author
int64
658
755k
date
stringdate
2012-06-12 08:34:29
2024-07-22 14:51:21
timezone
int64
-46,800
43.2k
hash
stringlengths
40
40
message
stringlengths
5
490
mods
listlengths
1
16
language
stringclasses
20 values
license
stringclasses
3 values
repo
stringlengths
5
68
original_message
stringlengths
12
491
260,411
31.08.2020 10:26:32
-10,800
b6b1f5e349a9e264324321f80bd3c16bf2ef71b7
[jax2tf] Turn on with_gradient by default As I was writing the demo I realized that it makes more sense for with_gradient to be set to True by default. I have also fixed a bug with tie_in in omnistaging.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "@@ -134,7 +134,7 @@ def _tfval_add_unit(vals: Sequence[TfValOrUnit],\ntf_impl: Dict[core.Primitive,\nCallable[..., Any]] = {}\n-def convert(fun, with_gradient=False):...
Python
Apache License 2.0
google/jax
[jax2tf] Turn on with_gradient by default (#4180) As I was writing the demo I realized that it makes more sense for with_gradient to be set to True by default. I have also fixed a bug with tie_in in omnistaging.
260,417
31.08.2020 17:00:34
-10,800
44bcf7e776ad2dd0636db534905708293ea21ad6
Fix axis checking and remove extra print statement A series of PRs renaming the frame entries have been submitted, one of them introducing a bug when using omnistaging. This PR fixes that and removes a print comment (assuming added for debugging purposes).
[ { "change_type": "MODIFY", "old_path": "jax/interpreters/batching.py", "new_path": "jax/interpreters/batching.py", "diff": "@@ -142,7 +142,7 @@ class BatchTrace(Trace):\naxis_names = (axis_names,)\nfor i, axis_name in enumerate(axis_names):\nframe = core.axis_frame(axis_name)\n- if frame.tag is not ...
Python
Apache License 2.0
google/jax
Fix axis checking and remove extra print statement (#4184) A series of PRs renaming the frame entries have been submitted, one of them introducing a bug when using omnistaging. This PR fixes that and removes a print comment (assuming added for debugging purposes).
260,298
01.09.2020 09:35:25
-7,200
0cdb1f7ee603b9ac1764192355e6111a939c6d8b
[jax2tf] Indicate the version of TF used in tests in README.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/README.md", "new_path": "jax/experimental/jax2tf/README.md", "diff": "@@ -8,6 +8,8 @@ primitives using TensorFlow operations. In practice this means that you can take\nsome code written in JAX and execute it using TensorFlow eager mode,...
Python
Apache License 2.0
google/jax
[jax2tf] Indicate the version of TF used in tests in README. (#4185)
260,335
01.09.2020 18:16:20
25,200
04f9a7e53db3eb0eb6d5d08a5f2a967bf9712048
better jax.numpy.tile implementation Use reshape, broadcast_to, reshape.
[ { "change_type": "MODIFY", "old_path": "jax/numpy/lax_numpy.py", "new_path": "jax/numpy/lax_numpy.py", "diff": "@@ -2148,14 +2148,11 @@ def stack(arrays, axis=0):\ndef tile(A, reps):\nif isinstance(reps, int):\nreps = (reps,)\n- A = reshape(A, (1,) * (len(reps) - ndim(A)) + shape(A))\n- reps = (1,) ...
Python
Apache License 2.0
google/jax
better jax.numpy.tile implementation (#4190) Use reshape, broadcast_to, reshape.
260,475
03.09.2020 00:13:17
-3,600
708d07d5ff0f7910865ec179b1cef76f873ce3f0
Add jax.numpy.array_split
[ { "change_type": "MODIFY", "old_path": "docs/jax.numpy.rst", "new_path": "docs/jax.numpy.rst", "diff": "@@ -67,6 +67,7 @@ Not every function in NumPy is implemented; contributions are welcome!\narray\narray_equal\narray_repr\n+ array_split\narray_str\nasarray\natleast_1d\n" }, { "change_type...
Python
Apache License 2.0
google/jax
Add jax.numpy.array_split (#4197)
260,411
03.09.2020 14:18:35
-10,800
5eac47726b4e02db4b909b56f97d8e1f767a14b4
[jax2tf] Implementation of random_gamma * [jax2tf] implementation of random_gamma The simplest implementation is by converting the JAX own impl_rule, which rewrites gamma into other JAX primitives. On TPU with use_vmap=True the performance is the same for JAX and TF, provided we use tf.function(compile=True).
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "@@ -134,7 +134,7 @@ def _tfval_add_unit(vals: Sequence[TfValOrUnit],\ntf_impl: Dict[core.Primitive,\nCallable[..., Any]] = {}\n-def convert(fun, with_gradient=True):\...
Python
Apache License 2.0
google/jax
[jax2tf] Implementation of random_gamma (#4192) * [jax2tf] implementation of random_gamma The simplest implementation is by converting the JAX own impl_rule, which rewrites gamma into other JAX primitives. On TPU with use_vmap=True the performance is the same for JAX and TF, provided we use tf.function(compile=True).
260,411
03.09.2020 14:24:04
-10,800
abdd13884b066a54413af1627f8e490cd4642306
[jax2tf] Flip the with_gradient=True; was flipped back by mistake
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "@@ -134,7 +134,7 @@ def _tfval_add_unit(vals: Sequence[TfValOrUnit],\ntf_impl: Dict[core.Primitive,\nCallable[..., Any]] = {}\n-def convert(fun, with_gradient=False):...
Python
Apache License 2.0
google/jax
[jax2tf] Flip the with_gradient=True; was flipped back by mistake (#4200)
260,298
03.09.2020 15:56:22
-7,200
bcf9777baca50d22f99b8adc45c472aecceaf33a
[jax2tf] Generator for the documentation of operations with limited support (WIP) * [jax2tf] Draft of a generator for the documentation of operations with limited support.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "import functools\nimport string\n-from typing import Any, Callable, Dict, Iterable, Sequence, Tuple, Union\n+from typing import Any, Callable, Dict, Iterable, List, N...
Python
Apache License 2.0
google/jax
[jax2tf] Generator for the documentation of operations with limited support (WIP) (#4193) * [jax2tf] Draft of a generator for the documentation of operations with limited support.
260,680
04.09.2020 19:21:43
-10,800
96278e67a23c3c43afa790c4ec58c7923e0003b7
Add reverse flag in associative scan Add optional 'reverse' argument in associative scan
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax_control_flow.py", "new_path": "jax/lax/lax_control_flow.py", "diff": "@@ -2324,7 +2324,7 @@ def _interleave(a, b):\nreturn jnp.reshape(jnp.stack([a, b], axis=1),\n(2 * half_num_elems,) + a.shape[1:])\n-def associative_scan(fn, elems):\n+def associa...
Python
Apache License 2.0
google/jax
Add reverse flag in associative scan (#4181) Add optional 'reverse' argument in associative scan
260,411
07.09.2020 14:41:50
-10,800
1e84cbe9cc68ec893e789afa00948971f5b01ff0
[jax2tf] Fix random.split when jax_exable_x64 Since we do the threefry with signed integers when converting to TF, we run into the type promotion 'uint32 - int32 = int64', which then results in lax.shift_right_logical(uint32, int64), which fails.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/tests/primitives_test.py", "new_path": "jax/experimental/jax2tf/tests/primitives_test.py", "diff": "@@ -617,8 +617,7 @@ class JaxPrimitiveTest(tf_test_util.JaxToTfTestCase):\nself.ConvertAndCompare(harness.dyn_fun, *harness.dyn_args_mak...
Python
Apache License 2.0
google/jax
[jax2tf] Fix random.split when jax_exable_x64 (#4208) Since we do the threefry with signed integers when converting to TF, we run into the type promotion 'uint32 - int32 = int64', which then results in lax.shift_right_logical(uint32, int64), which fails.
260,411
07.09.2020 17:13:11
-10,800
4413bb8a4f0bd8fb5ffd6ace5ca09f63fe3eac28
[jax2tf] Do not use jax.random.PRNGKey before in primitive harness We cannot execute JAX functions before the program is initialized
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/tests/primitive_harness.py", "new_path": "jax/experimental/jax2tf/tests/primitive_harness.py", "diff": "@@ -848,7 +848,7 @@ lax_reduce_window = tuple(\nrandom_gamma = tuple(\nHarness(f\"_shape={jtu.format_shape_dtype_string(shape, dtype...
Python
Apache License 2.0
google/jax
[jax2tf] Do not use jax.random.PRNGKey before in primitive harness (#4211) We cannot execute JAX functions before the program is initialized
260,287
07.09.2020 12:31:58
0
0aed1f4ddf1863f151cb7e4a2c5473df8dddba50
Add more context to the axis_frame error message. Some of the vmap and gmap collective tests have been failing on master and I can't seem to be able to reproduce them locally. Hopefully, if this happens again, this extra bit of information will be useful in debugging the problem.
