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 |
|---|---|---|---|---|---|---|---|---|---|
260,335 | 26.05.2020 20:01:36 | 25,200 | 9f8a4ad341acc8bab52c0746102193cb7a4da2ee | remove stray print statement from | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -775,7 +775,6 @@ def frexp(x):\nint_type = _INT_DTYPES[info.bits]\nx1, x2 = _normalize_float(x)\n- print(x1, x2)\nx1 = lax.bitcast_convert_type(x1, int_type)\nx2 += ((x1 >> info.nmant) ... | Python | Apache License 2.0 | google/jax | remove stray print statement from #1529 |
260,424 | 27.05.2020 08:59:31 | -3,600 | 1cc471928b0e677149a72518dc95fa2daa47f76e | Remove pe from name_stack and test. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/partial_eval.py",
"new_path": "jax/interpreters/partial_eval.py",
"diff": "@@ -167,8 +167,6 @@ class JaxprTrace(Trace):\nif (self.master.trace_type is StagingJaxprTrace\nand call_primitive in staged_out_calls):\ntracers = map(self.instantiate_... | Python | Apache License 2.0 | google/jax | Remove pe from name_stack and test. (#3209) |
260,485 | 27.05.2020 12:37:55 | 14,400 | ec3b593ca85d6f5c3b538b6615dfbd8c8ffe8148 | Added geometric distribution to scipy stats | [
{
"change_type": "MODIFY",
"old_path": "jax/scipy/stats/__init__.py",
"new_path": "jax/scipy/stats/__init__.py",
"diff": "@@ -26,3 +26,4 @@ from . import pareto\nfrom . import t\nfrom . import uniform\nfrom . import logistic\n+from . import geom\n"
},
{
"change_type": "ADD",
"old_path": ... | Python | Apache License 2.0 | google/jax | Added geometric distribution to scipy stats (#3205) |
260,287 | 28.05.2020 17:39:13 | -7,200 | c1ccbdf1a7d6738bd23971aa6d2718e8ba013f86 | Small cleanup for partial_eval
`partial_eval` uses some pretty tricky conventions for return values
(see `partial_eval_wrapper`), but it forces all call sites to deal with
untangling them. This commit inlines the postprocessing into
`partial_eval`, greatly simplifying its usage. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/partial_eval.py",
"new_path": "jax/interpreters/partial_eval.py",
"diff": "@@ -27,7 +27,7 @@ from .. import linear_util as lu\nfrom ..abstract_arrays import ShapedArray, ConcreteArray, raise_to_shaped\nfrom ..ad_util import zero\nfrom ..util i... | Python | Apache License 2.0 | google/jax | Small cleanup for partial_eval (#3210)
`partial_eval` uses some pretty tricky conventions for return values
(see `partial_eval_wrapper`), but it forces all call sites to deal with
untangling them. This commit inlines the postprocessing into
`partial_eval`, greatly simplifying its usage. |
260,335 | 28.05.2020 10:20:36 | 25,200 | 572928dfa309e625e221fd084bd25ee010afa2eb | fix custom_jvp_call_jaxpr transpose function
* make custom_jvp_call_jaxpr handle multilinear funs
see
* remove old comment | [
{
"change_type": "MODIFY",
"old_path": "jax/custom_derivatives.py",
"new_path": "jax/custom_derivatives.py",
"diff": "@@ -343,15 +343,11 @@ xla.initial_style_translations[custom_jvp_call_jaxpr_p] = \\\nxla.lower_fun_initial_style(_custom_jvp_call_jaxpr_impl)\n# If a (multi)linear function is defined... | Python | Apache License 2.0 | google/jax | fix custom_jvp_call_jaxpr transpose function (#3231)
* make custom_jvp_call_jaxpr handle multilinear funs
see #3226
* remove old comment |
260,366 | 28.05.2020 13:21:39 | 14,400 | 7c90023ddbbb598a2f34a805ac8c4b19f69b82e1 | Fix sign error in custom_jvp / custom_vjp.
f(x, y) = sin(x) * y.
