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
260,680
21.10.2021 16:12:33
-10,800
b056f796542c4a1714111f2450ed2dc659f8049f
Fix typo in doc for barrier_wait
[ { "change_type": "MODIFY", "old_path": "jax/experimental/host_callback.py", "new_path": "jax/experimental/host_callback.py", "diff": "@@ -181,7 +181,7 @@ of the computation, if all the callbacks are :func:`call`::\nres1 = jax.jit(device_fun)(1.)\nres2 = jax.jit(device_fun)(1.)\n- res2.block_until_re...
Python
Apache License 2.0
google/jax
Fix typo in doc for barrier_wait
260,631
21.10.2021 08:09:48
25,200
bc1c6b10904bb604b8a9fc664f766b003e0fdcc0
[JAX] Fix sphinx formatting issue
[ { "change_type": "MODIFY", "old_path": "jax/experimental/ann.py", "new_path": "jax/experimental/ann.py", "diff": "# See the License for the specific language governing permissions and\n# limitations under the License.\n-from functools import partial\n-from typing import (Any, Tuple)\n-\n-import nump...
Python
Apache License 2.0
google/jax
[JAX] Fix sphinx formatting issue PiperOrigin-RevId: 404797143
260,631
26.10.2021 06:56:05
25,200
7ae47cbc505698fd8a7fbad9493bf2260b340440
[JAX] Polish doc formatting for approx_top_k.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/ann.py", "new_path": "jax/experimental/ann.py", "diff": "@@ -33,11 +33,12 @@ top-1 from each window. The collision rate of the top-k affects the accuracy,\nwhich we can model by the number of people with shared birthdays in the Birthday\nprobl...
Python
Apache License 2.0
google/jax
[JAX] Polish doc formatting for approx_top_k. PiperOrigin-RevId: 405637140
260,335
22.07.2020 12:10:43
25,200
96623c3048351d4f43dee8d999fb440ef1a2c557
make iter(DeviceArray) return DeviceArrays w/o sync
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/lax.py", "new_path": "jax/_src/lax/lax.py", "diff": "@@ -919,7 +919,7 @@ def dynamic_slice(operand: Array, start_indices: Sequence[Array],\n\"\"\"\nstart_indices = _dynamic_slice_indices(operand, start_indices)\nreturn dynamic_slice_p.bind(operand...
Python
Apache License 2.0
google/jax
make iter(DeviceArray) return DeviceArrays w/o sync
260,510
06.10.2021 17:21:40
-7,200
ae9e69814ad4212ba0feec178c61970d5bad1196
Broadcast unmapped values in all_to_all batching rule Fixes
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/parallel.py", "new_path": "jax/_src/lax/parallel.py", "diff": "@@ -896,6 +896,12 @@ def _all_to_all_batched_collective(axis_size, frame_name, _, vals_in, dims_in,\nraise NotImplementedError(\"Please open a feature request!\")\nx, = vals_in\nd, = d...
Python
Apache License 2.0
google/jax
Broadcast unmapped values in all_to_all batching rule Fixes #7965. Co-authored-by: Sharad Vikram<sharad.vikram@gmail.com> Co-authored-by: Adam Paszke <apaszke@google.com>
260,335
27.10.2021 13:31:55
25,200
2bda894a30809340e2159c2973c327d82c3d8617
MeshComputation.hlo should be a method
[ { "change_type": "MODIFY", "old_path": "jax/interpreters/pxla.py", "new_path": "jax/interpreters/pxla.py", "diff": "@@ -1656,15 +1656,19 @@ def lower_mesh_computation(\nclass MeshComputation:\ndef __init__(self, hlo, *compile_args):\nself._executable = None\n- self.hlo = hlo\n+ self._hlo = hlo\nself...
Python
Apache License 2.0
google/jax
MeshComputation.hlo should be a method Co-authored-by: Brennan Saeta <saeta@google.com>
260,270
28.10.2021 18:37:15
-3,600
d890ae906854fa8a530e2e5b9670d21b2e7b3bf3
Use default backend if no backend supplied to xla_computation
[ { "change_type": "MODIFY", "old_path": "jax/_src/api.py", "new_path": "jax/_src/api.py", "diff": "@@ -772,6 +772,8 @@ def xla_computation(fun: Callable,\nfun_name = getattr(fun, \"__name__\", \"unknown\")\n+ backend = backend if backend is not None else xb.get_backend().platform\n+\ndef make_axis_en...
Python
Apache License 2.0
google/jax
Use default backend if no backend supplied to xla_computation
260,510
27.10.2021 15:20:45
25,200
b40245e38d7837a7777735ad60f3b5b1ac2d499d
Use `SourceInfo` named tuple to keep track of source information
[ { "change_type": "MODIFY", "old_path": "jax/_src/source_info_util.py", "new_path": "jax/_src/source_info_util.py", "diff": "@@ -16,7 +16,7 @@ import contextlib\nimport itertools\nimport os.path\nimport threading\n-from typing import Optional, Iterator\n+from typing import Optional, Iterator, NamedTu...
Python
Apache License 2.0
google/jax
Use `SourceInfo` named tuple to keep track of source information
260,631
28.10.2021 23:07:25
25,200
2ab00151edb3497f7fbbad394a866f1455743b5c
Copybara import of the project: by Sharad Vikram Use `SourceInfo` named tuple to keep track of source information
[ { "change_type": "MODIFY", "old_path": "jax/_src/source_info_util.py", "new_path": "jax/_src/source_info_util.py", "diff": "@@ -16,7 +16,7 @@ import contextlib\nimport itertools\nimport os.path\nimport threading\n-from typing import Optional, Iterator, NamedTuple\n+from typing import Optional, Itera...
Python
Apache License 2.0
google/jax
Copybara import of the project: -- b40245e38d7837a7777735ad60f3b5b1ac2d499d by Sharad Vikram <sharad.vikram@gmail.com>: Use `SourceInfo` named tuple to keep track of source information PiperOrigin-RevId: 406293469
260,447
29.10.2021 14:44:27
25,200
c5f73b3d8e84acbb7613c33c97b7317aad5907e9
[JAX] Added `jax.lax.linalg.qdwh`.
[ { "change_type": "MODIFY", "old_path": "CHANGELOG.md", "new_path": "CHANGELOG.md", "diff": "@@ -20,6 +20,8 @@ PLEASE REMEMBER TO CHANGE THE '..main' WITH AN ACTUAL TAG in GITHUB LINK.\n* Moved `jax.experimental.stax` to `jax.example_libraries.stax`\n* Moved `jax.experimental.optimizers` to `jax.exam...
Python
Apache License 2.0
google/jax
[JAX] Added `jax.lax.linalg.qdwh`. PiperOrigin-RevId: 406453671
260,510
29.10.2021 15:49:31
25,200
32319e1bc36e17ca270e9ff1a9545e6680f9eb28
Fix forward for PR (made `source_info` for `new_jaxpr_eqn` argument optional again)
[ { "change_type": "MODIFY", "old_path": "jax/_src/source_info_util.py", "new_path": "jax/_src/source_info_util.py", "diff": "@@ -16,7 +16,7 @@ import contextlib\nimport itertools\nimport os.path\nimport threading\n-from typing import Optional, Iterator\n+from typing import Optional, Iterator, NamedTu...
Python
Apache License 2.0
google/jax
Fix forward for PR #8392 (made `source_info` for `new_jaxpr_eqn` argument optional again) PiperOrigin-RevId: 406466709
260,447
02.11.2021 14:13:42
25,200
4814d75768d00901297103ea5fbb4750e649a4ab
Update coo_matvec and coo_matmat comments.
[ { "change_type": "MODIFY", "old_path": "jaxlib/cusparse.py", "new_path": "jaxlib/cusparse.py", "diff": "@@ -231,7 +231,7 @@ def coo_fromdense(c, mat, *, nnz, index_dtype):\nreturn tuple(_ops.GetTupleElement(out, i) for i in range(3))\ndef coo_matvec(c, data, row, col, x, *, shape, transpose=False, c...
Python
Apache License 2.0
google/jax
Update coo_matvec and coo_matmat comments.
260,411
04.11.2021 10:45:53
-3,600
27de09cc2b5c82a66fb6cf685e52fb3223372f43
[jax2tf] Improve the documentation for jax2tf and SavedModel In particular, document better how to avoid embedding large constants in the SavedModel.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/README.md", "new_path": "jax/experimental/jax2tf/README.md", "diff": "@@ -103,11 +103,12 @@ tf.saved_model.save(my_model, '/some/directory',\nrestored_model = tf.saved_model.load('/some/directory')\n```\n-An important point is that in t...