[ { "change_type": "MODIFY", "old_path": "jax/core.py", "new_path": "jax/core.py", "diff": "@@ -1588,4 +1588,5 @@ def omnistaging_enabler() -> None:\nif frame.name == axis_name:\nreturn frame\nelse:\n- raise NameError(\"unbound axis name: {}\".format(axis_name))\n+ raise NameError(f\"Unbound axis name...
Python
Apache License 2.0
google/jax
Add more context to the axis_frame error message. Some of the vmap and gmap collective tests have been failing on master and I can't seem to be able to reproduce them locally. Hopefully, if this happens again, this extra bit of information will be useful in debugging the problem.
260,298
07.09.2020 17:12:35
-7,200
e1340f3495619d260c31c41e45418e41165937f0
[jax2tf] Fix missing complex64 TPU corner case of scatter_{add,mul}
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/primitives_with_limited_support.md", "new_path": "jax/experimental/jax2tf/primitives_with_limited_support.md", "diff": "| reduce_window_sum | Missing TF support | reduce_window_sum is unimplemented for dtype uint32 | CPU, GPU, TPU |\n| ...
Python
Apache License 2.0
google/jax
[jax2tf] Fix missing complex64 TPU corner case of scatter_{add,mul} (#4213)
260,298
08.09.2020 10:32:53
-7,200
798a2648f55fc35bf58e2022cd38a1cf03229309
[jax2tf] Fix bug in population count and move expect_tf_exception into correctness stats. The code was using `tf.bitcast` instead of `tf.cast`, but using `expect_tf_exception` in every case was hiding the errors.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "@@ -454,7 +454,7 @@ tf_impl[lax.nextafter_p] = tf.math.nextafter\ndef _population_count(x):\norig_dtype = x.dtype\n- return tf.bitcast(tf.raw_ops.PopulationCount(x=x)...
Python
Apache License 2.0
google/jax
[jax2tf] Fix bug in population count and move expect_tf_exception (#4214) into correctness stats. The code was using `tf.bitcast` instead of `tf.cast`, but using `expect_tf_exception` in every case was hiding the errors.
260,335
08.09.2020 08:27:41
25,200
ed0d8c02f6860139347f5d3347b9e2f33f67977e
tweak lax.py shape broadcasting logic This new implementation is faster, and works for polymorphic shapes without weird tricks. (This new implementation is faster even if we remove the weird tricks for polymorphism.)
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax.py", "new_path": "jax/lax/lax.py", "diff": "@@ -60,12 +60,13 @@ DType = Any\nShape = Sequence[int]\ndef _try_broadcast_shapes(shapes):\n- # Replace 1 with 0 to avoid inconclusive comparisons for polymorphic dims:\n- out_shape = np.max(np.where(shap...
Python
Apache License 2.0
google/jax
tweak lax.py shape broadcasting logic (#4217) This new implementation is faster, and works for polymorphic shapes without weird tricks. (This new implementation is faster even if we remove the weird tricks for polymorphism.)
260,335
08.09.2020 08:54:13
25,200
7f3078b70d0ed9bea6228efa420879c56f72ef69
updtate version and changelog for pypi
[ { "change_type": "MODIFY", "old_path": "docs/CHANGELOG.rst", "new_path": "docs/CHANGELOG.rst", "diff": "@@ -9,6 +9,10 @@ Change Log\nThese are the release notes for JAX.\n+jax 0.1.76 (September 8, 2020)\n+--------------------------\n+* `GitHub commits <https://github.com/google/jax/compare/jax-v0.1....
Python
Apache License 2.0
google/jax
updtate version and changelog for pypi (#4224)
260,335
08.09.2020 21:14:25
25,200
745d90d03662453e64c2177134f4c0975e4d0987
improve lax.pad shape rule It's now: * better tested * better at catching errors * faster * easier to read
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax.py", "new_path": "jax/lax/lax.py", "diff": "@@ -3206,11 +3206,16 @@ def _pad_dtype_rule(operand, padding_value, *, padding_config):\nreturn _input_dtype(operand, padding_value)\ndef _pad_shape_rule(operand, padding_value, *, padding_config):\n- lo,...
Python
Apache License 2.0
google/jax
improve lax.pad shape rule (#4234) It's now: * better tested * better at catching errors * faster * easier to read
260,411
09.09.2020 11:34:22
-10,800
ee38e7166c6ac2f7d5f5cb39d39cdc672e8224d4
[jax2tf] Clean up code for XlaGather, experimental_compile not necessary * [jax2tf] Clean up code for XlaGather, experimental_compile not necessary Now that XlaGather has been fixed in XLA, we do not need to use experimental_compile workaround (which was not working anyway when put in a SavedModel). This fix requires a recent tf-nightly installation.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "# See the License for the specific language governing permissions and\n# limitations under the License.\n\"\"\"Experimental module transforms JAX functions to be exec...
Python
Apache License 2.0
google/jax
[jax2tf] Clean up code for XlaGather, experimental_compile not necessary (#4030) * [jax2tf] Clean up code for XlaGather, experimental_compile not necessary Now that XlaGather has been fixed in XLA, we do not need to use experimental_compile workaround (which was not working anyway when put in a SavedModel). This fix requires a recent tf-nightly installation.
260,298
09.09.2020 12:20:59
-7,200
f908f6f25c82bb2dcc8b54104083af43821e4d21
[jax2tf] Updated test_pad to test all dtypes and remove old skipped test.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/tests/primitive_harness.py", "new_path": "jax/experimental/jax2tf/tests/primitive_harness.py", "diff": "@@ -440,7 +440,7 @@ lax_pad = tuple(\nrng_factory=jtu.rand_small,\narg_shape=arg_shape, dtype=dtype, pads=pads)\nfor arg_shape in [(...
Python
Apache License 2.0
google/jax
[jax2tf] Updated test_pad to test all dtypes and remove old (#4235) skipped test.
260,380
09.09.2020 13:02:45
-3,600
bff24bddbbb3822b94eb63b804ac844a26a7386b
Add axis_index_groups support to all_gather.
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax_parallel.py", "new_path": "jax/lax/lax_parallel.py", "diff": "@@ -579,17 +579,17 @@ ad.deflinear(all_to_all_p, _all_to_all_transpose_rule)\npxla.multi_host_supported_collectives.add(all_to_all_p)\n-def _expand(dim, size, axis_name, x):\n+def _expan...
Python
Apache License 2.0
google/jax
Add axis_index_groups support to all_gather. (#4194)
260,298
09.09.2020 14:43:36
-7,200
053cd5aa391905772428e2a5b19af651f6e2d636
[jax2tf] Clean up test_dynamic_slice. * [jax2tf] Clean up test_dynamic_slice. With the XLA nested compilation bug fixed, this should now work fine.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/tests/primitives_test.py", "new_path": "jax/experimental/jax2tf/tests/primitives_test.py", "diff": "@@ -428,7 +428,6 @@ class JaxPrimitiveTest(tf_test_util.JaxToTfTestCase):\ndef test_dynamic_slice(self, harness):\n# JAX.dynamic_slice r...
Python
Apache License 2.0
google/jax
[jax2tf] Clean up test_dynamic_slice. (#4236) * [jax2tf] Clean up test_dynamic_slice. With the XLA nested compilation bug fixed, this should now work fine.
260,298
09.09.2020 16:48:00
-7,200
70891f46cdf42d0c9d3c6c5d2527a56a9500774e
[jax2tf] Add a template file for documentation generation. * [jax2tf] Add a template file for documentation generation. The documentation now gives instructions about how to regenerate it, as well as when it was last generated. * Added a list of conversions that are not yet implemented.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/primitives_with_limited_support.md", "new_path": "jax/experimental/jax2tf/primitives_with_limited_support.md", "diff": "# Primitives with limited support\n+*Last generated on (YYYY-MM-DD): 2020-09-09*\n+\n+## Updating the documentation\...
Python
Apache License 2.0
google/jax
[jax2tf] Add a template file for documentation generation. (#4219) * [jax2tf] Add a template file for documentation generation. The documentation now gives instructions about how to regenerate it, as well as when it was last generated. * Added a list of conversions that are not yet implemented.
260,411
10.09.2020 12:19:22
-10,800
f0a3fd4a8687baf348ca70fe168e43389bf7506e
[jax2tf] Moved the limitations for XlaSort to correctness_stats
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/primitives_with_limited_support.md", "new_path": "jax/experimental/jax2tf/primitives_with_limited_support.md", "diff": "@@ -42,6 +42,11 @@ certain data types.\n| scatter-add | Missing TF support | scatter-add is unimplemented for dtype ...
Python
Apache License 2.0
google/jax
[jax2tf] Moved the limitations for XlaSort to correctness_stats (#4237)
260,411
10.09.2020 13:55:57
-10,800
0962ceb05722beb26ee42100384095551d0bda9a
[jax2tf] Fix test failure on TPUs
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/primitives_with_limited_support.md", "new_path": "jax/experimental/jax2tf/primitives_with_limited_support.md", "diff": "This file is automatically generated by running the jax2tf tests in\n`jax/experimental/jax2tf/tests/primitives_test....