df/dy should be sin(x) instead of -sin(x). | [
{
"change_type": "MODIFY",
"old_path": "jax/custom_derivatives.py",
"new_path": "jax/custom_derivatives.py",
"diff": "@@ -118,7 +118,7 @@ class custom_jvp:\nx, y = primals\nx_dot, y_dot = tangents\nprimal_out = f(x, y)\n- tangent_out = jnp.cos(x) * x_dot * y - jnp.sin(x) * y_dot\n+ tangent_out = jnp... | Python | Apache License 2.0 | google/jax | Fix sign error in custom_jvp / custom_vjp. (#3213) (#3219)
f(x, y) = sin(x) * y.
df/dy should be sin(x) instead of -sin(x). |
260,280 | 28.05.2020 17:16:56 | 10,800 | e48a4e012bd253428e3bb0dd4ac5b163439a5270 | uses np.prod instead of jnp.prod for shapes | [
{
"change_type": "MODIFY",
"old_path": "tests/host_callback_test.py",
"new_path": "tests/host_callback_test.py",
"diff": "@@ -554,7 +554,7 @@ where: 10\nif jtu.device_under_test() == \"tpu\":\nif dtype in (jnp.int16,):\nraise SkipTest(f\"transfering {dtype} not supported on TPU\")\n- args = [jnp.ara... | Python | Apache License 2.0 | google/jax | uses np.prod instead of jnp.prod for shapes (#3236) |
260,411 | 29.05.2020 12:54:09 | -10,800 | 5b684fc695258c03ea5e04230e08f82ae4540ffc | Attempt to fix error in google3 import
See | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -1152,8 +1152,11 @@ def squeeze(a, axis: Union[int, Tuple[int, ...]] = None):\n@_wraps(np.expand_dims)\ndef expand_dims(a, axis: Union[int, Tuple[int, ...]]):\n- if isinstance(axis, int... | Python | Apache License 2.0 | google/jax | Attempt to fix error in google3 import (#3244)
See #3243 |
260,411 | 29.05.2020 13:06:50 | -10,800 | 9691f9b065f6400c3e13fc7f3a68cb31a3fc3058 | Fix travis; yaml parser does not like comments inside shell commands | [
{
"change_type": "MODIFY",
"old_path": ".travis.yml",
"new_path": ".travis.yml",
"diff": "@@ -44,8 +44,8 @@ install:\nconda install --yes sphinx sphinx_rtd_theme nbsphinx sphinx-autodoc-typehints jupyter_client matplotlib;\npip install sklearn;\nfi\n- - if [ \"$JAX_TO_TF\" = true ] ;then\n# jax_to_t... | Python | Apache License 2.0 | google/jax | Fix travis; yaml parser does not like comments inside shell commands (#3245) |
260,505 | 29.05.2020 20:58:10 | 14,400 | 0031075bb2d30c95ae4fb29d3457767d65ae07f6 | Replaced jnp.sum by sum when the argument is a list | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/optix.py",
"new_path": "jax/experimental/optix.py",
"diff": "@@ -96,7 +96,7 @@ def clip(max_delta) -> InitUpdate:\ndef global_norm(updates: Updates) -> Updates:\nreturn jnp.sqrt(\n- jnp.sum([jnp.sum(jnp.square(x)) for x in tree_leaves(updates)... | Python | Apache License 2.0 | google/jax | Replaced jnp.sum by sum when the argument is a list (#3253) |
260,411 | 30.05.2020 08:48:44 | -10,800 | d34debafe6b266fbc2482fd9f2dae0805d4fcc18 | Implementation of jax_to_tf.while | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax_to_tf/jax_to_tf.py",
"new_path": "jax/experimental/jax_to_tf/jax_to_tf.py",
"diff": "@@ -782,6 +782,68 @@ def _cond(pred: TfVal, *operands: TfVal,\ntf_impl[lax.cond_p] = _cond\n+\n+def _while(*args, cond_nconsts: int, cond_jaxpr: core.Type... | Python | Apache License 2.0 | google/jax | Implementation of jax_to_tf.while (#3241) |
260,485 | 01.06.2020 23:43:43 | 14,400 | 7a4b222387abb549b8629ac6815c093a456a5b9d | Added support for np.diagflat | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/__init__.py",
"new_path": "jax/numpy/__init__.py",