Python
Apache License 2.0
google/jax
[jax2tf] Improve the documentation for jax2tf and SavedModel In particular, document better how to avoid embedding large constants in the SavedModel.
260,335
04.11.2021 15:41:16
25,200
895f15d5e659569f30eae09ada01764d1e4a41d7
generalize lax.cond to accept any number of args
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/control_flow.py", "new_path": "jax/_src/lax/control_flow.py", "diff": "@@ -679,43 +679,45 @@ def switch(index, branches: Sequence[Callable], operand):\nreturn tree_unflatten(out_trees[0], out)\n-def _cond(pred, true_fun: Callable, false_fun: Calla...
Python
Apache License 2.0
google/jax
generalize lax.cond to accept any number of args
260,335
27.10.2021 20:27:09
25,200
05708aef2b43b8abd1705165b37e8498228d0969
jit(f).lower(...) works w/ duck typed shape/dtype
[ { "change_type": "MODIFY", "old_path": "jax/_src/api.py", "new_path": "jax/_src/api.py", "diff": "@@ -321,9 +321,6 @@ def _prepare_jit(fun, static_argnums, static_argnames, donate_argnums,\nelse:\ndonated_invars = (False,) * len(args_flat)\n- for arg in args_flat:\n- _check_arg(arg)\n-\nreturn f, in...
Python
Apache License 2.0
google/jax
jit(f).lower(...) works w/ duck typed shape/dtype
260,411
11.11.2021 13:21:55
-7,200
9175ed611df9e8037538f891740d531098228b47
[jax2tf] Improve the TF constant sharing Use fewer cache tables for constants: one per top-level converted function, and a separate table for the gradient.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "@@ -408,7 +408,8 @@ def convert(fun: Callable,\n@tf.custom_gradient\ndef converted_fun_flat_with_custom_gradient(*args_flat: TfVal) -> TfVal:\nout_with_avals = _inter...
Python
Apache License 2.0
google/jax
[jax2tf] Improve the TF constant sharing Use fewer cache tables for constants: one per top-level converted function, and a separate table for the gradient. Bug: #7992
260,496
11.11.2021 20:19:47
-3,600
3df11aee650ba1c9ee98b8d401eeef93c01cb8c2
Add numerical comparison for jax2tf -> TFLite
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/examples_eval/all_examples.py", "new_path": "jax/experimental/jax2tf/examples_eval/all_examples.py", "diff": "@@ -118,7 +118,7 @@ def _flax_examples():\nModuleSpec(\nmodule_path='ppo.models.ActorCritic',\ninput_shape=(1, 8, 8, 4),\n- mo...
Python
Apache License 2.0
google/jax
Add numerical comparison for jax2tf -> TFLite
260,456
11.11.2021 17:02:59
28,800
e511b280d8fcb286f1759c2ac84eb6ee594177fb
fix for translation to cudaDataType
[ { "change_type": "MODIFY", "old_path": "jaxlib/cusparse.cc", "new_path": "jaxlib/cusparse.cc", "diff": "@@ -59,7 +59,7 @@ cusparseIndexType_t DtypeToCuSparseIndexType(const py::dtype& np_type) {\ncudaDataType DtypeToCudaDataType(const py::dtype& np_type) {\nstatic auto* types =\nnew absl::flat_hash_...
Python
Apache License 2.0
google/jax
fix for translation to cudaDataType
260,496
15.11.2021 12:30:50
-3,600
2c17076fb4e10fc614b9f3b56bdd07b1664e81dc
Fix incorrect param ordering
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/examples_eval/converters.py", "new_path": "jax/experimental/jax2tf/examples_eval/converters.py", "diff": "@@ -95,7 +95,7 @@ def jax2tf_to_tflite(module: examples_converter.ModuleToConvert):\noutputs = tuple(interpreter.tensor(out[\"inde...
Python
Apache License 2.0
google/jax
Fix incorrect param ordering
260,496
15.11.2021 13:29:52
-3,600
775236aba6f5e0cb9cfe04492ee342e6d104863a
Update PPO shapes fixing numerical diff
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/examples_eval/all_examples.py", "new_path": "jax/experimental/jax2tf/examples_eval/all_examples.py", "diff": "@@ -117,7 +117,11 @@ def _flax_examples():\n'ppo':\nModuleSpec(\nmodule_path='ppo.models.ActorCritic',\n- input_shape=(1, 8, 8...
Python
Apache License 2.0
google/jax
Update PPO shapes fixing numerical diff
260,496
15.11.2021 15:19:19
-3,600
8bb2c766c76c1ac51c1f5bbcac849a68e740a3fa
Change random data and updates table
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/examples_eval/converters.py", "new_path": "jax/experimental/jax2tf/examples_eval/converters.py", "diff": "@@ -55,7 +55,7 @@ def _get_random_data(dtype: jnp.dtype, shape: Tuple[int, ...]) -> Any:\nif np.issubdtype(dtype, np.integer):\nre...
Python
Apache License 2.0
google/jax
Change random data and updates table
260,335
15.11.2021 13:24:33
28,800
57f4fc9bbb68442f0ef0ef27f04a4a14b49bdf7d
update readme to focus on most active libraries
[ { "change_type": "MODIFY", "old_path": "README.md", "new_path": "README.md", "diff": "| [**Reference docs**](https://jax.readthedocs.io/en/latest/)\n-**News:** [JAX tops largest-scale MLPerf Training 0.7 benchmarks!](https://cloud.google.com/blog/products/ai-machine-learning/google-breaks-ai-perform...
Python
Apache License 2.0
google/jax
update readme to focus on most active libraries
260,447
15.11.2021 22:36:27
28,800
21fa3b3f3b05b01938898942e1ae2941de739b9f
Change QR import in lax.linalg.qdwh.
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/qdwh.py", "new_path": "jax/_src/lax/qdwh.py", "diff": "@@ -37,7 +37,7 @@ def _use_qr(u, params):\na, b, c = params\nm, n = u.shape\ny = jnp.concatenate([jnp.sqrt(c) * u, jnp.eye(n)])\n- q, _ = jnp.linalg.qr(y)\n+ q, _ = lax_linalg.qr(y, full_matri...
Python
Apache License 2.0
google/jax
Change QR import in lax.linalg.qdwh.
260,335
16.11.2021 11:21:27
28,800
5d35b8a11903f8722519834f3748ee35e4cfee58
add donated_invars to xla.XlaComputation
[ { "change_type": "MODIFY", "old_path": "jax/_src/api.py", "new_path": "jax/_src/api.py", "diff": "@@ -488,17 +488,21 @@ class Lowered:\nquerying properties of lowered computations across JAX's various\nlowering paths (``jit``, ``pmap``, etc.).\n\"\"\"\n- __slots__ = ['in_tree', 'out_tree', '_lowerin...
Python
Apache License 2.0
google/jax
add donated_invars to xla.XlaComputation Co-authored-by: Brennan Saeta <saeta@google.com>
260,424
25.10.2021 15:53:52
-3,600
e14fea3b637ed4a67a2530106e1bf34c38676037
Overload jnp ops which are polymorphic to an array's value and support PRNGKeys.
[ { "change_type": "MODIFY", "old_path": "jax/_src/numpy/lax_numpy.py", "new_path": "jax/_src/numpy/lax_numpy.py", "diff": "@@ -30,7 +30,7 @@ import collections\nfrom functools import partial\nimport operator\nimport types\n-from typing import Sequence, FrozenSet, Optional, Tuple, Union\n+from typing ...
Python
Apache License 2.0
google/jax
Overload jnp ops which are polymorphic to an array's value and support PRNGKeys.
260,496
17.11.2021 15:09:06
-3,600
06d5e99ddfaf652fc9a4d4ce1f5b4d9c39970075
Add numerical comparison with TF
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/examples_eval/converters.py", "new_path": "jax/experimental/jax2tf/examples_eval/converters.py", "diff": "@@ -62,6 +62,22 @@ def _get_random_data(dtype: jnp.dtype, shape: Tuple[int, ...]) -> Any:\nraise ValueError(f\"Unsupported dtype f...
Python
Apache License 2.0
google/jax
Add numerical comparison with TF
260,411
05.11.2021 17:03:46
-7,200
75155f5eda4b298cbb9ac28a9ff0a2bd0176c24b
[shape_poly] Refactor arange and image_resize for shape polymorphism Small refactoring to jax.image.resize to make it compatible with shape polymorphismin jax2tf. In the process added also support for jnp.arange([dim_poly]). Note that the underlying lax.iota already supported shape polymorphism.