Python
Apache License 2.0
google/jax
[jax2tf] Fix test failure on TPUs (#4247)
260,423
11.09.2020 18:51:42
-3,600
ca1d8f41090e2befba903c4b20a1121f39ecf833
Fixing weird behavior in segment_sum when num_segments is None
[ { "change_type": "MODIFY", "old_path": "jax/ops/scatter.py", "new_path": "jax/ops/scatter.py", "diff": "@@ -316,8 +316,10 @@ def segment_sum(data, segment_ids, num_segments=None,\nneed not be sorted. Values outside of the range [0, num_segments) are\nwrapped into that range by applying jnp.mod.\nnum...
Python
Apache License 2.0
google/jax
Fixing weird behavior in segment_sum when num_segments is None (#4034) Co-authored-by: alvarosg <alvarosg@google.com>
260,335
11.09.2020 22:40:12
25,200
f039f6daf96f2e0abdd27f13142a97dc1dea2229
thread backend in pxla.replicate * thread backend in pxla.replicate fixes * add test for
[ { "change_type": "MODIFY", "old_path": "jax/interpreters/pxla.py", "new_path": "jax/interpreters/pxla.py", "diff": "@@ -925,7 +925,7 @@ def replicate(val, axis_size, nrep, devices=None, backend=None):\nA ShardedDeviceArray of length `axis_size` where each shard is equal to\n``val``.\n\"\"\"\n- devic...
Python
Apache License 2.0
google/jax
thread backend in pxla.replicate (#4272) * thread backend in pxla.replicate fixes #4223 * add test for #4223
260,639
12.09.2020 16:10:01
14,400
64bead2093b705820be0dcb46e27edcfd8b50d34
fixing typo I assume "...one of more type parameters..." was intended to read "...one or more type parameters..."
[ { "change_type": "MODIFY", "old_path": "docs/jaxpr.rst", "new_path": "docs/jaxpr.rst", "diff": "@@ -21,7 +21,7 @@ will apply transformations incrementally during tracing.\nNevertheless, if one wants to understand how JAX works internally, or to\nmake use of the result of JAX tracing, it is useful to...
Python
Apache License 2.0
google/jax
fixing typo (#4273) I assume "...one of more type parameters..." was intended to read "...one or more type parameters..."
260,296
13.09.2020 19:20:31
25,200
38b43ef95d0c445168ce416f0e7997b8b838c06e
Avoid rank promotion in np.outer
[ { "change_type": "MODIFY", "old_path": "jax/numpy/lax_numpy.py", "new_path": "jax/numpy/lax_numpy.py", "diff": "@@ -3198,7 +3198,7 @@ def outer(a, b, out=None):\nif out:\nraise NotImplementedError(\"The 'out' argument to outer is not supported.\")\na, b = _promote_dtypes(a, b)\n- return ravel(a)[:, ...
Python
Apache License 2.0
google/jax
Avoid rank promotion in np.outer (#4276)
260,298
14.09.2020 11:34:31
-7,200
2ff3479239bc172ccdd764ae579dae60e96d079e
[jax2tf] Fix tests when running with JAX_ENABLE_X64=1. Fixed tests: test_binary_elementwise dynamic_update_slice fft population_count test_unary_elementwise top_k select_and_gather_add
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/primitives_with_limited_support.md", "new_path": "jax/experimental/jax2tf/primitives_with_limited_support.md", "diff": "# Primitives with limited support\n-*Last generated on (YYYY-MM-DD): 2020-09-10*\n+*Last generated on (YYYY-MM-DD): ...
Python
Apache License 2.0
google/jax
[jax2tf] Fix tests when running with JAX_ENABLE_X64=1. (#4261) Fixed tests: - test_binary_elementwise - dynamic_update_slice - fft - population_count - test_unary_elementwise - top_k - select_and_gather_add
260,335
14.09.2020 12:31:51
25,200
7569e800146982bd4084599fc2f3cd95daa2a846
revert (google failure) * revert (google failure) Some downstream user is relying on the rank of stax's biases being 1. * only revert one change
[ { "change_type": "MODIFY", "old_path": "jax/experimental/stax.py", "new_path": "jax/experimental/stax.py", "diff": "@@ -49,8 +49,7 @@ def Dense(out_dim, W_init=glorot_normal(), b_init=normal()):\ndef init_fun(rng, input_shape):\noutput_shape = input_shape[:-1] + (out_dim,)\nk1, k2 = random.split(rng...
Python
Apache License 2.0
google/jax
revert #4277 (google failure) (#4281) * revert #4277 (google failure) Some downstream user is relying on the rank of stax's biases being 1. * only revert one change
260,298
15.09.2020 08:35:35
-7,200
4e04d4eaf36da15e9c14f4d5ea41850334607dc3
[jax2tf] Build a primitive harness for test_type_promotion. * [jax2tf] Build a primitive harness for test_type_promotion. We were previously generating the cases using `jtu.cases_from_list`, which by default dropped 2 test cases (JAX_NUM_GENERATED_CASES=10, number of generated cases = 12). * [jax2tf] Fix the generated test cases for test_type_promotion.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/tests/primitives_test.py", "new_path": "jax/experimental/jax2tf/tests/primitives_test.py", "diff": "@@ -81,13 +81,13 @@ class JaxPrimitiveTest(tf_test_util.JaxToTfTestCase):\nelse:\nself.assertIn(p, tf_impl)\n- @parameterized.named_para...
Python
Apache License 2.0
google/jax
[jax2tf] Build a primitive harness for test_type_promotion. (#4279) * [jax2tf] Build a primitive harness for test_type_promotion. We were previously generating the cases using `jtu.cases_from_list`, which by default dropped 2 test cases (JAX_NUM_GENERATED_CASES=10, number of generated cases = 12). * [jax2tf] Fix the generated test cases for test_type_promotion.
260,298
15.09.2020 10:40:07
-7,200
a5c2c4729e8e2a69b704df63d88311c50e6fe4ce
[jax2tf] Added support for x64 for the remaining test files * [jax2tf] Added support for x64 in other test files. This includes: control_flow_ops_test.py jax2tf_test.py saved_model_test.py stax_test
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/tests/control_flow_ops_test.py", "new_path": "jax/experimental/jax2tf/tests/control_flow_ops_test.py", "diff": "@@ -33,29 +33,29 @@ class ControlFlowOpsTest(tf_test_util.JaxToTfTestCase):\ndef f_jax(pred, x):\nreturn lax.cond(pred, lamb...
Python
Apache License 2.0
google/jax
[jax2tf] Added support for x64 for the remaining test files (#4282) * [jax2tf] Added support for x64 in other test files. This includes: - control_flow_ops_test.py - jax2tf_test.py - saved_model_test.py - stax_test
260,335
15.09.2020 12:36:53
25,200
5520948b0cb4e7390e6039f9a7cb65ed1a5f95a5
tweak traceback for unbound axis names
[ { "change_type": "MODIFY", "old_path": "jax/interpreters/xla.py", "new_path": "jax/interpreters/xla.py", "diff": "@@ -456,7 +456,7 @@ def axis_read(axis_env, axis_name):\ntry:\nreturn max(i for i, name in enumerate(axis_env.names) if name == axis_name)\nexcept ValueError:\n- raise NameError(\"unboun...
Python
Apache License 2.0
google/jax
tweak traceback for unbound axis names (#4295)
260,411
16.09.2020 11:44:07
-10,800
a9430561604c0547729911ef55b176c0550b110d
[jax2tf] Enable testing for SVD on TPU for float16
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/tests/correctness_stats.py", "new_path": "jax/experimental/jax2tf/tests/correctness_stats.py", "diff": "@@ -105,7 +105,7 @@ def categorize(prim: core.Primitive, *args, **kwargs) \\\nif prim is lax_linalg.svd_p:\nnp_dtype = _to_np_dtype(...
Python
Apache License 2.0
google/jax
[jax2tf] Enable testing for SVD on TPU for float16 (#4288)
260,298
16.09.2020 10:46:32
-7,200
1f95414f94ce5ed3f0b47c78c7d24e750f1af15f
[jax2tf] Add tests for the conversion of conv_general_dilated * [jax2tf] Add tests for the conversion of conv_general_dilated. This also adds the precision argument to the tfxla call which was previously ignored. * Separate orthogonal tests.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "@@ -625,7 +625,7 @@ def _concatenate(*operands, dimension=None):\ntf_impl[lax.concatenate_p] = _concatenate\n-def _conv_general_proto(dimension_numbers):\n+def _conv_...
Python
Apache License 2.0
google/jax
[jax2tf] Add tests for the conversion of conv_general_dilated (#4222) * [jax2tf] Add tests for the conversion of conv_general_dilated. This also adds the precision argument to the tfxla call which was previously ignored. * Separate orthogonal tests.