"diff": "@@ -28,7 +28,7 @@ from .lax_numpy import (\ncomplex128, complex64, complex_, complexfloating, compress, concatenate,\nconj, conjugate, convolve, copysign, corrcoef, correlate, cos, cosh,\nco... | Python | Apache License 2.0 | google/jax | Added support for np.diagflat (#3259) |
260,411 | 02.06.2020 12:35:28 | -10,800 | d36429b5fd992cb16081f44dfd787f28c296e0a8 | Implement jax_to_tf.scan
Also removed the enable_jit, which was needed only
to work around the lack of control flow primitive support. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax_to_tf/__init__.py",
"new_path": "jax/experimental/jax_to_tf/__init__.py",
"diff": "# See the License for the specific language governing permissions and\n# limitations under the License.\n-from .jax_to_tf import enable_jit, convert\n+from ... | Python | Apache License 2.0 | google/jax | Implement jax_to_tf.scan (#3260)
Also removed the enable_jit, which was needed only
to work around the lack of control flow primitive support. |
260,335 | 02.06.2020 17:37:20 | 25,200 | c42a7f7890205bbeacd5511161b7af04eb512417 | remove some trailing whitespace | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/linalg.py",
"new_path": "jax/numpy/linalg.py",
"diff": "@@ -170,7 +170,7 @@ def _cofactor_solve(a, b):\nIf a is rank n-1, then the lower right corner of u will be zero and the\ntriangular_solve will fail.\nLet x = solve(p @ l, b) and y = det(a)*solve... | Python | Apache License 2.0 | google/jax | remove some trailing whitespace (#3287) |
260,335 | 02.06.2020 20:28:21 | 25,200 | 538691b9f4b4edc4457b50e19a56d0cbd3bbc68d | remove `pack` from optimizers.py
It is vestigial, from a time when JaxTuples roamed free. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/optimizers.py",
"new_path": "jax/experimental/optimizers.py",
"diff": "@@ -88,12 +88,11 @@ zip = safe_zip\n# Since pytrees can be flattened, that structure is isomorphic to a list of\n# lists (with no further nesting).\n-pack = tuple\nOptimize... | Python | Apache License 2.0 | google/jax | remove `pack` from optimizers.py (#3305)
It is vestigial, from a time when JaxTuples roamed free. |
260,335 | 02.06.2020 20:28:59 | 25,200 | b58eec51ac28fcf49093adf8092726597ef6cfb8 | make pmap axis checking an exception, hoist | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -1116,6 +1116,9 @@ def pmap(fun: Callable, axis_name: Optional[AxisName] = None, *, in_axes=0,\ndonate_tuple = rebase_donate_argnums(_ensure_tuple(donate_argnums),\nstatic_broadcasted_tuple)\n+ if any(axis != ... | Python | Apache License 2.0 | google/jax | make pmap axis checking an exception, hoist (#3239) |
260,335 | 03.06.2020 07:32:44 | 25,200 | 0e229e4ea8c7361e1a72c2b15b9f18d9829c0a03 | keep old name 'packed_state' of OptimizerState | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/optimizers.py",
"new_path": "jax/experimental/optimizers.py",
"diff": "@@ -89,10 +89,10 @@ zip = safe_zip\n# lists (with no further nesting).\nOptimizerState = namedtuple(\"OptimizerState\",\n- [\"states_flat\", \"tree_def\", \"subtree_defs\"]... | Python | Apache License 2.0 | google/jax | keep old name 'packed_state' of OptimizerState |
260,519 | 04.06.2020 03:26:35 | -36,000 | b998044ffed2f36f683c9dce4fbd2914d5b7de44 | Add np.polyadd | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.numpy.rst",
"new_path": "docs/jax.numpy.rst",
"diff": "@@ -211,6 +211,7 @@ Not every function in NumPy is implemented; contributions are welcome!\npackbits\npad\npercentile\n+ polyadd\npolyval\npower\npositive\n"
},
{
"change_type": "MODIFY",
... | Python | Apache License 2.0 | google/jax | Add np.polyadd (#3261) |