[ { "change_type": "MODIFY", "old_path": "jax/_src/image/scale.py", "new_path": "jax/_src/image/scale.py", "diff": "@@ -44,7 +44,8 @@ def _fill_triangle_kernel(x):\nreturn jnp.maximum(0, 1 - jnp.abs(x))\n-def compute_weight_mat(input_size: int, output_size: int, scale,\n+def compute_weight_mat(input_s...
Python
Apache License 2.0
google/jax
[shape_poly] Refactor arange and image_resize for shape polymorphism Bug: 8367 Small refactoring to jax.image.resize to make it compatible with shape polymorphismin jax2tf. In the process added also support for jnp.arange([dim_poly]). Note that the underlying lax.iota already supported shape polymorphism.
260,411
18.11.2021 10:23:53
-7,200
3715fcb930d5cd9ef6a454136f442d2317a68536
Added workaround for bug in XLA
[ { "change_type": "MODIFY", "old_path": "jax/_src/image/scale.py", "new_path": "jax/_src/image/scale.py", "diff": "@@ -233,7 +233,9 @@ def _resize_nearest(x, output_shape: core.Shape):\nm = input_shape[d]\nn = output_shape[d]\noffsets = (jnp.arange(n) + 0.5) * core.dimension_as_value(m) / core.dimens...
Python
Apache License 2.0
google/jax
Added workaround for bug in XLA
260,335
18.11.2021 13:40:25
28,800
2cb235809aeb0ee5b42ef129f7a6179d74905034
make vmap ppermute consistent with pmap/docstring This was a bad bug! Unfortunately our tests didn't catch it, in part because permutations on size-two axes are either trivial or not. The simplest test might have a size-three axis.
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/parallel.py", "new_path": "jax/_src/lax/parallel.py", "diff": "@@ -795,7 +795,7 @@ def _ppermute_batcher(axis_size, frame_name, _, vals_in, dims_in, axis_name, per\nassert d is not batching.not_mapped\nperm_indices = np.zeros(axis_size, dtype=int)...
Python
Apache License 2.0
google/jax
make vmap ppermute consistent with pmap/docstring This was a bad bug! Unfortunately our tests didn't catch it, in part because permutations on size-two axes are either trivial or not. The simplest test might have a size-three axis.
260,631
19.11.2021 13:31:29
28,800
cad3fd808d68dd12ad2bb5c32c3114af27aa1b81
Make it easier to see what are the difference between two structures. When structures are very large, users can end up with pages and pages describing the two structures, and finding exactly where they differ can be tricky. This change makes these differences more obvious.
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/control_flow.py", "new_path": "jax/_src/lax/control_flow.py", "diff": "@@ -2033,6 +2033,11 @@ def _rng_bit_generator_batching_rule(batched_args, batch_dims, *, shape, dtype,\nbatching.primitive_batchers[lax.rng_bit_generator_p] = _rng_bit_generato...
Python
Apache License 2.0
google/jax
Make it easier to see what are the difference between two structures. When structures are very large, users can end up with pages and pages describing the two structures, and finding exactly where they differ can be tricky. This change makes these differences more obvious. PiperOrigin-RevId: 411131921
260,496
18.11.2021 11:13:26
-3,600
7b25e05fd1ea7547f165664e7ad29d2818aae8ac
Improve logics for numerical comparison
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/examples_eval/all_examples.py", "new_path": "jax/experimental/jax2tf/examples_eval/all_examples.py", "diff": "@@ -117,11 +117,10 @@ def _flax_examples():\n'ppo':\nModuleSpec(\nmodule_path='ppo.models.ActorCritic',\n- # TODO(marcvanzee):...
Python
Apache License 2.0
google/jax
Improve logics for numerical comparison
260,411
23.11.2021 11:03:10
-7,200
72d9d35555f72b39a415f0b1eb841d3220627a7d
Fix lint and mypy errors
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "@@ -875,7 +875,8 @@ class TensorFlowTrace(core.Trace):\nstore_tf_res_avals = None\ndef f_tf(*tf_args):\nnonlocal store_tf_res_avals\n- tf_res_out: Sequence[Tuple[TfVa...
Python
Apache License 2.0
google/jax
Fix lint and mypy errors
260,411
23.11.2021 11:35:05
-7,200
ddc3a126e23fcc33f074632803c95a5ac4c54db7
Improve error when jnp.arange is used with non-constant arguments
[ { "change_type": "MODIFY", "old_path": "jax/_src/numpy/lax_numpy.py", "new_path": "jax/_src/numpy/lax_numpy.py", "diff": "@@ -3761,12 +3761,16 @@ def arange(start: core.DimSize, stop: Optional[core.DimSize]=None,\nlax._check_user_dtype_supported(dtype, \"arange\")\nrequire = partial(core.concrete_or...
Python
Apache License 2.0
google/jax
Improve error when jnp.arange is used with non-constant arguments
260,335
23.11.2021 15:51:49
28,800
8430deda3e1e7a231e2908b721f8165eac675225
custom pp_eqn rules, simpler xla_call print
[ { "change_type": "MODIFY", "old_path": "docs/jaxpr.rst", "new_path": "docs/jaxpr.rst", "diff": "@@ -415,7 +415,6 @@ which the computation should run. For example\n{ lambda ; a:f32[]. let\nb:f32[] = sub a 2.0\nc:f32[1] = xla_call[\n- backend=None\ncall_jaxpr={ lambda ; d:f32[] e:f32[]. let\nf:f32[1] ...
Python
Apache License 2.0
google/jax
custom pp_eqn rules, simpler xla_call print
260,411
24.11.2021 11:47:11
-7,200
277a1d775e871c522144d7c264354e3abe96c18b
[hcb] Cleanup to account for changes in minimum jaxlib version We can assume now that jaxlib has the support for CustomCall.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/host_callback.py", "new_path": "jax/experimental/host_callback.py", "diff": "@@ -478,19 +478,7 @@ def _inline_host_callback() -> bool:\ndef _use_outfeed(platform: str) -> bool:\n- if platform in (\"tpu\", \"gpu\") or FLAGS.jax_host_callback_ou...
Python
Apache License 2.0
google/jax
[hcb] Cleanup to account for changes in minimum jaxlib version We can assume now that jaxlib has the support for CustomCall.
260,562
24.11.2021 10:02:35
28,800
e6950dd63ed3c054067c53bd4806dfbb85e0eef7
[XLA] Adjust the error tolerance of 'lax_control_flow_test.py' We are about to change the vectorization strategy for XLA:CPU. As a result, we have to increase the error tolerance of the impacted tests.
[ { "change_type": "MODIFY", "old_path": "tests/lax_control_flow_test.py", "new_path": "tests/lax_control_flow_test.py", "diff": "@@ -1453,7 +1453,12 @@ class LaxControlFlowTest(jtu.JaxTestCase):\nans = scan(f, c, as_)\nexpected = scan_reference(f, c, as_)\n- self.assertAllClose(ans, expected, check_d...
Python
Apache License 2.0
google/jax
[XLA] Adjust the error tolerance of 'lax_control_flow_test.py' We are about to change the vectorization strategy for XLA:CPU. As a result, we have to increase the error tolerance of the impacted tests. PiperOrigin-RevId: 412083490
260,335
23.11.2021 15:03:24
28,800
17add1ce389ac6470e412ee81f7fd7b6251a184e
sipmlify array_copy implementation Use nested calls rather than an add-zero.
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/lax.py", "new_path": "jax/_src/lax/lax.py", "diff": "@@ -83,19 +83,18 @@ Shape = core.Shape\ndef _array_copy(arr):\n\"\"\"Return an on-device copy of a DeviceArray.\n- This is a private method; users can access this via ``jnp.array(x, copy=True)``...
Python
Apache License 2.0
google/jax
sipmlify array_copy implementation Use nested calls rather than an add-zero.
260,335
23.11.2021 15:23:28
28,800
cd1655ed30f0a758b1aacb42bc1756c40434fbb9
re-implement lax._array_copy as a primitive
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/lax.py", "new_path": "jax/_src/lax/lax.py", "diff": "@@ -79,23 +79,6 @@ Array = Any\nDType = Any\nShape = core.Shape\n-@functools.partial(jax.jit, inline=True)\n-def _array_copy(arr):\n- \"\"\"Return an on-device copy of a DeviceArray.\n-\n- This ...