260,411
16.09.2020 15:08:46
-10,800
a433c16feb2bc0703a160ad931b13dc65b93760d
[jax2tf] Disable some convolution tests
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/tests/primitive_harness.py", "new_path": "jax/experimental/jax2tf/tests/primitive_harness.py", "diff": "@@ -905,10 +905,10 @@ lax_conv_general_dilated = tuple( # Validate dtypes and precision\n# This first harness runs the tests for all...
Python
Apache License 2.0
google/jax
[jax2tf] Disable some convolution tests (#4303)
260,411
16.09.2020 15:29:14
-10,800
dcaa28c624ef3402529786a5aa6dd3891f8779b0
[jax2tf] More convolution test disabling
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/tests/primitives_test.py", "new_path": "jax/experimental/jax2tf/tests/primitives_test.py", "diff": "@@ -455,6 +455,8 @@ class JaxPrimitiveTest(tf_test_util.JaxToTfTestCase):\n@primitive_harness.parameterized(primitive_harness.lax_conv_g...
Python
Apache License 2.0
google/jax
[jax2tf] More convolution test disabling (#4304)
260,298
17.09.2020 07:41:38
-7,200
504e28278822fd683cd4aa96d1bf38cedae7c22b
[jax2tf] Fix precision casting problem in convolution. In Python 3.6 (maybe 3.7 too?), the lax.Precision enumeration was not implicitly casted to int, which made the construction of the xla_data_pb2.PrecisionConfig object fail in the conversion of convolution.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "@@ -670,7 +670,7 @@ def _conv_general_precision_config_proto(precision):\nreturn None\nproto = xla_data_pb2.PrecisionConfig()\n- proto.operand_precision.append(precis...
Python
Apache License 2.0
google/jax
[jax2tf] Fix precision casting problem in convolution. (#4306) In Python 3.6 (maybe 3.7 too?), the lax.Precision enumeration was not implicitly casted to int, which made the construction of the xla_data_pb2.PrecisionConfig object fail in the conversion of convolution.
260,285
17.09.2020 08:58:32
-7,200
c6b7269480c773eb312e3fe3f15938110927db20
Support non-fragmenting mask of reshape
[ { "change_type": "MODIFY", "old_path": "jax/interpreters/masking.py", "new_path": "jax/interpreters/masking.py", "diff": "@@ -403,6 +403,8 @@ class MaskTrace(Trace):\nout_aval = primitive.abstract_eval(*(t.aval for t in tracers), **params)\nvals, polymorphic_shapes = unzip2((t.val, t.polymorphic_sha...
Python
Apache License 2.0
google/jax
Support non-fragmenting mask of reshape (#4264)
260,335
16.09.2020 23:59:58
25,200
b81c246a182a9be7df376f79b3b6a5d4e3add1bd
move the trace liveness check from
[ { "change_type": "MODIFY", "old_path": "jax/core.py", "new_path": "jax/core.py", "diff": "@@ -365,6 +365,7 @@ class Trace:\ndef full_raise(self, val) -> 'Tracer':\nif not isinstance(val, Tracer):\nreturn self.pure(val)\n+ val._assert_live()\nlevel = self.level\nsublevel = self.sublevel\nif val._trac...
Python
Apache License 2.0
google/jax
move the trace liveness check from #4312 (#4315)
260,298
17.09.2020 10:03:31
-7,200
b1d0f87648f73b06091ea3929a52b5d572391088
[jax2tf] Group error messages by dtype in pprint_limitations. * [jax2tf] Group error messages by dtype in pprint_limitations. This makes the output of the categorizer more synthetic in cases when the error is exactly the same for a given primitive on a set of devices for different dtypes.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/primitives_with_limited_support.md", "new_path": "jax/experimental/jax2tf/primitives_with_limited_support.md", "diff": "# Primitives with limited support\n-*Last generated on (YYYY-MM-DD): 2020-09-14*\n+*Last generated on (YYYY-MM-DD): ...
Python
Apache License 2.0
google/jax
[jax2tf] Group error messages by dtype in pprint_limitations. (#4307) * [jax2tf] Group error messages by dtype in pprint_limitations. This makes the output of the categorizer more synthetic in cases when the error is exactly the same for a given primitive on a set of devices for different dtypes.
260,411
17.09.2020 11:45:11
-10,800
2ff593747e16651521f84de67e762c17ca34363d
[jax2tf] Change precision of test_conv to fix TPU tests
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/tests/primitives_test.py", "new_path": "jax/experimental/jax2tf/tests/primitives_test.py", "diff": "@@ -455,15 +455,19 @@ class JaxPrimitiveTest(tf_test_util.JaxToTfTestCase):\n@primitive_harness.parameterized(primitive_harness.lax_conv...
Python
Apache License 2.0
google/jax
[jax2tf] Change precision of test_conv to fix TPU tests (#4317)
260,411
17.09.2020 16:31:17
-10,800
d74e81cc8b66260000bbe1f3fe90a0bfc2b9777d
[jax2tf] Disable complex convolution test on GPU: crash in TF
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/tests/primitives_test.py", "new_path": "jax/experimental/jax2tf/tests/primitives_test.py", "diff": "@@ -455,6 +455,9 @@ class JaxPrimitiveTest(tf_test_util.JaxToTfTestCase):\n@primitive_harness.parameterized(primitive_harness.lax_conv_g...
Python
Apache License 2.0
google/jax
[jax2tf] Disable complex convolution test on GPU: crash in TF (#4319)
260,604
17.09.2020 21:39:18
-19,080
40e20242db0ed1d8cceb1d93b78d07c248e693a6
Fix code quality issues Changes: Fix unnecessary generator Iterate dictionary directly instead of calling .keys() Remove global statement at the module level Use list() instead of a list comprehension Use with statement to open the file Merge isinstance calls
[ { "change_type": "MODIFY", "old_path": "build/build.py", "new_path": "build/build.py", "diff": "@@ -112,9 +112,9 @@ def download_and_verify_bazel():\nsys.stdout.write(\"\\n\")\n# Verify that the downloaded Bazel binary has the expected SHA256.\n- downloaded_file = open(tmp_path, \"rb\")\n+ with open...
Python
Apache License 2.0
google/jax
Fix code quality issues (#4302) Changes: - Fix unnecessary generator - Iterate dictionary directly instead of calling .keys() - Remove global statement at the module level - Use list() instead of a list comprehension - Use with statement to open the file - Merge isinstance calls
260,335
17.09.2020 09:57:43
25,200
11007ba0e3577dab0c19ac37f60061fdccecb016
test eval_context works w/ and w/o omnistaging
[ { "change_type": "MODIFY", "old_path": "jax/core.py", "new_path": "jax/core.py", "diff": "@@ -1493,7 +1493,8 @@ def pp_kv_pairs(kv_pairs):\n@config.register_omnistaging_disabler\ndef omnistaging_disabler() -> None:\nglobal thread_local_state, call_bind, find_top_trace, initial_style_staging, \\\n- n...
Python
Apache License 2.0
google/jax
test eval_context works w/ and w/o omnistaging (#4325)
260,298
17.09.2020 19:37:40
-7,200
8a4ee3d8516662b6f6fd95046f4d28119b3c1db2
Fix shape checking rule for conv_general_dilated. * Fix shape checking rule for conv_general_dilated. This closes google/jax#4316. * Added test based on google/jax#4316. * Change test name to be more accurate.
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax.py", "new_path": "jax/lax/lax.py", "diff": "@@ -2471,6 +2471,10 @@ def _conv_general_dilated_shape_rule(\nlhs_dilation, rhs_dilation, dimension_numbers, feature_group_count,\nbatch_group_count, **unused_kwargs) -> Tuple[int, ...]:\nassert type(dime...
Python
Apache License 2.0
google/jax
Fix shape checking rule for conv_general_dilated. (#4318) * Fix shape checking rule for conv_general_dilated. This closes google/jax#4316. * Added test based on google/jax#4316. * Change test name to be more accurate.
260,411
18.09.2020 10:30:45
-10,800
0ac25c760af2788ba0a14f8a0f5827035e4c29e0
[jax2tf] Replace tf.math.add with tf.raw_ops.AddV2 * [jax2tf] Replace tf.math.add with tf.raw_ops.AddV2 We now fixed tf.raw_ops.AddV2 to support uint32. It was already supporting uint8, so it is a better choice now than tf.math.add. This allowed us to use the threefry implementation using uint32.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/README.md", "new_path": "jax/experimental/jax2tf/README.md", "diff": "@@ -8,7 +8,7 @@ primitives using TensorFlow operations. In practice this means that you can take\nsome code written in JAX and execute it using TensorFlow eager mode,...
Python
Apache License 2.0
google/jax
[jax2tf] Replace tf.math.add with tf.raw_ops.AddV2 (#4278) * [jax2tf] Replace tf.math.add with tf.raw_ops.AddV2 We now fixed tf.raw_ops.AddV2 to support uint32. It was already supporting uint8, so it is a better choice now than tf.math.add. This allowed us to use the threefry implementation using uint32.