260,411 | 04.06.2020 09:41:45 | -10,800 | afa9276f0869305afe12cbaec88fe3fb535de807 | Implement jax_to_tf.scan | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax_to_tf/jax_to_tf.py",
"new_path": "jax/experimental/jax_to_tf/jax_to_tf.py",
"diff": "@@ -29,6 +29,7 @@ from jax import numpy as jnp\nfrom jax import tree_util\nfrom jax import util\nfrom jax.api_util import flatten_fun\n+from jax.lax impor... | Python | Apache License 2.0 | google/jax | Implement jax_to_tf.scan (#3307) |
260,299 | 04.06.2020 11:25:30 | -3,600 | c04dea1c84b657d014412a04a5312e7e525b7501 | Begin implementing mask(jit) | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/masking.py",
"new_path": "jax/interpreters/masking.py",
"diff": "@@ -26,7 +26,7 @@ from .. import abstract_arrays\nfrom .. import core\nfrom ..tree_util import tree_unflatten\nfrom ..core import Trace, Tracer\n-from ..util import safe_map, saf... | Python | Apache License 2.0 | google/jax | Begin implementing mask(jit) |
260,299 | 04.06.2020 12:28:38 | -3,600 | dfe3462746d701b08f3d1ee814534f228d2fa199 | Add device_put_p abstract_eval rule | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/xla.py",
"new_path": "jax/interpreters/xla.py",
"diff": "@@ -1202,6 +1202,7 @@ device_put_p = core.Primitive('device_put')\ndevice_put_p.def_impl(_device_put_impl)\npe.custom_partial_eval_rules[device_put_p] = lambda trace, x, **params: x\nad.... | Python | Apache License 2.0 | google/jax | Add device_put_p abstract_eval rule |
260,299 | 04.06.2020 12:50:47 | -3,600 | 0f0032727b68e3fc07164c3505ad5c80c9c08503 | Implement MaskTrace.post_process_call | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/masking.py",
"new_path": "jax/interpreters/masking.py",
"diff": "@@ -410,7 +410,12 @@ class MaskTrace(Trace):\nreturn [MaskTracer(self, v, s) for v, s in zip(vals_out, shapes_out())]\ndef post_process_call(self, call_primitive, out_tracers, pa... | Python | Apache License 2.0 | google/jax | Implement MaskTrace.post_process_call |
260,299 | 04.06.2020 15:05:14 | -3,600 | 38d483737d0e35f77ed79fc279574b4b2dc46937 | Fix x64 test | [
{
"change_type": "MODIFY",
"old_path": "tests/masking_test.py",
"new_path": "tests/masking_test.py",
"diff": "@@ -413,9 +413,9 @@ class MaskingTest(jtu.JaxTestCase):\n@jit\ndef concat(y):\nreturn lax.concatenate([x, y], 0)\n- return concat(jnp.array([1., 2., 3.]))\n+ return concat(jnp.array([1., 2.,... | Python | Apache License 2.0 | google/jax | Fix x64 test |
260,335 | 04.06.2020 10:13:15 | 25,200 | 9c0a58a8e774171e5e465bd81d2fad481c5264fc | add float dtype checks to random.py
fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/random.py",
"new_path": "jax/random.py",
"diff": "@@ -359,6 +359,9 @@ def uniform(key: jnp.ndarray,\nReturns:\nA random array with the specified shape and dtype.\n\"\"\"\n+ if not dtypes.issubdtype(dtype, np.floating):\n+ raise ValueError(f\"dtype argument... | Python | Apache License 2.0 | google/jax | add float dtype checks to random.py (#3320)
fixes #3317 |
260,485 | 04.06.2020 23:27:29 | 14,400 | 969ed6d162e6d92ec691cb5e07edee85d4aa0b03 | Initial implementation of polymul function | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.numpy.rst",
"new_path": "docs/jax.numpy.rst",
"diff": "@@ -212,6 +212,7 @@ Not every function in NumPy is implemented; contributions are welcome!\npad\npercentile\npolyadd\n+ polymul\npolyval\npower\npositive\n"
},
{
"change_type": "MODIFY",
... | Python | Apache License 2.0 | google/jax | Initial implementation of polymul function (#3303) |