Python
Apache License 2.0
google/jax
re-implement lax._array_copy as a primitive
260,335
24.11.2021 14:15:15
28,800
9a423bb4be4576c81a757009eed65db7601b74c2
add mlir lowering for copy_p
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/lax.py", "new_path": "jax/_src/lax/lax.py", "diff": "@@ -4120,6 +4120,7 @@ copy_p = core.Primitive('copy_p')\ncopy_p.def_impl(partial(xla.apply_primitive, copy_p))\ncopy_p.def_abstract_eval(lambda x: x)\nxla.register_translation(copy_p, lambda ctx...
Python
Apache License 2.0
google/jax
add mlir lowering for copy_p
260,335
24.11.2021 14:26:55
28,800
3d16a32986bc501a33c8fbec1ca1da15992d529d
add option for enabling custom jaxpr pprint rules
[ { "change_type": "MODIFY", "old_path": "jax/core.py", "new_path": "jax/core.py", "diff": "@@ -76,12 +76,14 @@ class Jaxpr:\nself.eqns = list(eqns)\ndef __str__(self):\n- return str(pp_jaxpr(self, JaxprPpContext()))\n+ return str(pp_jaxpr(self, JaxprPpContext(), custom_pp_eqn_rules=True))\n__repr__ =...
Python
Apache License 2.0
google/jax
add option for enabling custom jaxpr pprint rules
260,335
24.11.2021 14:30:07
28,800
63a8f62200467ec49231a7bfa3be4673d05922b8
add copy_p to jax2tf not implemented list
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "@@ -1003,6 +1003,7 @@ def _add(x: TfVal, y: TfVal) -> TfVal:\ntf_impl[ad_util.add_jaxvals_p] = _add\ntf_impl[dispatch.device_put_p] = lambda x, device=None: x\n+tf_im...
Python
Apache License 2.0
google/jax
add copy_p to jax2tf not implemented list
260,411
26.11.2021 09:26:54
-7,200
a94a8847c3802521725e08e91587079c6296fc1b
[jax2tf] Implement CLIP mode for scatter. Also handle a limited form of shape polymorphism, where the `operand.shape - update.shape` is a constant in the scatter dimensions, even when the shapes may contain dimension variables.
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/slicing.py", "new_path": "jax/_src/lax/slicing.py", "diff": "@@ -1430,10 +1430,6 @@ def _scatter_shape_rule(operand, indices, updates, *, update_jaxpr,\ndef _clamp_scatter_indices(operand, indices, updates, *, dnums):\n\"\"\"Clamps `indices` to be...
Python
Apache License 2.0
google/jax
[jax2tf] Implement CLIP mode for scatter. Also handle a limited form of shape polymorphism, where the `operand.shape - update.shape` is a constant in the scatter dimensions, even when the shapes may contain dimension variables.
260,562
29.11.2021 06:26:12
28,800
4f80e6af6eb6afb68ec68e1bedf4829cabf144e2
[XLA] More error tolerance adjustments due to incoming change in XLA:CPU We are about to change the vectorization strategy for XLA:CPU. As a result, we have to increase the error tolerance of the impacted tests.
[ { "change_type": "MODIFY", "old_path": "tests/lax_scipy_test.py", "new_path": "tests/lax_scipy_test.py", "diff": "@@ -406,7 +406,7 @@ class LaxBackedScipyTests(jtu.JaxTestCase):\nactual = jnp.real(\nlsp_special.sph_harm(jnp.array([0]), jnp.array([1]), theta, phi, n_max))\n- self.assertAllClose(actua...
Python
Apache License 2.0
google/jax
[XLA] More error tolerance adjustments due to incoming change in XLA:CPU We are about to change the vectorization strategy for XLA:CPU. As a result, we have to increase the error tolerance of the impacted tests. PiperOrigin-RevId: 412872724
260,424
30.11.2021 13:58:52
0
cb6a3f216f814f7687be3e1c4bbb2b5c5e0987cb
Leak checker: garbage collect before collecting hanging references.
[ { "change_type": "MODIFY", "old_path": "jax/core.py", "new_path": "jax/core.py", "diff": "@@ -793,6 +793,8 @@ def maybe_find_leaked_tracers(x: Optional[Union[MainTrace, Sublevel]]):\n\"\"\"\nif not getattr(threading.current_thread(), 'pydev_do_not_trace', True):\nwarnings.warn(TRACER_LEAK_DEBUGGER_W...
Python
Apache License 2.0
google/jax
Leak checker: garbage collect before collecting hanging references.
260,447
30.11.2021 15:47:50
28,800
19554e21d3da3d2caf4c1249e1e806d0e786bee5
Enable QDWH TPU tests.
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/qdwh.py", "new_path": "jax/_src/lax/qdwh.py", "diff": "@@ -180,6 +180,7 @@ def qdwh(x, is_symmetric, max_iterations=10):\n'`norm(x-x.H) / norm(x)` is {}, which is greater than '\n'the tolerance {}.'.format(relative_diff, tol))\n+ with jax.default_...
Python
Apache License 2.0
google/jax
Enable QDWH TPU tests.
260,411
29.11.2021 13:34:47
-7,200
6288d67fefec5dad8b0a42055e071d647ba1be06
Refactor the lowering of remat. Replace the XLA-specific lowering with the same logic written with LAX primitives. Then use this for HLO, MLIR, and jax2tf lowerings.
[ { "change_type": "MODIFY", "old_path": "jax/_src/ad_checkpoint.py", "new_path": "jax/_src/ad_checkpoint.py", "diff": "@@ -281,18 +281,6 @@ def remat_abstract_eval(*args, jaxpr, prevent_cse, differentiated, policy):\ndel args, prevent_cse, differentiated, policy # Unused.\nreturn [v.aval for v in jax...
Python
Apache License 2.0
google/jax
Refactor the lowering of remat. Replace the XLA-specific lowering with the same logic written with LAX primitives. Then use this for HLO, MLIR, and jax2tf lowerings.
260,411
30.11.2021 11:52:49
-7,200
b55e993be3aa2f0f846f6bb0935ae3d7c59a922a
[jax2tf] Add tests for rng_uniform These tests really only verify which dtypes are supported. The actual numeric comparison is disabled, because rng_uniform is statefull and calling multiple times produces different results.
[ { "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): 2021-07-31*\n+*Last gene...
Python
Apache License 2.0
google/jax
[jax2tf] Add tests for rng_uniform These tests really only verify which dtypes are supported. The actual numeric comparison is disabled, because rng_uniform is statefull and calling multiple times produces different results.
260,411
01.12.2021 11:34:37
-7,200
5b177967e62a25b63092d2a3b7dad146c5b8600b
[jax2tf] Improved error checking for call_tf for functions with dynamic shapes.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/README.md", "new_path": "jax/experimental/jax2tf/README.md", "diff": "@@ -988,6 +988,7 @@ The zero-copy does not yet work on TPU.\n### Limitations of call_tf\nThe TF function must be compileable (`tf.function(func, jit_compile=True`)\n+...
Python
Apache License 2.0
google/jax
[jax2tf] Improved error checking for call_tf for functions with dynamic shapes.
260,631
01.12.2021 08:42:35
28,800
610187cd332ee18b40bc9caf90e54db50f4d0659
Copybara import of the project: by George Necula Refactor the lowering of remat. Replace the XLA-specific lowering with the same logic written with LAX primitives. Then use this for HLO, MLIR, and jax2tf lowerings.
[ { "change_type": "MODIFY", "old_path": "jax/_src/ad_checkpoint.py", "new_path": "jax/_src/ad_checkpoint.py", "diff": "@@ -281,6 +281,18 @@ def remat_abstract_eval(*args, jaxpr, prevent_cse, differentiated, policy):\ndel args, prevent_cse, differentiated, policy # Unused.\nreturn [v.aval for v in jax...
Python
Apache License 2.0
google/jax
Copybara import of the project: -- 6288d67fefec5dad8b0a42055e071d647ba1be06 by George Necula <gcnecula@gmail.com>: Refactor the lowering of remat. Replace the XLA-specific lowering with the same logic written with LAX primitives. Then use this for HLO, MLIR, and jax2tf lowerings. PiperOrigin-RevId: 413425828
260,335
29.10.2021 09:23:27
25,200
659f8b794ff6b64c1264ce6398dfbf4e0a4b19bc
add skeleton checkify transformation
[ { "change_type": "MODIFY", "old_path": "jax/core.py", "new_path": "jax/core.py", "diff": "@@ -1631,9 +1631,9 @@ def call_bind(primitive: Union['CallPrimitive', 'MapPrimitive'],\nout_axes_transforms = _IgnoreElemList()\nif primitive.map_primitive:\nout_axes_thunk = params['out_axes_thunk']\n- # The n...