260,411
18.09.2020 11:37:34
-10,800
ded7b3854c37f93652e7eed4f4cd50522f3be70f
[jax2tf] Revert '[jax2tf] Replace tf.math.add with tf.raw_ops.AddV2 (#4278)' Generates errors due to Grappler replacing AddV2 with AddN, which is not implemented for uint32
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "@@ -441,14 +441,7 @@ except AttributeError:\npass\ntf_impl[ad_util.stop_gradient_p] = tf.stop_gradient\ntf_impl[ad_util.zeros_like_p] = tf.zeros_like\n-\n-def _add(*o...
Python
Apache License 2.0
google/jax
[jax2tf] Revert '[jax2tf] Replace tf.math.add with tf.raw_ops.AddV2 (#4278)' (#4332) Generates errors due to Grappler replacing AddV2 with AddN, which is not implemented for uint32
260,411
18.09.2020 11:55:26
-10,800
8376d92049624bf0784647b17b1f09015acd0947
Disable testExpmGrad on TPU, pending investigation of compiler error
[ { "change_type": "MODIFY", "old_path": "tests/linalg_test.py", "new_path": "tests/linalg_test.py", "diff": "@@ -1346,7 +1346,10 @@ class ScipyLinalgTest(jtu.JaxTestCase):\na = rng((n, n), dtype)\ndef expm(x):\nreturn jsp.linalg.expm(x, upper_triangular=False, max_squarings=16)\n- jtu.check_grads(exp...
Python
Apache License 2.0
google/jax
Disable testExpmGrad on TPU, pending investigation of compiler error (#4333)
260,335
18.09.2020 10:49:04
25,200
92c9713593df177501f6861afdcfe786f2d79738
add tests for new error message info
[ { "change_type": "MODIFY", "old_path": "jax/interpreters/partial_eval.py", "new_path": "jax/interpreters/partial_eval.py", "diff": "@@ -829,12 +829,12 @@ class DynamicJaxprTracer(core.Tracer):\n\"for `jit` to avoid tracing particular arguments of transformed \"\n\"functions, though at the cost of mo...
Python
Apache License 2.0
google/jax
add tests for new error message info
260,335
18.09.2020 11:09:03
25,200
be6cae35b7177a9a2c1fba9e35d46038fd44c055
only enable new tests with omnistaging
[ { "change_type": "MODIFY", "old_path": "tests/api_test.py", "new_path": "tests/api_test.py", "diff": "@@ -1793,6 +1793,9 @@ class APITest(jtu.JaxTestCase):\nf() # doesn't crash\ndef test_concrete_error_because_arg(self):\n+ if not config.omnistaging_enabled:\n+ raise unittest.SkipTest(\"test is omni...
Python
Apache License 2.0
google/jax
only enable new tests with omnistaging
260,325
08.09.2020 13:51:19
25,200
7fd7009c231bf9a86e009f6e86ccd4b374cb29dc
Expose scale_and_translate as a public function, fix a bug in implementation when translation is not 0. Change implementation to use native JAX everywhere allowing vmaping and gradients wrt scale and translation.
[ { "change_type": "MODIFY", "old_path": "docs/jax.image.rst", "new_path": "docs/jax.image.rst", "diff": "@@ -13,4 +13,5 @@ Image manipulation functions\n:toctree: _autosummary\nresize\n+ scale_and_translate\n" }, { "change_type": "MODIFY", "old_path": "jax/image/__init__.py", "new_pat...
Python
Apache License 2.0
google/jax
Expose scale_and_translate as a public function, fix a bug in implementation when translation is not 0. Change implementation to use native JAX everywhere allowing vmaping and gradients wrt scale and translation.
260,393
18.09.2020 14:58:03
25,200
2bc92f5593c04dfd9a7bbb65ed77b70092038031
Fixed ppermute translation rule
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax_parallel.py", "new_path": "jax/lax/lax_parallel.py", "diff": "@@ -547,7 +547,6 @@ ppermute_p.def_abstract_eval(lambda x, **params: raise_to_shaped(x))\nad.deflinear(ppermute_p, _ppermute_transpose_rule)\nxla.parallel_translations[ppermute_p] = _ppe...
Python
Apache License 2.0
google/jax
Fixed ppermute translation rule (#4349)
260,335
18.09.2020 17:39:05
25,200
f172fb74e17f250f769ca398d2eb563d88153ac2
plumb donate_argnums into jax.xla_computation
[ { "change_type": "MODIFY", "old_path": "jax/api.py", "new_path": "jax/api.py", "diff": "@@ -437,7 +437,8 @@ def xla_computation(fun: Callable,\nbackend: Optional[str] = None,\ntuple_args: bool = False,\ninstantiate_const_outputs: Optional[bool] = None,\n- return_shape: bool = False) -> Callable:\n+ ...
Python
Apache License 2.0
google/jax
plumb donate_argnums into jax.xla_computation
260,335
19.09.2020 19:41:17
25,200
50dd9c5016ce0bceaad7256208e790b0b02a3142
don't initialize backend in xla_computation This should allow us to use donate_argnums *and* build HLO computations for backends not available at build time.
[ { "change_type": "MODIFY", "old_path": "jax/api.py", "new_path": "jax/api.py", "diff": "@@ -676,12 +676,9 @@ def _xla_computation(\nout_tuple = build_out_tuple()\nif any(donated_invars):\n- # TODO(tomhennigan): At call time we should mark these buffers as deleted.\n- backend_ = xb.get_backend(backen...
Python
Apache License 2.0
google/jax
don't initialize backend in xla_computation This should allow us to use donate_argnums *and* build HLO computations for backends not available at build time.
260,335
19.09.2020 22:19:29
25,200
1092fa1d9b3de680aca5a883b781e6ba97c940be
fix logic, skip test
[ { "change_type": "MODIFY", "old_path": "jax/api.py", "new_path": "jax/api.py", "diff": "@@ -675,7 +675,7 @@ def _xla_computation(\nelse:\nout_tuple = build_out_tuple()\n- if any(donated_invars) and backend in (\"gpu\", \"tpu\"):\n+ if any(donated_invars):\ndonated_invars = xla.set_up_aliases(c, xla_...
Python
Apache License 2.0
google/jax
fix logic, skip test
260,287
21.09.2020 14:14:52
0
c4f98eb8fa2947e0db4a65bb67daa3243e7d103d
Add back the batching rule for ppermute Just make sure it's correct this time and add a test.
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax_parallel.py", "new_path": "jax/lax/lax_parallel.py", "diff": "@@ -547,6 +547,7 @@ ppermute_p.def_abstract_eval(lambda x, **params: raise_to_shaped(x))\nad.deflinear(ppermute_p, _ppermute_transpose_rule)\nxla.parallel_translations[ppermute_p] = _ppe...
Python
Apache License 2.0
google/jax
Add back the batching rule for ppermute Just make sure it's correct this time and add a test.
260,287
21.09.2020 16:25:50
0
2081e5acee2039930c250ccb91483ed6d6cfe580
Test pmap/vmap interactions of all reduction collectives
[ { "change_type": "MODIFY", "old_path": "tests/pmap_test.py", "new_path": "tests/pmap_test.py", "diff": "@@ -1522,14 +1522,18 @@ class VmapOfPmapTest(jtu.JaxTestCase):\nexpected = np.stack([fun(*args_slice(i)) for i in range(vmapped_size)])\nself.assertAllClose(ans, expected)\n+ @parameterized.named_...
Python
Apache License 2.0
google/jax
Test pmap/vmap interactions of all reduction collectives
260,631
21.09.2020 14:18:31
25,200
55c6bdfe9c5d0631200cb76e4f56481b3656b03f
Clean-up todos related to the upgrade of jaxlib.
[ { "change_type": "MODIFY", "old_path": "jax/api.py", "new_path": "jax/api.py", "diff": "@@ -49,6 +49,7 @@ from .tree_util import (tree_map, tree_flatten, tree_unflatten, tree_structure,\ntreedef_is_leaf, Partial)\nfrom .util import (unzip2, curry, partial, safe_map, safe_zip, prod, split_list,\nexte...
Python
Apache License 2.0
google/jax
Clean-up todos related to the upgrade of jaxlib. PiperOrigin-RevId: 332932271
260,325
21.09.2020 16:20:17
25,200
be50847ceeafc3d08de31bae21e9404b6283f05f
Make scale_and_translate take spatial dimensions
[ { "change_type": "MODIFY", "old_path": "jax/image/scale.py", "new_path": "jax/image/scale.py", "diff": "@@ -52,7 +52,7 @@ def compute_weight_mat(input_size: int, output_size: int, scale,\n# When downsampling the kernel should be scaled since we want to low pass\n# filter and interpolate, but when up...
Python
Apache License 2.0
google/jax
Make scale_and_translate take spatial dimensions
260,325
21.09.2020 16:21:48
25,200
ae910cdd311800fd1f2057b62d44028575820f60
Updating image_test
[ { "change_type": "MODIFY", "old_path": "tests/image_test.py", "new_path": "tests/image_test.py", "diff": "@@ -21,6 +21,7 @@ import numpy as np\nfrom absl.testing import absltest\nfrom absl.testing import parameterized\n+import jax\nfrom jax import image\nfrom jax import numpy as jnp\nfrom jax import...
