | 518fe6656ca2aab66dcfc8cd7866c10f476a17b1 | Pickling of Sharding classes: use module level functions when deserializing.
This avoids having to pickle the sharding class (which references the module and the Python source file) in the serialized bytes, which happens when deserializing using `classmethod`s. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/sharding.py",
"new_path": "jax/_src/sharding.py",
"diff": "@@ -242,12 +242,8 @@ class NamedSharding(XLACompatibleSharding):\nself._parsed_pspec = _parsed_pspec\nself._preprocess()\n- @classmethod\n- def unpickle(cls, mesh, spec):\n- return cls(mesh, s... | Python | Apache License 2.0 | google/jax | Pickling of Sharding classes: use module level functions when deserializing.
This avoids having to pickle the sharding class (which references the module and the Python source file) in the serialized bytes, which happens when deserializing using `classmethod`s.
PiperOrigin-RevId: 490249959 |
260,287 | 23.11.2022 09:14:55 | 28,800 | 7dd958947daa1aaaf16f48e85e7d2b5ff8b64e36 | Make testManyArgs actually test pmap with many args
For some reason the test has always been passing a single array since it was added,
which seems contradictory with its purpose. | [
{
"change_type": "MODIFY",
"old_path": "tests/pmap_test.py",
"new_path": "tests/pmap_test.py",
"diff": "@@ -1690,7 +1690,7 @@ class PythonPmapTest(jtu.JaxTestCase):\nvals = list(range(500))\nndevices = jax.device_count()\n- self.assertAllClose(f(jnp.array([vals] * ndevices)),\n+ self.assertAllClose(... | Python | Apache License 2.0 | google/jax | Make testManyArgs actually test pmap with many args
For some reason the test has always been passing a single array since it was added,
which seems contradictory with its purpose.
PiperOrigin-RevId: 490519068 |
260,403 | 23.11.2022 12:02:34 | 28,800 | 074e4ec813dcaa0a49162117741dd7d3fab3c8d3 | Enable faster test-runners for PR/push CI runs. | [
{
"change_type": "MODIFY",
"old_path": ".github/workflows/ci-build.yaml",
"new_path": ".github/workflows/ci-build.yaml",
"diff": "@@ -22,7 +22,7 @@ permissions:\njobs:\nlint_and_typecheck:\n- runs-on: ubuntu-latest\n+ runs-on: ubuntu-20.04-16core\ntimeout-minutes: 5\nsteps:\n- name: Cancel previous\... | Python | Apache License 2.0 | google/jax | Enable faster test-runners for PR/push CI runs. |
260,287 | 24.11.2022 09:13:11 | 28,800 | a71116656956827807baed602cf85c4eef98517a | Make eager pmap take advantage of pmap cache
The current strategy of creating a `partial(primitive.bind, **params)` has the downside
of completely confusing the pmap cache and resulting in a new compilation for every single
primitive. Replacing it with a `HashableFunction` should fix it.
Also, pmap_test is now 2x faster! | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/pxla.py",
"new_path": "jax/interpreters/pxla.py",
"diff": "@@ -32,7 +32,7 @@ from __future__ import annotations\nimport enum\nfrom contextlib import contextmanager, ContextDecorator\n-from collections import defaultdict, OrderedDict\n+from col... | Python | Apache License 2.0 | google/jax | Make eager pmap take advantage of pmap cache
The current strategy of creating a `partial(primitive.bind, **params)` has the downside
of completely confusing the pmap cache and resulting in a new compilation for every single
primitive. Replacing it with a `HashableFunction` should fix it.
Also, pmap_test is now 2x faster!
PiperOrigin-RevId: 490749153 |
260,589 | 24.11.2022 09:56:27 | 28,800 | 575c2f37837d6e10991544389522cdc1db1b1741 | Skip unsupported tests on XLA:CPU MLIR. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/test_util.py",
"new_path": "jax/_src/test_util.py",
"diff": "@@ -303,6 +303,20 @@ def set_host_platform_device_count(nr_devices: int):\nxla_bridge.get_backend.cache_clear()\nreturn undo\n+\n+def skip_on_xla_cpu_mlir(test_method):\n+ \"\"\"A decorator ... | Python | Apache License 2.0 | google/jax | Skip unsupported tests on XLA:CPU MLIR.
PiperOrigin-RevId: 490754048 |
260,589 | 25.11.2022 06:25:11 | 28,800 | cc1d2aaaeda6c0e10ffc66b9fc7f6f446e1bc6db | Disable more {cost,memory}_analysis tests when MLIR lowering is enabled. | [
{
"change_type": "MODIFY",
"old_path": "tests/pjit_test.py",
"new_path": "tests/pjit_test.py",
"diff": "@@ -998,6 +998,7 @@ class PJitTest(jtu.BufferDonationTestCase):\nself.assertIsInstance(f.as_text(), (str, type(None)))\n@jtu.with_mesh([('x', 2), ('y', 2)])\n+ @jtu.skip_on_xla_cpu_mlir\ndef testL... | Python | Apache License 2.0 | google/jax | Disable more {cost,memory}_analysis tests when MLIR lowering is enabled.