Python
Apache License 2.0
google/jax
add skeleton checkify transformation
260,335
02.12.2021 14:26:58
28,800
c1f71d17c0b34fbf987147c095843dc6601b41e0
generalize assert primitive, allow recharging
[ { "change_type": "MODIFY", "old_path": "jax/experimental/checkify.py", "new_path": "jax/experimental/checkify.py", "diff": "@@ -88,8 +88,19 @@ def assert_func(error: Error, pred: Bool, msg: str) -> Error:\n## Checkify transformation for plumbing functional error values.\n+class ErrorTracer(core.Trac...
Python
Apache License 2.0
google/jax
generalize assert primitive, allow recharging
260,303
02.12.2021 11:31:42
28,800
6f9c5abd388d9ad333dac67bb2e8e267f239adfa
Add profiler.annotate_function on some internals
[ { "change_type": "MODIFY", "old_path": "jax/_src/device_array.py", "new_path": "jax/_src/device_array.py", "diff": "@@ -25,6 +25,7 @@ from jax import core\nfrom jax._src.config import config\nfrom jax._src import dtypes\nfrom jax._src.lib import xla_client as xc\n+from jax._src.profiler import annot...
Python
Apache License 2.0
google/jax
Add profiler.annotate_function on some internals
260,303
02.12.2021 12:11:12
28,800
643a97c83b424d1903ad6d74e634bb26df34e39a
Add profiler to traceback exclusions
[ { "change_type": "MODIFY", "old_path": "jax/_src/profiler.py", "new_path": "jax/_src/profiler.py", "diff": "@@ -18,6 +18,9 @@ import threading\nfrom typing import Callable, Optional\nimport warnings\n+from jax._src import traceback_util\n+traceback_util.register_exclusion(__file__)\n+\nfrom jax._src...
Python
Apache License 2.0
google/jax
Add profiler to traceback exclusions
260,303
02.12.2021 12:33:18
28,800
6efb1444666d098c00666be6e88310cf4fa73a39
fix pytype errors
[ { "change_type": "MODIFY", "old_path": "jax/_src/device_array.py", "new_path": "jax/_src/device_array.py", "diff": "from functools import partial, partialmethod\nimport operator\n-from typing import (Any, Optional, Union)\n+from typing import (Any, List, Optional, Union)\nimport weakref\nimport nump...
Python
Apache License 2.0
google/jax
fix pytype errors
260,303
03.12.2021 10:37:04
28,800
ac0d02743cd43dfc5767eb48d645957193be97e2
Use profiler.annotate_function
[ { "change_type": "MODIFY", "old_path": "jax/_src/device_array.py", "new_path": "jax/_src/device_array.py", "diff": "@@ -24,8 +24,8 @@ import numpy as np\nfrom jax import core\nfrom jax._src.config import config\nfrom jax._src import dtypes\n+from jax._src import profiler\nfrom jax._src.lib import xl...
Python
Apache License 2.0
google/jax
Use profiler.annotate_function
260,424
06.12.2021 19:36:35
0
bbf1a9ba9788933db25cd8459ff9a3daa91609b4
Add checkify rule for scan.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/checkify.py", "new_path": "jax/experimental/checkify.py", "diff": "@@ -30,7 +30,7 @@ from jax._src import source_info_util, traceback_util\nfrom jax._src.lax import lax\nfrom jax._src.lax import slicing\nfrom jax._src.lax import control_flow\n...
Python
Apache License 2.0
google/jax
Add checkify rule for scan.
260,424
08.12.2021 15:07:43
0
0dc5a33a88cd9f51567ae5e10a2f34172368c166
Add checkify rule for while_loop.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/checkify.py", "new_path": "jax/experimental/checkify.py", "diff": "@@ -198,17 +198,20 @@ def check_errors_subtrace(main, msgs, err, code, *args):\ndel main.error\nyield (err, code, *out_vals), msgs\n-# TODO take (error_aval, code_aval) instead...
Python
Apache License 2.0
google/jax
Add checkify rule for while_loop.
260,496
09.12.2021 14:54:23
-3,600
9bd117f44da7dad5934c356e02f1f8a302aa29ce
Finish jax2tf no_xla documentation
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/converters_eval/no_xla_limitations.md", "new_path": "jax/experimental/jax2tf/converters_eval/no_xla_limitations.md", "diff": "@@ -17,23 +17,26 @@ are supported in [impl_no_xla.py](../impl_no_xla.py).\n## Summary Table\nThe table below s...
Python
Apache License 2.0
google/jax
Finish jax2tf no_xla documentation
260,287
10.12.2021 11:29:41
28,800
1b9a9fff0c0c43409ecf34b3821c4e1df59462bd
Clarify a comment in ParsedPartitionSpec.eq_given_rank Previous one used an overly lax definition of "semantic equality".
[ { "change_type": "MODIFY", "old_path": "jax/experimental/pjit.py", "new_path": "jax/experimental/pjit.py", "diff": "@@ -393,12 +393,22 @@ class ParsedPartitionSpec:\nself.sync == other.sync)\ndef eq_given_rank(self, other, rank):\n- # ParsedPartitionSpecs may contain trailing empty tuples that don't...
Python
Apache License 2.0
google/jax
Clarify a comment in ParsedPartitionSpec.eq_given_rank Previous one used an overly lax definition of "semantic equality". PiperOrigin-RevId: 415567823
260,447
10.12.2021 14:08:21
28,800
8abdd9eceb7bba66de4d3a2554e50b9bbf0b8aec
[sparse] Update bcoo_dot_general GPU translation rule.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/sparse/bcoo.py", "new_path": "jax/experimental/sparse/bcoo.py", "diff": "@@ -25,7 +25,7 @@ from jax import lax\nfrom jax import tree_util\nfrom jax import vmap\nfrom jax.experimental.sparse._base import JAXSparse\n-from jax.experimental.sparse...
Python
Apache License 2.0
google/jax
[sparse] Update bcoo_dot_general GPU translation rule. Co-authored-by: Jake VanderPlas <jakevdp@google.com>
260,411
02.12.2021 08:19:16
-7,200
c9691a5c0f5a2a5b256c4c4c3d1c46520e170670
Refactor the lowering of remat. Replace the XLA-specific lowering with the same logic written with LAX primitives. This will allow reusing the logic for different lowerings, e.g., jax2tf or MLIR. This is a new attempt at which was rolled-back in due to test failures.
[ { "change_type": "MODIFY", "old_path": "jax/_src/ad_checkpoint.py", "new_path": "jax/_src/ad_checkpoint.py", "diff": "@@ -291,6 +291,7 @@ def remat_translation(ctx, avals_in, avals_out, *in_nodes,\nreturn xla._remat_using_cond(ctx, in_nodes, \"checkpoint\", jaxpr)\nelse:\nreturn xla.jaxpr_subcomp(ct...
Python
Apache License 2.0
google/jax
Refactor the lowering of remat. Replace the XLA-specific lowering with the same logic written with LAX primitives. This will allow reusing the logic for different lowerings, e.g., jax2tf or MLIR. This is a new attempt at #8716, which was rolled-back in #8749 due to test failures.
260,643
13.12.2021 15:27:10
0
bd2724683ccf550bb8d31fc4a76554fabd00f6ee
fix: thread fill_value option
[ { "change_type": "MODIFY", "old_path": "jax/_src/numpy/lax_numpy.py", "new_path": "jax/_src/numpy/lax_numpy.py", "diff": "@@ -2875,8 +2875,8 @@ def nonzero(a, *, size=None, fill_value=None):\nreturn out\n@_wraps(np.flatnonzero, lax_description=_NONZERO_DOC)\n-def flatnonzero(a, *, size=None):\n- ret...
Python
Apache License 2.0
google/jax
fix: thread fill_value option
260,573
13.12.2021 16:24:12
0
c508e04ca51bc1734ed3ae9a1fba3bb55cfcba9b
test: add tests for argwhere, flatnonzero
[ { "change_type": "MODIFY", "old_path": "tests/lax_numpy_test.py", "new_path": "tests/lax_numpy_test.py", "diff": "@@ -998,6 +998,29 @@ class LaxBackedNumpyTests(jtu.JaxTestCase):\njnp_fun = lambda x: jnp.flatnonzero(x, size=np.size(x) // 2)\nself._CompileAndCheck(jnp_fun, args_maker)\n+ @parameteriz...