260,285 | 05.06.2020 09:04:22 | -7,200 | ea782220c4a95f85933b37955d597ab73783ab22 | Allow mask(split) | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -1353,17 +1353,19 @@ def broadcast_to(arr, shape):\n@_wraps(np.split)\ndef split(ary, indices_or_sections, axis=0):\n- dummy_val = np.broadcast_to(0, ary.shape) # zero strides\n+ axis =... | Python | Apache License 2.0 | google/jax | Allow mask(split) |
260,287 | 05.06.2020 17:22:55 | -7,200 | 74d160f5e0772b05c249ed6a7d71a846719c394c | Don't keep primal arguments and results in the linearized jaxpr
Linearized functions are supposed to take tangent types to tangent
types, and so all primal arguments are unused and primal results get
replaced by units. | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -1476,9 +1476,7 @@ def _lift_linearized(jaxpr, primal_avals, consts, io_tree, out_pvals, *py_args):\nmsg = (\"linearized function called on tangent values inconsistent with \"\n\"the original primal values.\")... | Python | Apache License 2.0 | google/jax | Don't keep primal arguments and results in the linearized jaxpr (#3233)
Linearized functions are supposed to take tangent types to tangent
types, and so all primal arguments are unused and primal results get
replaced by units. |
260,287 | 27.05.2020 13:57:47 | 0 | adb442eb8ab7e126f34abf6571477a882d7d7a9a | Make ad_util.zero a class that carries avals (similar to UndefinedPrimal)
This is useful for remat transpose rule submitted in and e.g.
allowed me to catch a slight overuse of defjvp2 for `random_gamma_p` (it
was unnecessarily declared as having multiple outputs). | [
{
"change_type": "MODIFY",
"old_path": "jax/ad_util.py",
"new_path": "jax/ad_util.py",
"diff": "from .core import (lattice_join, Primitive, Unit, unit, AbstractUnit,\n- valid_jaxtype)\n+ valid_jaxtype, raise_to_shaped, get_aval)\nfrom .tree_util import register_pytree_node\nfrom typing import Any, D... | Python | Apache License 2.0 | google/jax | Make ad_util.zero a class that carries avals (similar to UndefinedPrimal)
This is useful for remat transpose rule submitted in #3162 and e.g.
allowed me to catch a slight overuse of defjvp2 for `random_gamma_p` (it
was unnecessarily declared as having multiple outputs). |
260,287 | 27.05.2020 18:09:35 | 0 | c5d870738c571d75eacd92b14540b2705147b346 | Fix host_callback | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/host_callback.py",
"new_path": "jax/experimental/host_callback.py",
"diff": "@@ -389,7 +389,7 @@ id_tap_p.def_abstract_eval(_id_tap_abstract_eval)\n# The AttributeError is for regular values, the KeyError is for ConcreteArray\ndef _instantiate... | Python | Apache License 2.0 | google/jax | Fix host_callback |
260,287 | 28.05.2020 13:20:56 | 0 | 3f1d3a73ace426540f81167fbda9f2068898075a | Remove example from ad.instantiate_zeros, fix vmap bug | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -1881,7 +1881,7 @@ def defjvp_all(fun, custom_jvp):\nmsg = (\"Detected differentiation with respect to closed-over values with \"\n\"custom JVP rule, which isn't supported.\")\nraise ValueError(msg)\n- args_do... | Python | Apache License 2.0 | google/jax | Remove example from ad.instantiate_zeros, fix vmap bug |
260,287 | 29.05.2020 09:43:08 | 0 | 17472b97ab2d6953f953d1f339ed9ee8d4ad5424 | Optimize zeros_like_shaped_array
This function is used a lot more now, because `ad.instantiate_zeros` now
goes through that and not `zeros_like_array`. | [
{
"change_type": "MODIFY",
"old_path": "jax/abstract_arrays.py",
"new_path": "jax/abstract_arrays.py",