PiperOrigin-RevId: 490898616 |
260,383 | 27.11.2022 20:05:38 | 28,800 | 4011d1796582cf0fc999e72b2df9ebe036640f66 | Change documentation to state the correct usage of XLA_PYTHON_CLIENT_MEM_FRACTION | [
{
"change_type": "MODIFY",
"old_path": "docs/gpu_memory_allocation.rst",
"new_path": "docs/gpu_memory_allocation.rst",
"diff": "@@ -16,7 +16,7 @@ override the default behavior:\n``XLA_PYTHON_CLIENT_MEM_FRACTION=.XX``\nIf preallocation is enabled, this makes JAX preallocate XX% of\n- currently-availa... | Python | Apache License 2.0 | google/jax | Change documentation to state the correct usage of XLA_PYTHON_CLIENT_MEM_FRACTION |
260,383 | 28.11.2022 09:30:25 | 28,800 | 7456b66d354160b1cdb07a0c1e7c19bb4dc78a0b | changed both "currently available" to "total" for mem allocation doc | [
{
"change_type": "MODIFY",
"old_path": "docs/gpu_memory_allocation.rst",
"new_path": "docs/gpu_memory_allocation.rst",
"diff": "GPU memory allocation\n=====================\n-**JAX will preallocate 90% of currently-available GPU memory when the first JAX\n+**JAX will preallocate 90% of the total GPU... | Python | Apache License 2.0 | google/jax | changed both "currently available" to "total" for mem allocation doc |
260,383 | 28.11.2022 10:37:42 | 28,800 | 5fb0215d4dd2a7fcf031d2f6251634ea46223292 | updated jaxlib CHANGELOG | [
{
"change_type": "MODIFY",
"old_path": "CHANGELOG.md",
"new_path": "CHANGELOG.md",
"diff": "@@ -21,6 +21,13 @@ Remember to align the itemized text with the first line of an item within a list\n## jaxlib 0.4.0\n+* Changes\n+ * The behavior of `XLA_PYTHON_CLIENT_MEM_FRACTION=.XX` has been changed to a... | Python | Apache License 2.0 | google/jax | updated jaxlib CHANGELOG |
260,545 | 16.11.2022 20:26:20 | 28,800 | a35fe206a1d0fbc01eb7d0adb1be7868d371843c | Added more accurate version of the betaln function. | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.scipy.rst",
"new_path": "docs/jax.scipy.rst",
"diff": "@@ -107,6 +107,7 @@ jax.scipy.special\n:toctree: _autosummary\nbetainc\n+ betaln\ndigamma\nentr\nerf\n"
},
{
"change_type": "MODIFY",
"old_path": "jax/_src/scipy/special.py",
"new_... | Python | Apache License 2.0 | google/jax | Added more accurate version of the betaln function. |
260,428 | 29.11.2022 15:15:40 | 28,800 | 92fd8534cda6e464478201879a0d5a5a266dd4d8 | Fixes b/259636412, all-gather failing when called within xmap in pjit. Piece by piece making xmap in pjit work with all collectives so that we can use it to write 'manual kernels' safely! | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/parallel.py",
"new_path": "jax/_src/lax/parallel.py",
"diff": "@@ -1157,6 +1157,9 @@ def _all_gather_lowering(ctx, x, *, all_gather_dimension, axis_name,\n# TODO(jekbradbury): enable for all_gather_dimension > 0\nx_aval, = ctx.avals_in\nout_aval, ... | Python | Apache License 2.0 | google/jax | Fixes b/259636412, all-gather failing when called within xmap in pjit. Piece by piece making xmap in pjit work with all collectives so that we can use it to write 'manual kernels' safely!
PiperOrigin-RevId: 491749906 |
260,267 | 30.11.2022 16:38:27 | 0 | b5dc0638a2b35d9e0e41b49f7c79f402e61d38fa | Fix typo in docs/multi_process.md | [
{
"change_type": "MODIFY",
"old_path": "docs/multi_process.md",
"new_path": "docs/multi_process.md",
"diff": "@@ -91,7 +91,7 @@ jax.distributed.initialize()\n```\nOn TPU at present calling {func}`jax.distributed.initialize` is optional, but\n-recommanded since it enables additional checkpointing and... | Python | Apache License 2.0 | google/jax | Fix typo in docs/multi_process.md |
260,455 | 30.11.2022 09:56:21 | 28,800 | d5a058c7a885208bfd6f23d36e89ccd954795233 | doc improvement on initilaizer | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/nn/initializers.py",
"new_path": "jax/_src/nn/initializers.py",
"diff": "@@ -241,7 +241,8 @@ def variance_scaling(\n* a uniform interval, if `dtype` is real, or\n* a uniform disk, if `dtype` is complex,\n- with a mean of zero and a standard deviation ... | Python | Apache License 2.0 | google/jax | doc improvement on initilaizer
PiperOrigin-RevId: 491947286 |
260,411 | 01.12.2022 11:18:47 | -3,600 | fcaf7f1169ca94c7dea1b24f298b0e02a98b8d2d | [jax2tf] Fix the handling of jnp.roll for polymorphic shapes | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/numpy/lax_numpy.py",
"new_path": "jax/_src/numpy/lax_numpy.py",
"diff": "@@ -3507,7 +3507,7 @@ def _roll(a, shift, axis):\ni = _canonicalize_axis(i, a_ndim)\nx = remainder(x, (a_shape[i] or 1))\na = lax.concatenate((a, a), i)\n- a = lax.dynamic_slice_... | Python | Apache License 2.0 | google/jax | [jax2tf] Fix the handling of jnp.roll for polymorphic shapes |
260,578 | 01.12.2022 14:27:45 | 28,800 | 70d50814b1598e1dbcfb69fb748ea2c2d99ad594 | Add cross-linking for the migration guide and the parallelism with JAX
tutorial | [
{
"change_type": "MODIFY",
"old_path": "docs/jax_array_migration.md",
"new_path": "docs/jax_array_migration.md",
"diff": "+(jax-array-migration)=\n# jax.Array migration\n**yashkatariya@**\n@@ -17,6 +18,8 @@ the unified jax.Array\nAfter the migration is complete `jax.Array` will be the only type of a... | Python | Apache License 2.0 | google/jax | Add cross-linking for the migration guide and the parallelism with JAX
tutorial
Co-authored-by: Skye Wanderman-Milne <skyewm@google.com> |
260,578 | 01.12.2022 15:38:38 | 28,800 | a2870a182e1d935442362957087d92a34fa60ae5 | Add jax.Array to the index page | [
{
"change_type": "MODIFY",
"old_path": "docs/index.rst",
"new_path": "docs/index.rst",
"diff": "@@ -5,6 +5,11 @@ JAX is Autograd_ and XLA_, brought together for high-performance numerical compu\nIt provides composable transformations of Python+NumPy programs: differentiate, vectorize,\nparallelize, ... | Python | Apache License 2.0 | google/jax | Add jax.Array to the index page |
260,290 | 03.12.2022 21:22:25 | 0 | 542b38a1c232f7c28e8c25ac4d0fd8dd0598ae00 | Updated jax.tree_leaves --> jax.tree_util.tree_leaves to remove deprecation notice in jax101-pytrees tutorial | [
{
"change_type": "MODIFY",
"old_path": "docs/jax-101/05.1-pytrees.ipynb",
"new_path": "docs/jax-101/05.1-pytrees.ipynb",
"diff": "\"\\n\",\n\"# Let's see how many leaves they have:\\n\",\n\"for pytree in example_trees:\\n\",\n- \" leaves = jax.tree_leaves(pytree)\\n\",\n+ \" leaves = jax.tree_util.t... | Python | Apache License 2.0 | google/jax | Updated jax.tree_leaves --> jax.tree_util.tree_leaves to remove deprecation notice in jax101-pytrees tutorial
Signed-off-by: Simon Butt <simonbutt123@gmail.com> |
260,310 | 05.12.2022 18:02:05 | 28,800 | 2fc7bbce49bbe8c77deeda24e8118e818b8fe645 | Cache the exec time `tf.get_current_name_scope` into `_thread_local_state.exec_time_tf_name_scope` and add it as prefix during tracing.