Python
Apache License 2.0
google/jax
test: add tests for argwhere, flatnonzero
260,335
13.12.2021 21:51:08
28,800
2eb788d3f63c1333e1d4085d66fe44349a6b0a1c
log tracing and compilation times
[ { "change_type": "MODIFY", "old_path": "jax/_src/dispatch.py", "new_path": "jax/_src/dispatch.py", "diff": "# Primitive dispatch and jit dispatch.\n+import contextlib\nfrom functools import partial\nimport itertools\n+import time\nfrom typing import (\nAny, Callable, Dict, Optional, Sequence, Set, T...
Python
Apache License 2.0
google/jax
log tracing and compilation times
260,573
14.12.2021 09:29:07
0
9280a392fa28eee8e16552ddaff58bac980a2a76
test: use appropriate args in flatnonzero
[ { "change_type": "MODIFY", "old_path": "tests/lax_numpy_test.py", "new_path": "tests/lax_numpy_test.py", "diff": "@@ -1004,7 +1004,7 @@ class LaxBackedNumpyTests(jtu.JaxTestCase):\n\"shape\": shape, \"dtype\": dtype, \"size\": size, \"fill_value\": fill_value}\nfor shape in nonempty_array_shapes\nfo...
Python
Apache License 2.0
google/jax
test: use appropriate args in flatnonzero
260,573
14.12.2021 09:38:57
0
83d9a5fe31076844b4c53b46bc43a93c0181d9c3
tests: pass list to np.stack, not generator
[ { "change_type": "MODIFY", "old_path": "tests/lax_numpy_test.py", "new_path": "tests/lax_numpy_test.py", "diff": "@@ -1058,8 +1058,8 @@ class LaxBackedNumpyTests(jtu.JaxTestCase):\nreturn result[:size]\nelse:\nfillvals = fill_value if np.ndim(fill_value) else len(result) * [fill_value or 0]\n- retur...
Python
Apache License 2.0
google/jax
tests: pass list to np.stack, not generator
260,573
14.12.2021 10:01:32
0
7980379dffdb548033d3370b9260544021e57175
tests: fix shape inference dimension
[ { "change_type": "MODIFY", "old_path": "tests/lax_numpy_test.py", "new_path": "tests/lax_numpy_test.py", "diff": "@@ -1057,7 +1057,7 @@ class LaxBackedNumpyTests(jtu.JaxTestCase):\nif size <= len(result):\nreturn result[:size]\nelse:\n- fillvals = fill_value if np.ndim(fill_value) else len(result) *...
Python
Apache License 2.0
google/jax
tests: fix shape inference dimension
260,573
14.12.2021 10:01:54
0
cba9852b91e4d7dd4edea35510fced76a428d334
tests: cover 0 dim case
[ { "change_type": "MODIFY", "old_path": "tests/lax_numpy_test.py", "new_path": "tests/lax_numpy_test.py", "diff": "@@ -1058,7 +1058,7 @@ class LaxBackedNumpyTests(jtu.JaxTestCase):\nreturn result[:size]\nelse:\nfillvals = fill_value if np.ndim(fill_value) else result.shape[-1] * [fill_value or 0]\n- ...
Python
Apache License 2.0
google/jax
tests: cover 0 dim case
260,540
14.12.2021 07:16:10
28,800
b28a1c5240fbf7e433f308d4919d511b320df504
Tweak documentation on error handling for host_callback.call()
[ { "change_type": "MODIFY", "old_path": "jax/experimental/host_callback.py", "new_path": "jax/experimental/host_callback.py", "diff": "@@ -393,7 +393,10 @@ in one of the tap functions. This exception will include the text and the\nstack trace of the last exception encountered.\nOne further complicati...
Python
Apache License 2.0
google/jax
Tweak documentation on error handling for host_callback.call()
260,335
14.12.2021 10:42:05
28,800
c555f5f0e44224c65b356b63a9d68983ec14e5fa
handle trivial case for ppermute batching rule fixes
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/parallel.py", "new_path": "jax/_src/lax/parallel.py", "diff": "@@ -793,7 +793,8 @@ def _ppermute_batcher(axis_size, frame_name, _, vals_in, dims_in, axis_name, per\nraise NotImplementedError(\"ppermute batcher only supports a single axis\")\nasser...
Python
Apache License 2.0
google/jax
handle trivial case for ppermute batching rule fixes #8688
260,335
14.12.2021 11:02:14
28,800
c8a34fe5cc8de51bbf6aacd30b38453248440f9e
add jax.block_until_ready function fixes
[ { "change_type": "MODIFY", "old_path": "jax/__init__.py", "new_path": "jax/__init__.py", "diff": "@@ -52,6 +52,7 @@ from jax._src.config import (\n)\nfrom jax._src.api import (\nad, # TODO(phawkins): update users to avoid this.\n+ block_until_ready,\ncheckpoint as checkpoint,\ncheckpoint_policies as...
Python
Apache License 2.0
google/jax
add jax.block_until_ready function fixes #8536
260,335
14.12.2021 12:57:52
28,800
d8e28400ba80d4d636bde449ac4ae296a5de5cae
fix leak checker interaction with custom_jvp fixes
[ { "change_type": "MODIFY", "old_path": "jax/interpreters/partial_eval.py", "new_path": "jax/interpreters/partial_eval.py", "diff": "@@ -1408,8 +1408,9 @@ class DynamicJaxprTrace(core.Trace):\nwith core.new_sublevel():\nfun_jaxpr, out_avals, consts = trace_to_subjaxpr_dynamic(fun, self.main, in_avals...
Python
Apache License 2.0
google/jax
fix leak checker interaction with custom_jvp fixes #8171
260,335
14.12.2021 13:36:46
28,800
0c68605bf1af90662472966657d7a489ff3329a2
add jax.block_until_ready to docs and changelog also unrelatedly fix a couple of the uses of rst in changelog.md (though many others remain)
[ { "change_type": "MODIFY", "old_path": "CHANGELOG.md", "new_path": "CHANGELOG.md", "diff": "@@ -21,6 +21,9 @@ PLEASE REMEMBER TO CHANGE THE '..main' WITH AN ACTUAL TAG in GITHUB LINK.\nAdditionally, added documentation for how to implement the old behavior\nusing JAX custom AD APIs ({jax-issue}`#783...
Python
Apache License 2.0
google/jax
add jax.block_until_ready to docs and changelog also unrelatedly fix a couple of the uses of rst in changelog.md (though many others remain)
260,335
14.12.2021 14:40:13
28,800
984e8d79f0dc45de8d92cbd4a44bf9169ef50ee3
make ravel_pytree unraveler dtype-polymorphic fixes
[ { "change_type": "MODIFY", "old_path": "jax/_src/flatten_util.py", "new_path": "jax/_src/flatten_util.py", "diff": "@@ -57,8 +57,24 @@ def _ravel_list(lst):\nsizes, shapes = unzip2((jnp.size(x), jnp.shape(x)) for x in lst)\nindices = np.cumsum(sizes)\n+ if all(dt == to_dtype for dt in from_dtypes):\...
Python
Apache License 2.0
google/jax
make ravel_pytree unraveler dtype-polymorphic fixes #7809
260,411
24.11.2021 12:58:16
-7,200
3021d3e2e2e3d8d97e9cba620bf233e807ff20b6
[hcb] Add support for remat2 to host_callback A callback under ad_checkpoint.checkpoint will be invoked twice when taking the gradient: once during the forward pass and once again during the backward pass when the residuals for the forward pass are rematerialized.
[ { "change_type": "MODIFY", "old_path": "CHANGELOG.md", "new_path": "CHANGELOG.md", "diff": "@@ -19,7 +19,10 @@ PLEASE REMEMBER TO CHANGE THE '..main' WITH AN ACTUAL TAG in GITHUB LINK.\nobtained (for a limited time) by setting the ``JAX_HOST_CALLBACK_AD_TRANSFORMS``\nenvironment variable, or the ```...
Python
Apache License 2.0
google/jax
[hcb] Add support for remat2 to host_callback A callback under ad_checkpoint.checkpoint will be invoked twice when taking the gradient: once during the forward pass and once again during the backward pass when the residuals for the forward pass are rematerialized.
260,411
16.12.2021 11:30:53
-7,200
06ce9b9b1ac25d2fc51e0bee2645ad6027a74e8a
[jax2tf] Simplify the shape_poly support. The goal is to make is shareable outside of jax2tf.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "@@ -50,7 +50,6 @@ from jax.interpreters import xla\nfrom jax._src.lib import xla_client\nfrom jax.experimental.jax2tf import shape_poly\n-from jax.experimental.jax2tf...