"diff": "@@ -37,7 +37,7 @@ def make_shaped_array(x):\ndef zeros_like_array(x):\ndtype = dtypes.canonicalize_dtype(dtypes.result_type(x))\n- return np.broadcast_to(np.array(0, dtype), np.shape(x))\n... | Python | Apache License 2.0 | google/jax | Optimize zeros_like_shaped_array
This function is used a lot more now, because `ad.instantiate_zeros` now
goes through that and not `zeros_like_array`. |
260,519 | 06.06.2020 02:44:10 | -36,000 | 29740de63ecd35073e66f3502f1d7dfe79d990e7 | Add np.polysub | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.numpy.rst",
"new_path": "docs/jax.numpy.rst",
"diff": "@@ -213,6 +213,7 @@ Not every function in NumPy is implemented; contributions are welcome!\npercentile\npolyadd\npolymul\n+ polysub\npolyval\npower\npositive\n"
},
{
"change_type": "MODIFY... | Python | Apache License 2.0 | google/jax | Add np.polysub (#3319) |
260,335 | 06.06.2020 21:44:14 | 25,200 | fd886b17a48f23637f3fcae810117721f155e16c | make jnp.array faster
fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -35,20 +35,21 @@ import warnings\nimport numpy as np\nimport opt_einsum\n-from jax import jit, device_put, custom_jvp\n+from jax import jit, custom_jvp\nfrom .vectorize import vectorize... | Python | Apache License 2.0 | google/jax | make jnp.array faster (#3350)
fixes #2919 |
260,411 | 07.06.2020 14:45:15 | -10,800 | 7863f817c7a853b4968f1f2635aab01d89aa2321 | Commented-out the literal jaxpr checks in host_callback
* Commented-out the literal jaxpr checks in host_callback
Will re-enable when we change the host_callback, or when we have better
tools for updating goldens | [
{
"change_type": "MODIFY",
"old_path": "tests/host_callback_test.py",
"new_path": "tests/host_callback_test.py",
"diff": "@@ -132,20 +132,8 @@ class HostCallbackTest(jtu.JaxTestCase):\nxla_bridge.get_backend.cache_clear()\ndef test_eval(self):\n- assertMultiLineStrippedEqual(self, \"\"\"\n-{ lambda ... | Python | Apache License 2.0 | google/jax | Commented-out the literal jaxpr checks in host_callback (#3351)
* Commented-out the literal jaxpr checks in host_callback
Will re-enable when we change the host_callback, or when we have better
tools for updating goldens |
260,335 | 08.06.2020 09:47:32 | 25,200 | 508821cc184155b86c10850dcbb61c81daeee3e4 | fix a one-character issue from a bad merge
cf. | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax_control_flow.py",
"new_path": "jax/lax/lax_control_flow.py",
"diff": "@@ -776,7 +776,7 @@ def _cond_batching_rule(args, dims, branches, linear):\nreturn out, out_dims\ndef _cond_jvp(primals, tangents, branches, linear):\n- nonzeros = [type(t) is no... | Python | Apache License 2.0 | google/jax | fix a one-character issue from a bad merge (#3362)
cf. #3222 |
260,411 | 08.06.2020 19:59:25 | -10,800 | 65d95f10eaea6f29910e77bee87b5b84c4d6f277 | A couple of ad_util.zero were missed in | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -1905,7 +1905,7 @@ def defjvp(fun, *jvprules):\nans = fun(*primals)\ntangents_out = [rule(t, ans, *primals) for rule, t in zip(jvprules, tangents)\nif rule is not None and type(t) is not ad_util.Zero]\n- retur... | Python | Apache License 2.0 | google/jax | A couple of ad_util.zero were missed in #3222 (#3363) |
260,519 | 09.06.2020 03:06:20 | -36,000 | b1adef40d49d4c8e995f2468af2e07fa13fb0690 | Fix polyadd and test | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -2626,13 +2626,11 @@ def polyadd(a, b):\na = asarray(a)\nb = asarray(b)\n- shape_diff = a.shape[0] - b.shape[0]\n- if shape_diff > 0:\n- b = concatenate((np.zeros(shape_diff, dtype=b.dt... | Python | Apache License 2.0 | google/jax | Fix polyadd and test (#3344) |