Also move the test cases to right code location. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -160,7 +160,6 @@ _has_registered_tf_source_path = False\nclass _ThreadLocalState(threading.local):\ndef __init__(self):\n- self.name_stack = \"\"\n# XLA is not lin... | Python | Apache License 2.0 | google/jax | Cache the exec time `tf.get_current_name_scope` into `_thread_local_state.exec_time_tf_name_scope` and add it as prefix during tracing.
Also move the test cases to right code location.
PiperOrigin-RevId: 493163018 |
260,667 | 07.12.2022 04:24:29 | 28,800 | 6e0c8029cc2c108c7fb14d555575403a1556ea46 | Fix type annotation for bcoo_update_layout. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/sparse/bcoo.py",
"new_path": "jax/experimental/sparse/bcoo.py",
"diff": "@@ -1565,7 +1565,7 @@ mlir.register_lowering(bcoo_sum_duplicates_p, _bcoo_sum_duplicates_mhlo)\n# BCOO functions that maybe should be primitives?\ndef bcoo_update_layout(... | Python | Apache License 2.0 | google/jax | Fix type annotation for bcoo_update_layout.
PiperOrigin-RevId: 493567424 |
260,710 | 15.11.2022 18:40:52 | -32,400 | 1ade5f859210ff5e70f671ed6f468dda9b8cf2ef | Add `jax.scipy.linalg.toeplitz`. | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.scipy.rst",
"new_path": "docs/jax.scipy.rst",
"diff": "@@ -45,6 +45,7 @@ jax.scipy.linalg\nsolve_triangular\nsqrtm\nsvd\n+ toeplitz\ntril\ntriu\n"
},
{
"change_type": "MODIFY",
"old_path": "jax/_src/scipy/linalg.py",
"new_path": "jax/_... | Python | Apache License 2.0 | google/jax | Add `jax.scipy.linalg.toeplitz`. |
260,447 | 08.12.2022 16:54:15 | 28,800 | 942aa7a9070122e63953a1cd7a8f34c3baf6ca97 | [sparse] Move _dot_general_validated_shape to sparse util. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/sparse/bcoo.py",
"new_path": "jax/experimental/sparse/bcoo.py",
"diff": "@@ -31,7 +31,8 @@ from jax import vmap\nfrom jax.config import config\nfrom jax.experimental.sparse._base import JAXSparse\nfrom jax.experimental.sparse.util import (\n- ... | Python | Apache License 2.0 | google/jax | [sparse] Move _dot_general_validated_shape to sparse util.
PiperOrigin-RevId: 494031113 |
260,411 | 07.12.2022 10:43:33 | -7,200 | 86a70ab811c2e1dfd124ede7a90887f65eff9822 | [jax2tf] Fix for jnp.roll with shape polymorphism
There was a partial fix before, in but it was incomplete
and the x64 mode was not handled properly.
There are no tests added here; this was discovered by running the
tests with --jax2tf_default_experimental_native_lowering, which
will become default soon. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/numpy/lax_numpy.py",
"new_path": "jax/_src/numpy/lax_numpy.py",
"diff": "@@ -3506,9 +3506,11 @@ def _roll(a, shift, axis):\nfor x, i in zip(broadcast_to(shift, b_shape),\nnp.broadcast_to(axis, b_shape)):\ni = _canonicalize_axis(i, a_ndim)\n- x = remai... | Python | Apache License 2.0 | google/jax | [jax2tf] Fix for jnp.roll with shape polymorphism
There was a partial fix before, in #13470, but it was incomplete
and the x64 mode was not handled properly.
There are no tests added here; this was discovered by running the
tests with --jax2tf_default_experimental_native_lowering, which
will become default soon. |
260,424 | 08.12.2022 16:16:49 | 0 | 7fe466c548c603807ac708fa66c76dc774d6c810 | Small fix to scan type-check error message. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/control_flow/loops.py",
"new_path": "jax/_src/lax/control_flow/loops.py",
"diff": "@@ -965,7 +965,7 @@ def _scan_typecheck(bind_time, *in_atoms, reverse, length, num_consts, num_carry\nif not all(_map(core.typecompat, x_avals_jaxpr, x_avals_mapped... | Python | Apache License 2.0 | google/jax | Small fix to scan type-check error message. |
260,447 | 09.12.2022 10:09:48 | 28,800 | 11fbe5542ea7e215bb00de5d1cf38fec385c8f34 | [sparse] Add rand_bcsr to generate a random BCSR array. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/sparse/test_util.py",
"new_path": "jax/experimental/sparse/test_util.py",
"diff": "# See the License for the specific language governing permissions and\n# limitations under the License.\n\"\"\"Sparse test utilities.\"\"\"\n+import functools\n... | Python | Apache License 2.0 | google/jax | [sparse] Add rand_bcsr to generate a random BCSR array.