Python
Apache License 2.0
google/jax
Updating image_test
260,335
21.09.2020 17:30:56
25,200
2abb37c286c92d70b8fb0104aacea60cda7675ce
move a _device_put_raw under broadcast impl Before this change, we had to interact with the device to construct an array of zeros, even if we were staging everything out (e.g. with jax.xla_computation and omnistaging).
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax.py", "new_path": "jax/lax/lax.py", "diff": "@@ -657,7 +657,8 @@ def broadcast_in_dim(operand: Array, shape: Shape,\n\"\"\"\nshape = _broadcast_in_dim_shape_rule(\noperand, shape=shape, broadcast_dimensions=broadcast_dimensions)\n- if np.ndim(operan...
Python
Apache License 2.0
google/jax
move a _device_put_raw under broadcast impl Before this change, we had to interact with the device to construct an array of zeros, even if we were staging everything out (e.g. with jax.xla_computation and omnistaging).
260,335
21.09.2020 19:33:14
25,200
34e04609610bd94498a0eb0eb0b902e4cb2fe287
only test if omnistaging is enabled
[ { "change_type": "MODIFY", "old_path": "tests/api_test.py", "new_path": "tests/api_test.py", "diff": "@@ -1798,6 +1798,9 @@ class APITest(jtu.JaxTestCase):\nf() # doesn't crash\ndef test_xla_computation_zeros_doesnt_device_put(self):\n+ if not config.omnistaging_enabled:\n+ raise unittest.SkipTest(\...
Python
Apache License 2.0
google/jax
only test if omnistaging is enabled
260,548
21.09.2020 20:59:08
-3,600
6aa35a2a7351e3c2bbf2fa1f58628801e95c8791
Adding an option to return the output tree in make_jaxpr
[ { "change_type": "MODIFY", "old_path": "jax/api.py", "new_path": "jax/api.py", "diff": "@@ -1938,7 +1938,8 @@ def linear_transpose(fun: Callable, *primals) -> Callable:\ndef make_jaxpr(fun: Callable,\n- static_argnums: Union[int, Iterable[int]] = ()\n+ static_argnums: Union[int, Iterable[int]] = (),...
Python
Apache License 2.0
google/jax
Adding an option to return the output tree in make_jaxpr
260,287
22.09.2020 13:08:38
0
e0d1b375fa593c5ef777e7e650a3a3e75996cedc
Delete dead axis_index code The primitive was moved to `lax_parallel.py` some time ago, so the one in `core` should no longer be used. This is probably a result of a botched rebase.
[ { "change_type": "MODIFY", "old_path": "jax/core.py", "new_path": "jax/core.py", "diff": "@@ -1191,48 +1191,6 @@ def axis_frame(axis_name):\nraise NameError(\"unbound axis name: {}\".format(axis_name))\n-def axis_index(axis_name):\n- \"\"\"Return the index along the mapped axis ``axis_name``.\n-\n- ...
Python
Apache License 2.0
google/jax
Delete dead axis_index code The primitive was moved to `lax_parallel.py` some time ago, so the one in `core` should no longer be used. This is probably a result of a botched rebase.
260,287
22.09.2020 16:05:24
0
8ac19c722211ffe4c7fb0bdbeeb236818d000291
Fix a faulty soft_pmap rule for axis_index The rule didn't specify the precision for the `np.arange` constant, which caused an accidental dtype promotion in X64 mode. Previously the error has luckicly been hidden behind a coerction that followed `axis_index` in that test, but the new implementation has surfaced it.
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax_parallel.py", "new_path": "jax/lax/lax_parallel.py", "diff": "@@ -678,7 +678,7 @@ def _axis_index_translation_rule(c, *, axis_name, axis_env, platform):\ndef _axis_index_soft_pmap_rule(vals, mapped, chunk_size, *, axis_name):\nassert not vals and n...
Python
Apache License 2.0
google/jax
Fix a faulty soft_pmap rule for axis_index The rule didn't specify the precision for the `np.arange` constant, which caused an accidental dtype promotion in X64 mode. Previously the error has luckicly been hidden behind a coerction that followed `axis_index` in that test, but the new implementation has surfaced it.
260,411
23.09.2020 13:14:36
-10,800
625be69333ad46c7e0b4cb765060054faf5f2927
[host_callback] Update the documentation The module-level documentation was out of date.
[ { "change_type": "MODIFY", "old_path": "jax/experimental/host_callback.py", "new_path": "jax/experimental/host_callback.py", "diff": "This module introduces the host callback functions :func:`id_tap` and\n:func:`id_print`, which behave like the identity function but have the\n-side-effect of sending...
Python
Apache License 2.0
google/jax
[host_callback] Update the documentation The module-level documentation was out of date.
260,287
23.09.2020 12:59:01
0
2b7580c2d206e692a83fbc37eaeeb846e0a4462f
Consider lists as groups of axis names too
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax_parallel.py", "new_path": "jax/lax/lax_parallel.py", "diff": "@@ -692,7 +692,7 @@ pxla.multi_host_supported_collectives.add(axis_index_p)\n# wants to bind an axis name has to additionally implement `process_axis_index`\n# and put its main trace on ...
Python
Apache License 2.0
google/jax
Consider lists as groups of axis names too
260,287
23.09.2020 10:45:23
0
0d5f15f5c0f862d3d5cbf8f6853409cdc1091de1
Fix the abstract eval and translation rule for all_to_all The previous rules assumed that `split_axis == concat_axis` (i.e. that the used collective is equivalent to `pswapaxes`). Since we expose this as part of our API, we should probably make sure that we handle other cases too. Fixes
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax_parallel.py", "new_path": "jax/lax/lax_parallel.py", "diff": "@@ -557,13 +557,25 @@ def _all_to_all_translation_rule(c, x, *, split_axis, concat_axis, axis_name,\nreplica_groups = _replica_groups(axis_env, axis_name, None)\nif len(replica_groups[0]...
Python
Apache License 2.0
google/jax
Fix the abstract eval and translation rule for all_to_all The previous rules assumed that `split_axis == concat_axis` (i.e. that the used collective is equivalent to `pswapaxes`). Since we expose this as part of our API, we should probably make sure that we handle other cases too. Fixes #1332.
260,335
23.09.2020 19:37:34
25,200
c42d736e347d059290ab20c013635d62a1ee6c45
remove limit on size of random arrays
[ { "change_type": "MODIFY", "old_path": "jax/random.py", "new_path": "jax/random.py", "diff": "@@ -308,12 +308,17 @@ def _random_bits(key, bit_width, shape):\nraise TypeError(\"requires 8-, 16-, 32- or 64-bit field width.\")\nsize = prod(shape)\nmax_count = int(np.ceil(bit_width * size / 32))\n- if m...
Python
Apache License 2.0
google/jax
remove limit on size of random arrays
260,335
23.09.2020 19:39:22
25,200
96f5a3c4026c929664e75e07262bba7a4c8d2044
fix test for non-omnistaging
[ { "change_type": "MODIFY", "old_path": "tests/random_test.py", "new_path": "tests/random_test.py", "diff": "@@ -858,9 +858,9 @@ class LaxRandomTest(jtu.JaxTestCase):\nrandom.choice(key, 5, 2, replace=True)\ndef test_eval_shape_big_random_array(self):\n- def f():\n- return random.normal(random.PRNGKe...
Python
Apache License 2.0
google/jax
fix test for non-omnistaging
260,335
23.09.2020 20:15:32
25,200
71f5f9972cd305d4060637115a7ff316087d229e
skip checks in big randomness test
[ { "change_type": "MODIFY", "old_path": "tests/random_test.py", "new_path": "tests/random_test.py", "diff": "@@ -859,7 +859,8 @@ class LaxRandomTest(jtu.JaxTestCase):\ndef test_eval_shape_big_random_array(self):\ndef f(x):\n- return random.normal(random.PRNGKey(x), (int(1e10),))\n+ return random.norm...
Python
Apache License 2.0
google/jax
skip checks in big randomness test
260,335
23.09.2020 20:41:57
25,200
d607164d35e05077d60969a8e6b145f10aca95e0
make_jaxpr return_shape use ShapeDtypeStruct, test
[ { "change_type": "MODIFY", "old_path": "jax/api.py", "new_path": "jax/api.py", "diff": "@@ -1957,7 +1957,11 @@ def make_jaxpr(fun: Callable,\nReturns:\nA wrapped version of ``fun`` that when applied to example arguments returns\n- a ``ClosedJaxpr`` representation of ``fun`` on those arguments.\n+ a ...