260,335 | 08.06.2020 11:48:58 | 25,200 | 982f86702e55c33e063e02285dad77376ceebfda | clean up handling of aux data in jvp_subtrace_aux
related to indirectly, in that we now won't try to do something
crazy like `JVPTracer(trace, object(), zero)`, which didn't like | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/ad.py",
"new_path": "jax/interpreters/ad.py",
"diff": "@@ -74,10 +74,10 @@ def jvp_subtrace_aux(master, primals, tangents):\nassert x._trace.level < trace.level\nans, aux = yield map(partial(JVPTracer, trace), primals, tangents), {}\nans_trace... | Python | Apache License 2.0 | google/jax | clean up handling of aux data in jvp_subtrace_aux
related to #3222 indirectly, in that we now won't try to do something
crazy like `JVPTracer(trace, object(), zero)`, which #3222 didn't like |
260,335 | 08.06.2020 13:16:19 | 25,200 | fb8b3a4df9c2746fb1768d1175511f59eef0c802 | symbolic zeros tweak to work with div rule | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/ad.py",
"new_path": "jax/interpreters/ad.py",
"diff": "@@ -144,7 +144,8 @@ def backward_pass(jaxpr: core.Jaxpr, consts, primals_in, cotangents_in):\ndef write_cotangent(v, ct):\n# assert v not in primal_env\n- if ct is not None and type(v) is ... | Python | Apache License 2.0 | google/jax | symbolic zeros tweak to work with div rule |
260,335 | 08.06.2020 13:22:13 | 25,200 | 866c17c32ec10d46c79ccaf3069189e776b3536f | fix a couple ad_util.Zero type checks | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -2871,7 +2871,7 @@ def _concatenate_translation_rule(c, *operands, **kwargs):\ndef _concatenate_transpose_rule(t, *operands, dimension):\noperand_shapes = [o.aval.shape if ad.is_undefined_primal(o) els... | Python | Apache License 2.0 | google/jax | fix a couple ad_util.Zero type checks |
260,335 | 08.06.2020 13:30:00 | 25,200 | 0a716978adb67194a7ce5c4f373e1f60f14c9b9a | fix a docs bug | [
{
"change_type": "MODIFY",
"old_path": "docs/notebooks/How_JAX_primitives_work.ipynb",
"new_path": "docs/notebooks/How_JAX_primitives_work.ipynb",
"diff": "\" else:\\n\",\n\" # This use of multiply_add is with a constant \\\"y\\\"\\n\",\n\" assert ad.is_undefined_primal(x)\\n\",\n- \" ct_x = ad.zero... | Python | Apache License 2.0 | google/jax | fix a docs bug |
260,335 | 08.06.2020 13:46:10 | 25,200 | 04191ab9de7ed262e4b2b3fd885435e272ae157e | improve broadcast handling in batching.py
We can avoid a circular import just by `import jax`! | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/batching.py",
"new_path": "jax/interpreters/batching.py",
"diff": "import numpy as onp\nfrom typing import Any, Callable, Dict, Optional, Tuple, Union\n+import jax\nfrom .. import core\nfrom ..core import Trace, Tracer, new_master\nfrom ..abst... | Python | Apache License 2.0 | google/jax | improve broadcast handling in batching.py
We can avoid a circular import just by `import jax`! |
260,335 | 08.06.2020 14:13:01 | 25,200 | 2a6d3f417c8fba37e109099d1618378b50c7c7b8 | fix another docs bug | [
{
"change_type": "MODIFY",
"old_path": "docs/notebooks/How_JAX_primitives_work.ipynb",
"new_path": "docs/notebooks/How_JAX_primitives_work.ipynb",
"diff": "\"\\n\",\n\" In our case, multiply_add is not a linear primitive. However, it is used linearly \\n\",\n\" w.r.t. tangents in multiply_add_value_... | Python | Apache License 2.0 | google/jax | fix another docs bug |