PiperOrigin-RevId: 494201364 |
260,447 | 09.12.2022 15:22:15 | 28,800 | a8b90b325a2d90aa0c2ca16cd10feac97f9b51ae | [sparse] Fix a bug in BCSR tree_flatten. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/sparse/bcsr.py",
"new_path": "jax/experimental/sparse/bcsr.py",
"diff": "@@ -22,6 +22,7 @@ from typing import NamedTuple, Optional, Sequence, Tuple\nimport numpy as np\nfrom jax import core\n+from jax import tree_util\nfrom jax.experimental.sp... | Python | Apache License 2.0 | google/jax | [sparse] Fix a bug in BCSR tree_flatten.
PiperOrigin-RevId: 494276415 |
260,424 | 09.12.2022 14:01:44 | 0 | 3db909ee8df4567f1f51670cc2b79fe9191a01a0 | Checkify: Add documentation for adding run-time values to error message. | [
{
"change_type": "MODIFY",
"old_path": "docs/debugging/checkify_guide.md",
"new_path": "docs/debugging/checkify_guide.md",
"diff": "@@ -8,16 +8,16 @@ import jax\nimport jax.numpy as jnp\ndef f(x, i):\n- checkify.check(i >= 0, \"index needs to be non-negative!\")\n+ checkify.check(i >= 0, \"index nee... | Python | Apache License 2.0 | google/jax | Checkify: Add documentation for adding run-time values to error message. |
260,464 | 13.12.2022 13:02:40 | 28,800 | 0e254a1f8146f912482f2a2660058b1482d23b01 | jupytext version bump | [
{
"change_type": "MODIFY",
"old_path": "docs/notebooks/Custom_Operation_for_GPUs.md",
"new_path": "docs/notebooks/Custom_Operation_for_GPUs.md",
"diff": "@@ -5,7 +5,7 @@ jupytext:\nextension: .md\nformat_name: myst\nformat_version: 0.13\n- jupytext_version: 1.13.8\n+ jupytext_version: 1.14.1\nkernel... | Python | Apache License 2.0 | google/jax | jupytext version bump |
260,424 | 12.12.2022 17:49:56 | 0 | 3134797968011e67168a74068e43606bd10cb89a | Add checkify.debug_check which is a noop outside of checkify. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/checkify.py",
"new_path": "jax/_src/checkify.py",
"diff": "@@ -243,7 +243,7 @@ class Error:\nreturn None\ndef throw(self):\n- check_error(self)\n+ _check_error(self)\ndef __str__(self):\nreturn f'Error({self.get()})'\n@@ -605,7 +605,7 @@ def check(pre... | Python | Apache License 2.0 | google/jax | Add checkify.debug_check which is a noop outside of checkify. |
260,411 | 04.12.2022 09:15:11 | -7,200 | ac7740513d0b47894d9170af6aaa6b9355fb2059 | Raise error for unsupported shape polymorphism for custom call and fallback lowering | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/fft.py",
"new_path": "jax/_src/lax/fft.py",
"diff": "@@ -19,7 +19,7 @@ from typing import Union, Sequence\nimport numpy as np\nfrom jax._src.api import jit, linear_transpose, ShapeDtypeStruct\n-from jax.core import Primitive\n+from jax.core import... | Python | Apache License 2.0 | google/jax | Raise error for unsupported shape polymorphism for custom call and fallback lowering |
260,464 | 14.12.2022 23:41:57 | 28,800 | 3cacb9a35a9e0d144490f31be9576d4279a8f8c6 | Remove unnecessary lines with +++ | [
{
"change_type": "MODIFY",
"old_path": "docs/Custom_Operation_for_GPUs.md",
"new_path": "docs/Custom_Operation_for_GPUs.md",
"diff": "@@ -5,16 +5,12 @@ To accommodate such scenarios, JAX allows users to define custom operations and\nThis tutorial contains information from [Extending JAX with custom ... | Python | Apache License 2.0 | google/jax | Remove unnecessary lines with +++ |
260,464 | 14.12.2022 23:43:59 | 28,800 | ef1a0cba6825d96fc78154307291b71a154b16ca | Set code block highlighting with python | [
{
"change_type": "MODIFY",
"old_path": "docs/Custom_Operation_for_GPUs.md",
"new_path": "docs/Custom_Operation_for_GPUs.md",
"diff": "@@ -106,7 +106,7 @@ We first create primitive operations, `_rms_norm_fwd_p` and `_rms_norm_bwd_p`, w\nWe set the `multipe_results` attribute to `True` for these opera... | Python | Apache License 2.0 | google/jax | Set code block highlighting with python |
260,411 | 16.12.2022 08:50:30 | -7,200 | fe6418a8a0c3a588a56df6b83ef857b1e33cad85 | [jax2tf] Force keep_unused for native lowering when we have shape polymorphism
In presence of dimension variables we conservatively do not drop unused inputs
because we may drop the only inputs from whose shape we can infer the
values of the dimension variables.
See b/261971607. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/api.py",
"new_path": "jax/_src/api.py",
"diff": "@@ -742,6 +742,12 @@ def _jit_lower(fun, static_argnums, static_argnames, device, backend,\nif abstracted_axes:\nraise ValueError(\"abstracted_axes must be used with --jax_dynamic_shapes\")\nin_avals, _... | Python | Apache License 2.0 | google/jax | [jax2tf] Force keep_unused for native lowering when we have shape polymorphism
In presence of dimension variables we conservatively do not drop unused inputs
because we may drop the only inputs from whose shape we can infer the
values of the dimension variables.