Python
Apache License 2.0
google/jax
make_jaxpr return_shape use ShapeDtypeStruct, test
260,335
23.09.2020 21:09:44
25,200
ebf7c1b6127d6df3a64a4d969297ca70e126a341
add jax logo file
[ { "change_type": "ADD", "old_path": null, "new_path": "images/jax_logo.svg", "diff": "+<svg xmlns=\"http://www.w3.org/2000/svg\" viewBox=\"0 0 451 260.81\"><defs><style>.cls-1{fill:#5e97f6;}.cls-1,.cls-2,.cls-3,.cls-4,.cls-5,.cls-6,.cls-7,.cls-8,.cls-9{stroke:#dce0df;stroke-linejoin:round;}.cls-2{fi...
Python
Apache License 2.0
google/jax
add jax logo file
260,411
24.09.2020 15:08:07
-10,800
03970e9787bcf7a04257ea9b0b6f537ceac68d4f
Tag tests that require omnistaging
[ { "change_type": "MODIFY", "old_path": "tests/host_callback_test.py", "new_path": "tests/host_callback_test.py", "diff": "@@ -294,6 +294,8 @@ class HostCallbackTest(jtu.JaxTestCase):\ndef test_jit_result_unused(self):\n\"\"\"We can id_print even if we don't use the result.\"\"\"\n+ if not config.omn...
Python
Apache License 2.0
google/jax
Tag tests that require omnistaging
260,287
28.08.2020 15:21:50
0
acd4cc573722b2f5069e69d64f15b71e4502da55
Allow vmapping all_to_all and implement a (slow) CPU and GPU translation This allows pmapping vmapped computations that use `all_to_all` or `pswapaxes` inside. It also includes a very slow CPU and GPU translation rule that might be useful for debugging programs locally, since XLA only implements the `AllToAll` collective on TPUs. Fixes
[ { "change_type": "MODIFY", "old_path": "jax/interpreters/xla.py", "new_path": "jax/interpreters/xla.py", "diff": "@@ -857,7 +857,7 @@ def _tuple_output(*args, **kwargs):\nans = yield args, kwargs\nyield (ans,)\n-def lower_fun(fun, multiple_results):\n+def lower_fun(fun, multiple_results, parallel=Fa...
Python
Apache License 2.0
google/jax
Allow vmapping all_to_all and implement a (slow) CPU and GPU translation This allows pmapping vmapped computations that use `all_to_all` or `pswapaxes` inside. It also includes a very slow CPU and GPU translation rule that might be useful for debugging programs locally, since XLA only implements the `AllToAll` collective on TPUs. Fixes #4141.
260,411
25.09.2020 15:28:23
-10,800
f46a42206c0917d4da5f040f273aa0fefef050bd
[host_callback] improved logging, and add instructions for debugging
[ { "change_type": "MODIFY", "old_path": "jax/experimental/host_callback.py", "new_path": "jax/experimental/host_callback.py", "diff": "@@ -56,7 +56,8 @@ which eventually pauses the computation on devices. The runtime has one\nadditional thread that invokes the Python user functions with the received ...
Python
Apache License 2.0
google/jax
[host_callback] improved logging, and add instructions for debugging
260,335
25.09.2020 11:50:32
25,200
90b272aa88baa90ecbd89734eb5309f8d58eac2d
add author line, fix typo
[ { "change_type": "MODIFY", "old_path": "design_notes/omnistaging.md", "new_path": "design_notes/omnistaging.md", "diff": "# Omnistaging\n+_mattjj@_\n+_Sept 25 2020_\nThis is more of an upgrade guide than a design doc.\n@@ -6,7 +8,7 @@ This is more of an upgrade guide than a design doc.\n* [tl;dr](#t...
Python
Apache License 2.0
google/jax
add author line, fix typo
260,335
25.09.2020 12:10:03
25,200
be6e85a176ed4cb180c96082a2a779f85bf25e65
fix links in omnistaging doc (hopefully...)
[ { "change_type": "MODIFY", "old_path": "design_notes/omnistaging.md", "new_path": "design_notes/omnistaging.md", "diff": "@@ -7,8 +7,13 @@ This is more of an upgrade guide than a design doc.\n### Contents\n* [tl;dr](#tldr)\n-* [What is \"omnistaging\" and why is it useful?](what-is-omnistaging-and-w...
Python
Apache License 2.0
google/jax
fix links in omnistaging doc (hopefully...)
260,335
25.09.2020 15:35:44
25,200
61fcea93e52976d94899461d217070b1db59efde
tweak error to mention *flattened* positions
[ { "change_type": "MODIFY", "old_path": "jax/interpreters/partial_eval.py", "new_path": "jax/interpreters/partial_eval.py", "diff": "@@ -805,7 +805,7 @@ class DynamicJaxprTracer(core.Tracer):\norigin = (f\"While tracing the function {self._trace.main.source_info}, \"\n\"this concrete value was not av...
Python
Apache License 2.0
google/jax
tweak error to mention *flattened* positions
260,287
22.09.2020 11:19:06
0
fa38f250e314346d6d47586c99dd4fffb1b1daba
Add support for all_to_all over vmapped axes
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax_parallel.py", "new_path": "jax/lax/lax_parallel.py", "diff": "@@ -612,6 +612,17 @@ def _all_to_all_batcher(vals_in, dims_in, *, axis_name, split_axis, concat_axis)\nresult = all_to_all_p.bind(x, axis_name=axis_name, split_axis=split_axis, concat_ax...
Python
Apache License 2.0
google/jax
Add support for all_to_all over vmapped axes
260,287
22.09.2020 13:05:08
0
e61ca913606bbb8a92c57ef49852bd736c064558
Implement split_axis for all_to_all This allows us to use `all_to_all` over a mix of vmapped and pmapped dimensions, which will be useful for `gmap`.
[ { "change_type": "MODIFY", "old_path": "jax/interpreters/pxla.py", "new_path": "jax/interpreters/pxla.py", "diff": "@@ -964,6 +964,7 @@ def _pmap_translation_rule(c, axis_env,\n_xla_shard(c, aval, new_env, in_node) if in_node_mapped else in_node\nfor aval, in_node, in_node_mapped in zip(in_avals, in...
Python
Apache License 2.0
google/jax
Implement split_axis for all_to_all This allows us to use `all_to_all` over a mix of vmapped and pmapped dimensions, which will be useful for `gmap`.
260,424
01.10.2020 13:07:33
-3,600
cc0114a0a9dc330f5cf6e2f9318e593482eff43f
Fix dtype behavior with float0s in CustomVJP.
[ { "change_type": "MODIFY", "old_path": "jax/core.py", "new_path": "jax/core.py", "diff": "@@ -1594,5 +1594,7 @@ def omnistaging_disabler() -> None:\ntrace_state.initial_style = prev\n# Casting float0 array to a float-valued zero array.\n-def zeros_like_float0(array):\n- return np.zeros(array.shape, ...
Python
Apache License 2.0
google/jax
Fix dtype behavior with float0s in CustomVJP.
260,639
21.09.2020 16:59:46
14,400
0893b08a6fa71108992aea564beb6cbabfd5fce7
histogramdd implemenation
[ { "change_type": "MODIFY", "old_path": "docs/jax.numpy.rst", "new_path": "docs/jax.numpy.rst", "diff": "@@ -176,6 +176,7 @@ Not every function in NumPy is implemented; contributions are welcome!\nheaviside\nhistogram\nhistogram_bin_edges\n+ histogramdd\nhsplit\nhstack\nhypot\n" }, { "change_...
Python
Apache License 2.0
google/jax
histogramdd implemenation
260,325
05.10.2020 10:19:18
25,200
b357005e60440657321737b1465f3b900a2c9b8c
Update Common_Gotchas_in_JAX.ipynb Clarify that the index is clamped to the bounds of the array when accessing out of bounds.
[ { "change_type": "MODIFY", "old_path": "docs/notebooks/Common_Gotchas_in_JAX.ipynb", "new_path": "docs/notebooks/Common_Gotchas_in_JAX.ipynb", "diff": "\"id\": \"eoXrGARWypdR\"\n},\n\"source\": [\n- \"However, raising an error on other accelerators can be more difficult. Therefore, JAX does not rais...
Python
Apache License 2.0
google/jax
Update Common_Gotchas_in_JAX.ipynb Clarify that the index is clamped to the bounds of the array when accessing out of bounds.
260,411
07.10.2020 11:14:35
-10,800
ae7e5b99d6e12d7410564c429fb423126f810dbd
Another attempt to save the notebook
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/JAX2TF_getting_started.ipynb", "new_path": "jax/experimental/jax2tf/JAX2TF_getting_started.ipynb", "diff": "\"output_type\": \"stream\",\n\"text\": [\n\"Execute with TF eager: hello_tf(x) = -0.44657254219055176\\n\",\n+ \"WARNING:tensor...
Python
Apache License 2.0
google/jax
Another attempt to save the notebook
260,411
08.10.2020 12:28:42
-10,800
dc9168ba93764a6f282dd33c376e8dd84da7d8db
[jax2tf] Ensure that in tests TF does not constant-fold in eager mode before compiling
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/tests/control_flow_ops_test.py", "new_path": "jax/experimental/jax2tf/tests/control_flow_ops_test.py", "diff": "@@ -33,15 +33,15 @@ class ControlFlowOpsTest(tf_test_util.JaxToTfTestCase):\ndef f_jax(pred, x):\nreturn lax.cond(pred, lamb...