Python
Apache License 2.0
google/jax
[jax2tf] Simplify the shape_poly support. The goal is to make is shareable outside of jax2tf.
260,424
16.12.2021 21:48:37
0
98a54611323653b70d2bd1eed8e4e993c09950ea
Make sure while_loop cond generates an error even if it returns False.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/checkify.py", "new_path": "jax/experimental/checkify.py", "diff": "@@ -305,8 +305,9 @@ def checkify_while_body_jaxpr(cond_jaxpr, body_jaxpr, error):\ncond_f = core.jaxpr_as_fun(cond_jaxpr)\nbody_f = core.jaxpr_as_fun(body_jaxpr)\ndef new_body_...
Python
Apache License 2.0
google/jax
Make sure while_loop cond generates an error even if it returns False.
260,424
16.12.2021 22:44:05
0
03e0deac04f7379f52f93bf879a1d58380182e84
Add NaN checkify rule to all lax primitives.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/checkify.py", "new_path": "jax/experimental/checkify.py", "diff": "@@ -27,9 +27,7 @@ from jax.api_util import flatten_fun\nfrom jax.interpreters import partial_eval as pe\nfrom jax.tree_util import tree_flatten, tree_unflatten, register_pytree...
Python
Apache License 2.0
google/jax
Add NaN checkify rule to all lax primitives.
260,461
21.12.2021 20:21:39
0
cd333f0f5b97d1319b5cd3523faccfa0aa256202
Fix straight-through estimator example in docs
[ { "change_type": "MODIFY", "old_path": "docs/jax-101/04-advanced-autodiff.ipynb", "new_path": "docs/jax-101/04-advanced-autodiff.ipynb", "diff": "\" return jnp.round(x) # non-differentiable\\n\",\n\"\\n\",\n\"def straight_through_f(x):\\n\",\n- \" return x + jax.lax.stop_gradient(f(x) - x)\\n\",\n+ ...
Python
Apache License 2.0
google/jax
Fix straight-through estimator example in docs (#9032)
260,447
21.12.2021 14:56:09
28,800
6a0bfdd637a79edda35356621867dd758705062d
[JAX] Requires indices to be sorted and of int32 in `_sparse_bcoo_matvec` test.
[ { "change_type": "MODIFY", "old_path": "benchmarks/api_benchmark.py", "new_path": "benchmarks/api_benchmark.py", "diff": "@@ -404,13 +404,10 @@ def sparse_bcoo_todense_compile(state):\ndef _sparse_bcoo_matvec(state, jit: bool = False, compile: bool = False):\nshape = (2000, 2000)\nnse = 10000\n- siz...
Python
Apache License 2.0
google/jax
[JAX] Requires indices to be sorted and of int32 in `_sparse_bcoo_matvec` test. PiperOrigin-RevId: 417695937
260,335
14.12.2021 13:29:16
28,800
a582fa874807cdc9fba1dbe07d830ac7632b9abb
add limit to number of tracer provenance lines fixes
[ { "change_type": "MODIFY", "old_path": "jax/interpreters/partial_eval.py", "new_path": "jax/interpreters/partial_eval.py", "diff": "@@ -1159,11 +1159,13 @@ class DynamicJaxprTracer(core.Tracer):\nmsts = [\" operation \"\nf\"{core.pp_eqn(eqn, core.JaxprPpContext(), print_shapes=True)}\\n\"\nf\" from ...
Python
Apache License 2.0
google/jax
add limit to number of tracer provenance lines fixes #8910
260,335
16.12.2021 10:57:19
28,800
9e28bd5f4efc59de4c0a93a0b85b96c47d10799d
small changes to checkify
[ { "change_type": "MODIFY", "old_path": "jax/experimental/checkify.py", "new_path": "jax/experimental/checkify.py", "diff": "@@ -222,9 +222,16 @@ def check_errors_traceable(msgs, err, code, *args):\n## assert primitive\ndef assert_(pred: Bool, msg: str) -> None:\n+ if not is_scalar_pred(pred):\n+ rai...
Python
Apache License 2.0
google/jax
small changes to checkify Co-authored-by: Lena Martens <lenamartens@google.com>
260,377
27.12.2021 20:23:29
-10,800
f93531b0204914b04217674ef31dd1a29088fd26
replace deprecated jax.ops.index_* functions with the new index update operators
[ { "change_type": "MODIFY", "old_path": "docs/jax-101/01-jax-basics.ipynb", "new_path": "docs/jax-101/01-jax-basics.ipynb", "diff": "\"\\u001b[0;31mTypeError\\u001b[0m Traceback (most recent call last)\",\n\"\\u001b[0;32m<ipython-input-12-709e2d7ddd3f>\\u001b[0m in \\u001b[0;36m<module>\\u001b[0;34m(...
Python
Apache License 2.0
google/jax
replace deprecated jax.ops.index_* functions with the new index update operators
260,377
29.12.2021 12:29:16
-10,800
504728d8b65d3219cb26fc6400f775afac53361b
link directly to the documentation for the jnp.ndarray.at property
[ { "change_type": "MODIFY", "old_path": "docs/jax-101/01-jax-basics.ipynb", "new_path": "docs/jax-101/01-jax-basics.ipynb", "diff": "\"\\u001b[0;32m<ipython-input-12-709e2d7ddd3f>\\u001b[0m in \\u001b[0;36m<module>\\u001b[0;34m()\\u001b[0m\\n\\u001b[0;32m----> 1\\u001b[0;31m \\u001b[0min_place_modify...
Python
Apache License 2.0
google/jax
link directly to the documentation for the jnp.ndarray.at property
260,652
29.10.2021 00:33:40
-19,080
03adf8cc8922016d37e6187c1d33203e7597afe9
support null shape in jax.random.poisson
[ { "change_type": "MODIFY", "old_path": "jax/_src/random.py", "new_path": "jax/_src/random.py", "diff": "@@ -1109,20 +1109,21 @@ def _poisson(key, lam, shape, dtype):\ndef poisson(key: KeyArray,\nlam: RealArray,\n- shape: Sequence[int] = (),\n+ shape: Optional[Sequence[int]] = None,\ndtype: DTypeLike...
Python
Apache License 2.0
google/jax
support null shape in jax.random.poisson
260,695
21.12.2021 13:59:30
-3,600
b810e8be88c7bb4459e65ad271e6ca49a1c0b917
Add `where=` arg to jax.nn.{softmax, log_softmax, normalize}. This change adds a `where=` argument (analogous to `jnp.sum`) that can be used to specify which elements to include in the calculation.
[ { "change_type": "MODIFY", "old_path": "jax/_src/nn/functions.py", "new_path": "jax/_src/nn/functions.py", "diff": "@@ -261,7 +261,10 @@ def glu(x: Array, axis: int = -1) -> Array:\nlogsumexp = _logsumexp\n-def log_softmax(x: Array, axis: Optional[Union[int, Tuple[int, ...]]] = -1) -> Array:\n+def l...
Python
Apache License 2.0
google/jax
Add `where=` arg to jax.nn.{softmax, log_softmax, normalize}. This change adds a `where=` argument (analogous to `jnp.sum`) that can be used to specify which elements to include in the calculation.
260,424
20.12.2021 15:56:50
0
bbd127f8fa128bf686d858a6639d6527289891af
Add division by zero check.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/checkify.py", "new_path": "jax/experimental/checkify.py", "diff": "@@ -279,6 +279,14 @@ def gather_error_check(error, operand, start_indices, *,\nreturn out, assert_func(error, all_inbounds, msg)\nerror_checks[lax.gather_p] = gather_error_chec...
Python
Apache License 2.0
google/jax
Add division by zero check.
260,689
19.11.2021 21:21:46
-3,600
2c5fe8c40d8b530701c1088ea3446b07d844ce46
Implement SciPy's RegularGridInterpolator Resolves .
[ { "change_type": "ADD", "old_path": null, "new_path": "jax/_src/scipy/interpolate/__init__.py", "diff": "+# Copyright 2022 Google LLC\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may obtain a co...
Python
Apache License 2.0
google/jax
Implement SciPy's RegularGridInterpolator Resolves #8572 .