See b/261971607. |
260,464 | 16.12.2022 12:27:49 | 28,800 | 8a0b4d947d24e7d4dde5c405c7a3983a3ba33c57 | inline xmap_axes in the xmap call | [
{
"change_type": "MODIFY",
"old_path": "docs/Custom_Operation_for_GPUs.md",
"new_path": "docs/Custom_Operation_for_GPUs.md",
"diff": "@@ -542,11 +542,10 @@ from jax.experimental.maps import xmap\ndef xmap_rms_norm(x, weight, *, device_count):\nreshaped = x.reshape(device_count, x.shape[0] // device_... | Python | Apache License 2.0 | google/jax | inline xmap_axes in the xmap call |
260,447 | 16.12.2022 15:00:06 | 28,800 | 49086e84558c0ae749f6766e731104cffa0d2558 | [sparse] Use SparseInfo for both bcoo and bcsr format. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/sparse/bcoo.py",
"new_path": "jax/experimental/sparse/bcoo.py",
"diff": "@@ -33,7 +33,8 @@ from jax.experimental.sparse._base import JAXSparse\nfrom jax.experimental.sparse.util import (\n_broadcasting_vmap, _count_stored_elements,\n_dot_gener... | Python | Apache License 2.0 | google/jax | [sparse] Use SparseInfo for both bcoo and bcsr format.
PiperOrigin-RevId: 495968517 |
260,464 | 16.12.2022 15:05:55 | 28,800 | 40ae0d2b58f63776760e07ac3084998991e6d934 | Use bfloat16 instead of float16 and add jtu.check_grads | [
{
"change_type": "MODIFY",
"old_path": "docs/Custom_Operation_for_GPUs.md",
"new_path": "docs/Custom_Operation_for_GPUs.md",
"diff": "@@ -112,6 +112,7 @@ from functools import partial\nimport jax\nimport jax.numpy as jnp\n+import jax._src.test_util as jtu\nfrom build import gpu_ops\nfrom jax import ... | Python | Apache License 2.0 | google/jax | Use bfloat16 instead of float16 and add jtu.check_grads |
260,464 | 16.12.2022 17:06:43 | 28,800 | 4a63488825f39bf4b5a198370f854e2803385b50 | Add a hyperlink to jax.numpy` | [
{
"change_type": "MODIFY",
"old_path": "docs/Custom_Operation_for_GPUs.md",
"new_path": "docs/Custom_Operation_for_GPUs.md",
"diff": "@@ -8,11 +8,11 @@ This tutorial contains information from [Extending JAX with custom C++ and CUDA\n# RMS Normalization\nFor this tutorial, we are going to add the RMS... | Python | Apache License 2.0 | google/jax | Add a hyperlink to jax.numpy` |
260,411 | 17.12.2022 22:43:58 | -7,200 | 63a9e71b4e7cacbca71a490fcc53eff527bd3a35 | [jax2tf] Update the select_and_gather_add lowering for dynamic shapes
This enables lowering of select_and_gather_add for GPU when we have
dynamic shapes. The main change is to carry the abstract values and
to use them for broadcast, using mlir.broadcast_in_dim, which can
general BroadcastInDimOp or DynamicBroadcastInDimOp. This requires
some passing around of abstract values. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/windowed_reductions.py",
"new_path": "jax/_src/lax/windowed_reductions.py",
"diff": "@@ -654,7 +654,8 @@ def _select_and_gather_add_shape_rule(\nwindow_dilation)\ndef _select_and_gather_add_lowering(\n- ctx, tangents, operand, *, select_prim,\n+ c... | Python | Apache License 2.0 | google/jax | [jax2tf] Update the select_and_gather_add lowering for dynamic shapes
This enables lowering of select_and_gather_add for GPU when we have
dynamic shapes. The main change is to carry the abstract values and
to use them for broadcast, using mlir.broadcast_in_dim, which can
general BroadcastInDimOp or DynamicBroadcastInDimOp. This requires
some passing around of abstract values. |
260,294 | 19.12.2022 21:05:17 | 0 | 3391a5e3850a2aed8e189c04fe0624fcb0f9f301 | [ROCm]: Disable some tests on ROCm platform | [
{
"change_type": "MODIFY",
"old_path": "tests/experimental_rnn_test.py",
"new_path": "tests/experimental_rnn_test.py",
"diff": "@@ -33,7 +33,7 @@ class RnnTest(jtu.JaxTestCase):\nnum_layers=[1, 4],\nbidirectional=[True, False],\n)\n- @jtu.skip_on_devices(\"cpu\", \"tpu\")\n+ @jtu.skip_on_devices(\"c... | Python | Apache License 2.0 | google/jax | [ROCm]: Disable some tests on ROCm platform |
260,447 | 19.12.2022 14:16:50 | 28,800 | bc34af9f30e03ac34b7424dd979b35107ae11190 | [sparse] Add bcsr dot_general | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/sparse/__init__.py",
"new_path": "jax/experimental/sparse/__init__.py",
"diff": "@@ -223,6 +223,8 @@ from jax.experimental.sparse.bcoo import (\n)\nfrom jax.experimental.sparse.bcsr import (\n+ bcsr_dot_general as bcsr_dot_general,\n+ bcsr_dot... | Python | Apache License 2.0 | google/jax | [sparse] Add bcsr dot_general
PiperOrigin-RevId: 496489368 |
260,411 | 20.12.2022 15:29:51 | -7,200 | 61505d11d2b4512b613a73013cf27476dbfe3d08 | [jax2tf] More improvements to shape_poly_test
* Small cleanup inside PolyHarness.run_test to use contextlib.ExitContext,
and to run the TF eager mode conversion before the tf.function mode, because
the latter is unfriendly to breakpoints and makes debugging harder.