Python
Apache License 2.0
google/jax
[jax2tf] Ensure that in tests TF does not constant-fold in eager mode before compiling
260,424
08.10.2020 15:36:05
-3,600
e3d622cc67edb9c0ea18c336798d3f29cbfeae90
Recast int/bool tangents to float0 in custom_jvp/vjps (also in the initial_style path).
[ { "change_type": "MODIFY", "old_path": "jax/custom_derivatives.py", "new_path": "jax/custom_derivatives.py", "diff": "@@ -306,8 +306,14 @@ custom_jvp_call_jaxpr_p.def_abstract_eval(_custom_jvp_call_jaxpr_abstract_eval)\ndef _custom_jvp_call_jaxpr_jvp(primals, tangents, *, fun_jaxpr, jvp_jaxpr_thunk)...
Python
Apache License 2.0
google/jax
Recast int/bool tangents to float0 in custom_jvp/vjps (also in the initial_style path).
260,265
08.10.2020 12:12:15
25,200
a2a9409a8ac5bd3e60dbb2f1e854304e0ce0e64b
Mark pmin and pmax as multi-host supported.
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax_parallel.py", "new_path": "jax/lax/lax_parallel.py", "diff": "@@ -491,6 +491,7 @@ pmax_p = core.Primitive('pmax')\npmax_p.def_abstract_eval(lambda x, **params: raise_to_shaped(x))\nxla.parallel_translations[pmax_p] = \\\npartial(_allreduce_translat...
Python
Apache License 2.0
google/jax
Mark pmin and pmax as multi-host supported.
260,335
08.10.2020 13:00:32
25,200
09f2be15d29b6fc9c9df5808afca1ef399593e20
wait for result in debug_nans_test
[ { "change_type": "MODIFY", "old_path": "tests/debug_nans_test.py", "new_path": "tests/debug_nans_test.py", "diff": "@@ -34,20 +34,24 @@ class DebugNaNsTest(jtu.JaxTestCase):\ndef testSingleResultPrimitiveNoNaN(self):\nA = jnp.array([[1., 2.], [2., 3.]])\n- _ = jnp.tanh(A)\n+ ans = jnp.tanh(A)\n+ ans...
Python
Apache License 2.0
google/jax
wait for result in debug_nans_test
260,335
08.10.2020 13:34:56
25,200
24de811a392040bfe85a3fcc24a1ca3be715d4b0
move a debug_nans test into debug_nans test file
[ { "change_type": "MODIFY", "old_path": "tests/api_test.py", "new_path": "tests/api_test.py", "diff": "@@ -399,20 +399,6 @@ class PythonJitTest(CPPJitTest):\njit_fun(a) # doesn't crash\n- def test_debug_nans(self):\n- @self.jit\n- def f(x):\n- return jnp.add(x, np.nan)\n- f(1)\n-\n- msg = r\"invalid ...
Python
Apache License 2.0
google/jax
move a debug_nans test into debug_nans test file
260,335
08.10.2020 15:36:42
25,200
55be9dd91bc4a1ce04d250d2d7500c6819bd2076
remove mysterious code that is no longer needed
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax_control_flow.py", "new_path": "jax/lax/lax_control_flow.py", "diff": "@@ -1566,10 +1566,6 @@ def _scan_partial_eval(trace, *tracers, reverse, length, num_consts, num_carry,\nlu.wrap_init(core.jaxpr_as_fun(jaxpr_1)), in_pvals_1,\ninstantiate=[True] ...
Python
Apache License 2.0
google/jax
remove mysterious code that is no longer needed
260,335
08.10.2020 20:31:39
25,200
52fe026c099cd3572c12b6819bb19c55c58f3313
optimize scan partial_eval to fix fixes
[ { "change_type": "MODIFY", "old_path": "jax/core.py", "new_path": "jax/core.py", "diff": "@@ -232,9 +232,6 @@ class Literal:\ndef __hash__(self):\nassert False\n- def __eq__(self, other):\n- assert False\n-\ndef __repr__(self):\nif hasattr(self, 'hash'):\nreturn '{}'.format(self.val)\n" }, { ...
Python
Apache License 2.0
google/jax
optimize scan partial_eval to fix #4510 fixes #4510
260,335
09.10.2020 10:26:28
25,200
5355776e574f5fb0dd38486926a05532384fd03f
make scan fwd raw extensive inputs as DeviceArray follow-up on
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax_control_flow.py", "new_path": "jax/lax/lax_control_flow.py", "diff": "@@ -1582,6 +1582,8 @@ def _scan_partial_eval(trace, *tracers, reverse, length, num_consts, num_carry,\n# outputs from the jaxpr here, and updating out_flat below.\nextensive_inva...
Python
Apache License 2.0
google/jax
make scan fwd raw extensive inputs as DeviceArray follow-up on #4517
260,335
09.10.2020 10:35:47
25,200
0aeeb63f85c52a6f35a9408d4581d3a56b83f34a
add test for scan input forwarding
[ { "change_type": "MODIFY", "old_path": "tests/lax_control_flow_test.py", "new_path": "tests/lax_control_flow_test.py", "diff": "@@ -34,6 +34,7 @@ from jax import random\nfrom jax import test_util as jtu\nfrom jax.util import unzip2\nfrom jax.lib import xla_bridge\n+from jax.interpreters import xla\n...
Python
Apache License 2.0
google/jax
add test for scan input forwarding
260,411
10.10.2020 10:30:23
-10,800
9d9762b714c8fd83a4e566784edf4750074cef95
[jax2tf] Ensure jax2tf still works without omnistaging This refines changes from PR 4470 to ensure that jax2tf works even without omnistaging. Some tests would fail though, e.g., converting a function with no arguments
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/jax2tf.py", "new_path": "jax/experimental/jax2tf/jax2tf.py", "diff": "@@ -20,6 +20,7 @@ import jax\nfrom jax import abstract_arrays\nfrom jax import ad_util\nfrom jax import api\n+from jax import config\nfrom jax import core\nfrom jax i...
Python
Apache License 2.0
google/jax
[jax2tf] Ensure jax2tf still works without omnistaging This refines changes from PR 4470 to ensure that jax2tf works even without omnistaging. Some tests would fail though, e.g., converting a function with no arguments
260,639
04.10.2020 17:46:13
14,400
0485fd87366bfa80477de4ac4c268887459d0231
adding histogram2d implementation
[ { "change_type": "MODIFY", "old_path": "docs/jax.numpy.rst", "new_path": "docs/jax.numpy.rst", "diff": "@@ -178,6 +178,7 @@ Not every function in NumPy is implemented; contributions are welcome!\nheaviside\nhistogram\nhistogram_bin_edges\n+ histogram2d\nhistogramdd\nhsplit\nhstack\n" }, { "c...
Python
Apache License 2.0
google/jax
adding histogram2d implementation
260,335
10.10.2020 21:08:52
25,200
4e65a6f0a9d792ebcd22ea29efffa485f1d76f8c
don't generate lazy iota/eye/tri/delta omnistaging
[ { "change_type": "MODIFY", "old_path": "jax/lax/lax.py", "new_path": "jax/lax/lax.py", "diff": "@@ -1409,6 +1409,11 @@ def iota(dtype: DType, size: int) -> Array:\n<https://www.tensorflow.org/xla/operation_semantics#iota>`_\noperator.\n\"\"\"\n+ if config.omnistaging_enabled:\n+ dtype = dtypes.canon...
Python
Apache License 2.0
google/jax
don't generate lazy iota/eye/tri/delta omnistaging
260,268
11.10.2020 11:30:45
14,400
00e70492e4e8aa0caf9377124fd6d7af35a310c3
Export expit and logsumexp in jax.nn.functions
[ { "change_type": "MODIFY", "old_path": "jax/nn/__init__.py", "new_path": "jax/nn/__init__.py", "diff": "@@ -19,6 +19,7 @@ from . import initializers\nfrom .functions import (\ncelu,\nelu,\n+ expit,\ngelu,\nglu,\nhard_sigmoid,\n@@ -28,6 +29,7 @@ from .functions import (\nleaky_relu,\nlog_sigmoid,\nlo...
Python
Apache License 2.0
google/jax
Export expit and logsumexp in jax.nn.functions
260,335
11.10.2020 10:15:48
25,200
e1708f5c91684bea7f51fe48dd65456c5e5992cb
fix typo "differentiaion"
[ { "change_type": "MODIFY", "old_path": "jax/experimental/jax2tf/README.md", "new_path": "jax/experimental/jax2tf/README.md", "diff": "@@ -60,7 +60,7 @@ restored_model = tf.saved_model.load('/some/directory')\nMore involved examples of using SavedModel are described in the\n[getting started Colab](JA...
Python
Apache License 2.0
google/jax
fix typo "differentiaion"