260,335
10.12.2021 23:22:11
28,800
4db899007b86f51da18940be35bf77575652e11f
add staging logic for polymorphic shapes in jaxprs
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/lax.py", "new_path": "jax/_src/lax/lax.py", "diff": "@@ -105,7 +105,7 @@ def _try_broadcast_shapes(shapes):\nif not rank: return () # scalar case\nresult_shape = [None] * rank\nfor i, sizes in enumerate(zip(*shapes)):\n- non_1s = set([d for d in s...
Python
Apache License 2.0
google/jax
add staging logic for polymorphic shapes in jaxprs Co-authored-by: Dougal Maclaurin <dougalm@google.com>
260,335
05.01.2022 13:25:52
28,800
6ce38acca85791bb6f290cb65e1fdc1170e873e4
remove axis name logic from Primitive / bind Instead, just give AxisPrimitive its own bind function. This way the logic is nicely separated by concerns. In addition, this factorization will let us more easily experiment with other ways to find the top trace (e.g. for assert_p in checkify).
[ { "change_type": "MODIFY", "old_path": "jax/core.py", "new_path": "jax/core.py", "diff": "@@ -260,7 +260,6 @@ class Primitive:\nmultiple_results = False # set for multi-output primitives\ncall_primitive = False # set for call primitives processed in final style\nmap_primitive = False # set for map p...
Python
Apache License 2.0
google/jax
remove axis name logic from Primitive / bind Instead, just give AxisPrimitive its own bind function. This way the logic is nicely separated by concerns. In addition, this factorization will let us more easily experiment with other ways to find the top trace (e.g. for assert_p in checkify).
260,315
07.01.2022 12:06:23
-3,600
6bf7b0d325c79f2cfede938b7ed1bbf9d0099199
Fix host_callback docs There was a missing ':' causing invalid rendering of the docs.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/host_callback.py", "new_path": "jax/experimental/host_callback.py", "diff": "@@ -229,7 +229,7 @@ Behavior under JAX autodiff transformations\n-------------------------------------------\nWhen used under a JAX autodiff transformation, the host ...
Python
Apache License 2.0
google/jax
Fix host_callback docs There was a missing ':' causing invalid rendering of the docs.
260,382
01.12.2021 15:29:49
18,000
38e98d2d7bdd14419a1d668a928f6743c30d3470
Fix a bug that promoted t to a complex in odeint, and modify a test so it would have caught it In odeint, raise error if t is not an array of floats
[ { "change_type": "MODIFY", "old_path": "jax/experimental/ode.py", "new_path": "jax/experimental/ode.py", "diff": "@@ -92,7 +92,7 @@ def initial_step_size(fun, t0, y0, order, rtol, atol, f0):\ndef runge_kutta_step(func, y0, f0, t0, dt):\n# Dopri5 Butcher tableaux\n- alpha = jnp.array([1 / 5, 3 / 10, ...
Python
Apache License 2.0
google/jax
Fix a bug that promoted t to a complex in odeint, and modify a test so it would have caught it In odeint, raise error if t is not an array of floats
260,424
23.12.2021 15:23:58
0
7b5b9cefbd928e1f8b8e8ccf50416c0609f78e39
Add scatter OOB error.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/checkify.py", "new_path": "jax/experimental/checkify.py", "diff": "@@ -287,6 +287,52 @@ def div_error_check(error, x, y):\nreturn nan_error_check(lax.div_p, div_by_zero_err, x, y)\nerror_checks[lax.div_p] = div_error_check\n+def scatter_in_bou...
Python
Apache License 2.0
google/jax
Add scatter OOB error.
260,296
07.01.2022 14:54:41
28,800
8977998b5c8cd8db240e36d87dd8dfd6192d0797
Update type annotations and use the new `convolution.py` file
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/other.py", "new_path": "jax/_src/lax/other.py", "diff": "@@ -122,10 +122,10 @@ def conv_general_dilated_local(\nwindow_strides: Sequence[int],\npadding: Union[str, Sequence[Tuple[int, int]]],\nfilter_shape: Sequence[int],\n- lhs_dilation: Sequence...
Python
Apache License 2.0
google/jax
Update type annotations and use the new `convolution.py` file
260,335
08.01.2022 12:57:28
28,800
15e6098beb5c12f4fe9d8ccc5f3341a099086cb7
add JAX_BACKEND_TARGET env var, fixes
[ { "change_type": "MODIFY", "old_path": "jax/_src/lib/xla_bridge.py", "new_path": "jax/_src/lib/xla_bridge.py", "diff": "@@ -57,7 +57,8 @@ flags.DEFINE_string(\n'jax_xla_backend', '',\n'Deprecated, please use --jax_platforms instead.')\nflags.DEFINE_string(\n- 'jax_backend_target', '',\n+ 'jax_backen...
Python
Apache License 2.0
google/jax
add JAX_BACKEND_TARGET env var, fixes #9034
260,442
08.01.2022 23:36:50
28,800
4a17c78605e7fc69a69a999e2f6298db79d3837a
[mhlo] Remove tuples from mhlo If/Case op.
[ { "change_type": "MODIFY", "old_path": "jax/_src/lax/control_flow.py", "new_path": "jax/_src/lax/control_flow.py", "diff": "@@ -1405,6 +1405,8 @@ xla.register_translation(cond_p, _cond_translation_rule, initial_style=True)\ncore.custom_typechecks[cond_p] = _cond_typecheck\npe.partial_eval_jaxpr_cust...
Python
Apache License 2.0
google/jax
[mhlo] Remove tuples from mhlo If/Case op. PiperOrigin-RevId: 420554920
260,574
10.01.2022 16:19:57
-10,800
78977d6f5ab33012fa7171bb744190efa4a2e9c8
fix broken links and update texts in thinking_in_jax.ipynb
[ { "change_type": "MODIFY", "old_path": "docs/notebooks/thinking_in_jax.ipynb", "new_path": "docs/notebooks/thinking_in_jax.ipynb", "diff": "\"id\": \"yRYF0YgO3F4H\"\n},\n\"source\": [\n- \"For updating individual elements, JAX provides an [indexed update syntax](https://jax.readthedocs.io/en/latest/...
Python
Apache License 2.0
google/jax
fix broken links and update texts in thinking_in_jax.ipynb
260,574
10.01.2022 17:34:09
-10,800
d2c6c06546c7e68cc0e34bc594e9d69bf58db9a4
Fix DeviceArray class reference
[ { "change_type": "MODIFY", "old_path": "docs/glossary.rst", "new_path": "docs/glossary.rst", "diff": "@@ -13,7 +13,7 @@ JAX Glossary of Terms\nby JAX to perform computations.\nDeviceArray\n- JAX's analog of the :class:`numpy.ndarray`. See :class:`jax.interpreters.xla.DeviceArray`.\n+ JAX's analog of...
Python
Apache License 2.0
google/jax
Fix DeviceArray class reference
260,335
10.01.2022 20:57:56
28,800
1cf7d4ab5dcea4619a2b6c1b471f2be340b6abfc
Copybara import of the project: by Matthew Johnson add jax.ensure_compile_time_eval to public api aka jax.core.eval_context COPYBARA_INTEGRATE_REVIEW=https://github.com/google/jax/pull/7987 from google:issue7535
[ { "change_type": "MODIFY", "old_path": "CHANGELOG.md", "new_path": "CHANGELOG.md", "diff": "@@ -32,6 +32,7 @@ PLEASE REMEMBER TO CHANGE THE '..main' WITH AN ACTUAL TAG in GITHUB LINK.\n* Added a new debugging flag/environment variable `JAX_DUMP_IR_TO=/path`.\nIf set, JAX dumps the MHLO/HLO IR it gen...
Python
Apache License 2.0
google/jax
Copybara import of the project: -- 4fcdadbfb3f4c484fd4432203cf13b88782b9311 by Matthew Johnson <mattjj@google.com>: add jax.ensure_compile_time_eval to public api aka jax.core.eval_context COPYBARA_INTEGRATE_REVIEW=https://github.com/google/jax/pull/7987 from google:issue7535 4fcdadbfb3f4c484fd4432203cf13b88782b9311 PiperOrigin-RevId: 420928687
260,335
10.01.2022 21:29:12
28,800
6850833c3a6c94060aaf48a4c26c2cc39cba8ba9
checkify: tweak some organization and names
[ { "change_type": "MODIFY", "old_path": "jax/experimental/checkify.py", "new_path": "jax/experimental/checkify.py", "diff": "from dataclasses import dataclass\nfrom functools import partial\nimport itertools as it\n-from typing import Union, Optional, Callable, Dict\n+from typing import Union, Option...
Python
Apache License 2.0
google/jax
checkify: tweak some organization and names