* Cleanup of the code to exclude some tests for native serialization. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/tests/shape_poly_test.py",
"new_path": "jax/experimental/jax2tf/tests/shape_poly_test.py",
"diff": "# See the License for the specific language governing permissions and\n# limitations under the License.\n\"\"\"Tests for the shape-polym... | Python | Apache License 2.0 | google/jax | [jax2tf] More improvements to shape_poly_test
* Small cleanup inside PolyHarness.run_test to use contextlib.ExitContext,
and to run the TF eager mode conversion before the tf.function mode, because
the latter is unfriendly to breakpoints and makes debugging harder.
* Cleanup of the code to exclude some tests for native serialization. |
260,411 | 20.12.2022 15:42:09 | -7,200 | a51b1744607479906f5dfbdada3cf368db21d817 | [jax2tf] Improved error checking for jnp.squeeze with shape polymorphism | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/numpy/lax_numpy.py",
"new_path": "jax/_src/numpy/lax_numpy.py",
"diff": "@@ -908,6 +908,9 @@ def squeeze(a: ArrayLike, axis: Optional[Union[int, Tuple[int, ...]]] = None) ->\ndef _squeeze(a: Array, axis: Tuple[int]) -> Array:\nif axis is None:\na_shap... | Python | Apache License 2.0 | google/jax | [jax2tf] Improved error checking for jnp.squeeze with shape polymorphism |
260,335 | 20.12.2022 12:00:46 | 28,800 | 580fdb6e15728acffbda0b863e76f46d3f997134 | tweak print_saved_residuals | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/ad_checkpoint.py",
"new_path": "jax/_src/ad_checkpoint.py",
"diff": "@@ -396,8 +396,12 @@ def saved_residuals(f, *args, **kwargs) -> List[Tuple[core.AbstractValue, str]]:\nif v in res_vars:\nif eqn.primitive is name_p:\nresults.append((v.aval, f\"name... | Python | Apache License 2.0 | google/jax | tweak print_saved_residuals |
260,335 | 20.12.2022 21:06:09 | 28,800 | 6ba0ef650558086322d4b1380d3650732ece9dd3 | relax tanh test tols for upcoming xla change | [
{
"change_type": "MODIFY",
"old_path": "tests/jet_test.py",
"new_path": "tests/jet_test.py",
"diff": "@@ -281,7 +281,8 @@ class JetTest(jtu.JaxTestCase):\n@jtu.skip_on_devices(\"tpu\")\ndef test_cosh(self): self.unary_check(jnp.cosh)\n@jtu.skip_on_devices(\"tpu\")\n- def test_tanh(self): self.unary_... | Python | Apache License 2.0 | google/jax | relax tanh test tols for upcoming xla change |
260,411 | 20.12.2022 16:20:14 | -7,200 | 0a0ee7873801b64d4da2d01ca1eddc9013c0ae6c | [jax2tf] Fixes for handling of opaque dtypes for slice/update_slice/gather | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -857,8 +857,10 @@ def _jax_physical_aval(aval: core.ShapedArray) -> core.ShapedArray:\nthere is only one and return it.\n\"\"\"\nif core.is_opaque_dtype(aval.dtype... | Python | Apache License 2.0 | google/jax | [jax2tf] Fixes for handling of opaque dtypes for slice/update_slice/gather |
260,411 | 21.12.2022 09:14:46 | -7,200 | 5812957186dd8024fe20f3cba08b2be7b098ceb8 | [jax2tf] Add autograph=False everywhere we use tf.function | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -87,7 +87,8 @@ f_tf_graph = tf.function(f_tf, autograph=False)\nThe Autograph feature of `tf.function` cannot be expected to work on\nfunctions lowered from JAX as... | Python | Apache License 2.0 | google/jax | [jax2tf] Add autograph=False everywhere we use tf.function |
260,411 | 21.12.2022 10:22:12 | -7,200 | a74c74c25572eec23c28e08dbe67781a23be19fb | Fix scatter in CLIP mode with uint32 and uint64 indices
Clipping uses np.iinfo(indices.dtype).max and those values
are too large to be converted to Python or C constants. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/slicing.py",
"new_path": "jax/_src/lax/slicing.py",
"diff": "@@ -1562,7 +1562,8 @@ def _clamp_scatter_indices(operand, indices, updates, *, dnums):\nfor i in dnums.scatter_dims_to_operand_dims)\n# Stack upper_bounds into a Array[n]\nupper_bound = ... | Python | Apache License 2.0 | google/jax | Fix scatter in CLIP mode with uint32 and uint64 indices
Clipping uses np.iinfo(indices.dtype).max and those values
are too large to be converted to Python or C constants. |
260,447 | 21.12.2022 09:54:38 | 28,800 | e89b60e3832d2ee16b1ae31cd73a23b8bc4f64d5 | [sparse] Propagate SparseInfo to BCSR todense() and tree_(un)flatten(). | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/sparse/bcsr.py",
"new_path": "jax/experimental/sparse/bcsr.py",
"diff": "@@ -173,7 +173,7 @@ def _bcsr_fromdense_impl(mat, *, nse, n_batch, n_dense, index_dtype):\n@bcsr_fromdense_p.def_abstract_eval\n-def _bcoo_fromdense_abstract_eval(mat, *,... | Python | Apache License 2.0 | google/jax | [sparse] Propagate SparseInfo to BCSR todense() and tree_(un)flatten().
PiperOrigin-RevId: 496945167 |
260,335 | 21.12.2022 10:16:18 | 28,800 | c2d9b5cee6b5eaa94618c2bef73866182a583e5a | tweak dot_general pretty-printing rule to suppress default params | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/lax/lax.py",
"new_path": "jax/_src/lax/lax.py",
"diff": "@@ -2566,7 +2566,8 @@ def _dot_general_transpose_lhs(g, y, *, dimension_numbers, precision,\ndims = ((ans_y, y_kept), (ans_batch, y_batch))\nx_contract_sorted_by_y = list(np.take(x_contract, np.... | Python | Apache License 2.0 | google/jax | tweak dot_general pretty-printing rule to suppress default params |
260,599 | 16.12.2022 10:12:40 | 0 | c0d4ae0cc3252811861b65036f247a583c50bdff | Added scipy.stats.bernoulli cdf and ppf. | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.scipy.rst",
"new_path": "docs/jax.scipy.rst",
"diff": "@@ -159,6 +159,8 @@ jax.scipy.stats.bernoulli\nlogpmf\npmf\n+ cdf\n+ ppf\njax.scipy.stats.beta\n~~~~~~~~~~~~~~~~~~~~\n"
},
{
"change_type": "MODIFY",
"old_path": "jax/_src/scipy/stats/... | Python | Apache License 2.0 | google/jax | Added scipy.stats.bernoulli cdf and ppf. |
260,631 | 23.12.2022 22:46:21 | 28,800 | 398aaaacc7eb40337d097ff0b4ff53d46f7f76ea | Add support for Ellipsis as an index for stateful operations. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/state/primitives.py",
"new_path": "jax/_src/state/primitives.py",
"diff": "@@ -65,9 +65,12 @@ def _get_impl(ref: Ref, *idx: int, **_):\nget_p.def_impl(_get_impl)\nIndexer = Tuple[Union[int, slice, Array], ...]\n+# or Ellipsis, but that can't be annota... | Python | Apache License 2.0 | google/jax | Add support for Ellipsis as an index for stateful operations.
PiperOrigin-RevId: 497466879 |
260,411 | 27.12.2022 15:28:04 | -7,200 | 0752655ff0338b6842fafb6c888f6da8d50a6e63 | [jax2tf] Move pjit test into the right TestCase class
This test uses tfxla.call_module directly, belongs to
XlaCallModuleTest. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/tests/jax2tf_test.py",
"new_path": "jax/experimental/jax2tf/tests/jax2tf_test.py",
"diff": "@@ -1264,38 +1264,6 @@ class Jax2TfTest(tf_test_util.JaxToTfTestCase):\n# TODO(b/243146552) We can switch to ConvertAndCompare after this bug fi... | Python | Apache License 2.0 | google/jax | [jax2tf] Move pjit test into the right TestCase class
This test uses tfxla.call_module directly, belongs to
XlaCallModuleTest. |
260,411 | 28.12.2022 12:07:46 | -7,200 | 1e09602c08e31ef7e9c210013b0bb89ee2b437d6 | [jax2tf] Improve handling of dimension polynomials used with JAX arrays.
See new description in README.md. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/README.md",
"new_path": "jax/experimental/jax2tf/README.md",
"diff": "@@ -518,11 +518,16 @@ def reshape_tf(x):\nreturn tf.reshape(x, [tf.math.multiply(4, b)])\n```\n-While operations among dimension polynomials and constants are handled... | Python | Apache License 2.0 | google/jax | [jax2tf] Improve handling of dimension polynomials used with JAX arrays.
See new description in README.md. |
260,411 | 28.12.2022 17:53:13 | -7,200 | 9593741b731841cbf494409daf6630e2c939093d | [jax2tf] Fix lowering for tf.while_loop
Sometimes TF infers more specific shapes for the init_carry, and this has
led to errors: "enters the loop with shape (1,), but has shape (None,) after one iteration"
Unfortunately, I cannot construct a small test. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -2751,7 +2751,12 @@ def _while(*args: TfVal, cond_nconsts: int, cond_jaxpr: core.ClosedJaxpr,\nbody_tf_func = partial(_interpret_jaxpr, body_jaxpr, *body_consts,\n... | Python | Apache License 2.0 | google/jax | [jax2tf] Fix lowering for tf.while_loop
Sometimes TF infers more specific shapes for the init_carry, and this has
led to errors: "enters the loop with shape (1,), but has shape (None,) after one iteration"
Unfortunately, I cannot construct a small test. |
260,411 | 20.12.2022 11:01:51 | -7,200 | 71ce60012747bd4f4945a0c0c82ed57993c6d198 | [jax2tf] Ensure that dim_as_value returns int64 in x64 mode and int32 otherwise
Changes all the computations with dimensions to work in int64 if
JAX_ENABLE_X64 and int32 otherwise. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/core.py",
"new_path": "jax/_src/core.py",
"diff": "@@ -1871,7 +1871,10 @@ def stride_shape(s: Shape, window_size: Shape, window_stride: Shape) -> Shape:\ndef dimension_as_value(d: DimSize):\n\"\"\"Turns a dimension size into a JAX value that we can co... | Python | Apache License 2.0 | google/jax | [jax2tf] Ensure that dim_as_value returns int64 in x64 mode and int32 otherwise
Changes all the computations with dimensions to work in int64 if
JAX_ENABLE_X64 and int32 otherwise. |
260,510 | 28.12.2022 22:55:13 | 28,800 | 48eb39a9b893bede0d072eaf6bca5b51cb6c4f2a | Remove dummy recv in TPU callback lowering, seems like it is no longer necessary | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/mlir.py",
"new_path": "jax/interpreters/mlir.py",
"diff": "@@ -1705,26 +1705,6 @@ def _emit_tpu_python_callback(\nrecv_channels = []\noutputs = []\n- # `send-to-host`s can be interleaved by the transfer manager so we add in a\n- # dummy recv t... | Python | Apache License 2.0 | google/jax | Remove dummy recv in TPU callback lowering, seems like it is no longer necessary
PiperOrigin-RevId: 498318923 |
260,511 | 30.12.2022 22:42:45 | 18,000 | 8df97ff2a304b20bd9730dc1aba4cd0e581314ed | Add `Optional` to `boundary` arg of `stft`. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/scipy/signal.py",
"new_path": "jax/_src/scipy/signal.py",
"diff": "@@ -439,7 +439,7 @@ def _spectral_helper(x: Array, y: Optional[ArrayLike], fs: ArrayLike = 1.0,\n@_wraps(osp_signal.stft)\ndef stft(x: Array, fs: ArrayLike = 1.0, window: str = 'hann',... | Python | Apache License 2.0 | google/jax | Add `Optional` to `boundary` arg of `stft`. |
260,335 | 04.01.2023 08:25:57 | 28,800 | 4f7cf622d4410d88b600feadd3224c48fa0ba0ed | tweak remat named policies | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/ad_checkpoint.py",
"new_path": "jax/_src/ad_checkpoint.py",
"diff": "@@ -71,8 +71,17 @@ def dot_with_no_batch_dims(prim, *_, **params) -> bool:\nname_p = core.Primitive('name')\n+def save_anything_except_these_names(*names_not_to_save):\n+ \"\"\"Save ... | Python | Apache License 2.0 | google/jax | tweak remat named policies |
260,287 | 05.01.2023 03:09:22 | 28,800 | f1635ca8757b72a0099ff586fb1719038c9599d5 | Skip flaky test on TPU | [
{
"change_type": "MODIFY",
"old_path": "tests/jaxpr_effects_test.py",
"new_path": "tests/jaxpr_effects_test.py",
"diff": "@@ -597,6 +597,7 @@ class EffectOrderingTest(jtu.JaxTestCase):\njax.effects_barrier()\nself.assertListEqual(log, [2., 3.])\n+ @jtu.skip_on_devices(\"tpu\")\ndef test_ordered_effe... | Python | Apache License 2.0 | google/jax | Skip flaky test on TPU
PiperOrigin-RevId: 499794466 |
260,287 | 05.01.2023 05:19:32 | 28,800 | 6655f2ba8d37a0d5ef59c30f732abddd10e8e996 | Skip gather and reduce scatter grad tests on GPU
Recent changes in XLA:GPU seem to be causing deadlocks. | [
{
"change_type": "MODIFY",
"old_path": "tests/pmap_test.py",
"new_path": "tests/pmap_test.py",
"diff": "@@ -974,6 +974,8 @@ class PythonPmapTest(jtu.JaxTestCase):\n))\n@ignore_slow_all_to_all_warning()\ndef testGradOf(self, prim, tiled, use_axis_index_groups):\n+ if jtu.device_under_test() == \"gpu\... | Python | Apache License 2.0 | google/jax | Skip gather and reduce scatter grad tests on GPU
Recent changes in XLA:GPU seem to be causing deadlocks.
PiperOrigin-RevId: 499832080 |
260,424 | 05.01.2023 14:27:17 | 0 | 0fe159b67e4fda1921fd6124ea166365c532ed36 | Make Shard.device and Shard.data read-only properties. | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/array.py",
"new_path": "jax/_src/array.py",
"diff": "@@ -55,10 +55,10 @@ class Shard:\ndef __init__(self, device: Device, sharding: Sharding, global_shape: Shape,\ndata: Optional[ArrayImpl] = None):\n- self.device = device\n+ self._device = device\nse... | Python | Apache License 2.0 | google/jax | Make Shard.device and Shard.data read-only properties. |
260,424 | 05.01.2023 17:04:53 | 0 | 8a4b9d6aad72da965bbce52ffd8dba4ab740746f | Fix typo in checkify guide. | [
{
"change_type": "MODIFY",
"old_path": "docs/debugging/checkify_guide.md",
"new_path": "docs/debugging/checkify_guide.md",
"diff": "@@ -187,7 +187,7 @@ def f(x, i):\ncheckify.check(i >= 0, \"index needs to be non-negative!\")\nreturn x[i]\n-checked_f = checkify.checkify(f, errors=checkify.all_errors... | Python | Apache License 2.0 | google/jax | Fix typo in checkify guide. |
260,411 | 08.01.2023 05:57:35 | -7,200 | 2844f506492c0202dab5d085953c9452cf81c799 | [jax2tf] Add more pjit tests
In particular, add a test with a closed-over constant, to test
for a pjit lowering behavior that will change soon. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/tests/jax2tf_test.py",
"new_path": "jax/experimental/jax2tf/tests/jax2tf_test.py",
"diff": "Specific JAX primitive conversion tests are in primitives_test.\"\"\"\nimport collections\n+from functools import partial\nimport os\nfrom typin... | Python | Apache License 2.0 | google/jax | [jax2tf] Add more pjit tests
In particular, add a test with a closed-over constant, to test
for a pjit lowering behavior that will change soon. |
260,717 | 10.01.2023 18:11:08 | -32,400 | 41b9c5e8cdb1d4bf65b643c3f0a07922a74812cc | [docs] donate_argnums FAQ link to rst format | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/api.py",
"new_path": "jax/_src/api.py",
"diff": "@@ -228,7 +228,7 @@ def jit(\narguments will not be donated.\nFor more details on buffer donation see the\n- [FAQ](https://jax.readthedocs.io/en/latest/faq.html#buffer-donation).\n+ `FAQ <https://jax.re... | Python | Apache License 2.0 | google/jax | [docs] donate_argnums FAQ link to rst format |
260,539 | 10.01.2023 09:14:10 | 28,800 | 7e395c9bbec6e9faa22eb4d553f57c20743fdaa0 | DOC: add note about localhost & friends in jax.distributed.initialize | [
{
"change_type": "MODIFY",
"old_path": "jax/_src/distributed.py",
"new_path": "jax/_src/distributed.py",
"diff": "@@ -131,6 +131,8 @@ def initialize(coordinator_address: Optional[str] = None,\nport does not matter, so long as the port is available on the coordinator\nand all processes agree on the p... | Python | Apache License 2.0 | google/jax | DOC: add note about localhost & friends in jax.distributed.initialize |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.