260,335 | 08.06.2020 15:06:00 | 25,200 | ee428008c4bdc271a5c962a5b315a2f45056a290 | yet another doc fix | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -830,7 +830,7 @@ def vmap(fun: Callable, in_axes=0, out_axes=0) -> Callable:\nacross the mapped axis:\n>>> print(vmap(lambda x, y: (x + y, y * 2.), in_axes=(0, None), out_axes=0)(np.arange(2.), 4.))\n- (Device... | Python | Apache License 2.0 | google/jax | yet another doc fix |
260,411 | 09.06.2020 12:28:03 | -10,800 | 67927b071e1c6c0ded453761ca8ebd0e043df731 | Fix imports for jax_to_tf tests | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax_to_tf/tests/tf_ops_test.py",
"new_path": "jax/experimental/jax_to_tf/tests/tf_ops_test.py",
"diff": "@@ -27,13 +27,14 @@ import numpy as np\nimport tensorflow as tf # type: ignore[import]\nfrom jax.experimental import jax_to_tf\n-from . im... | Python | Apache License 2.0 | google/jax | Fix imports for jax_to_tf tests (#3377) |
260,335 | 09.06.2020 15:19:53 | 25,200 | 0011fd5e9429fc463e585737c0d69ca0b03cb08f | fix defjvps with None as first rule
fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/custom_derivatives.py",
"new_path": "jax/custom_derivatives.py",
"diff": "@@ -75,7 +75,7 @@ def _initial_style_jaxpr(fun, in_avals):\ntyped_jaxpr = core.TypedJaxpr(jaxpr, consts, in_avals, out_avals)\nreturn typed_jaxpr\n-def sum_tangents(x, *xs):\n+def su... | Python | Apache License 2.0 | google/jax | fix defjvps with None as first rule
fixes #3389 |
260,335 | 09.06.2020 19:20:15 | 25,200 | 650fb494ad9f955ac8ddc979b1bdd8e64faf57fd | tweak t logpdf tolerance for float64 | [
{
"change_type": "MODIFY",
"old_path": "tests/scipy_stats_test.py",
"new_path": "tests/scipy_stats_test.py",
"diff": "@@ -373,7 +373,8 @@ class LaxBackedScipyStatsTests(jtu.JaxTestCase):\nself._CheckAgainstNumpy(scipy_fun, lax_fun, args_maker, check_dtypes=False,\ntol=1e-3)\n- self._CompileAndCheck(... | Python | Apache License 2.0 | google/jax | tweak t logpdf tolerance for float64 |
260,411 | 10.06.2020 08:13:40 | -10,800 | b3c348cc07ec31b560532a48c0c759263341b003 | Moved shift_right implementation from tfxla to jax_to_tf
* Implemented shift_right without tfxla
These ops don't actually need XLA, they should not depend on tfxla.
* Small fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax_to_tf/jax_to_tf.py",
"new_path": "jax/experimental/jax_to_tf/jax_to_tf.py",
"diff": "@@ -175,7 +175,7 @@ class TensorFlowTracer(core.Tracer):\nself._trace = trace\nif not isinstance(val, (tf.Tensor, tf.Variable)):\naval = xla.abstractify(v... | Python | Apache License 2.0 | google/jax | Moved shift_right implementation from tfxla to jax_to_tf (#3378)
* Implemented shift_right without tfxla
These ops don't actually need XLA, they should not depend on tfxla.
* Small fixes |
260,285 | 10.06.2020 11:47:47 | -7,200 | 98a32f2c5999c69d9367bf2bd9dcaacaaab114c2 | Remove special case for scan masking
The scan masking rule is now simplified to a standard masking rule, and the special case handling in masking.py has been removed. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/masking.py",
"new_path": "jax/interpreters/masking.py",
"diff": "@@ -33,7 +33,6 @@ from .. import linear_util as lu\nmap = safe_map\nzip = safe_zip\n-shape_parameterized_primitive_rules: Dict[core.Primitive, Callable] = {}\nmasking_rules: Dict... | Python | Apache License 2.0 | google/jax | Remove special case for scan masking
The scan masking rule is now simplified to a standard masking rule, and the special case handling in masking.py has been removed. |
260,285 | 10.06.2020 12:56:26 | -7,200 | c39b6a40652218fbd67923ea57499a6cfc2d75fa | Remove unused scan shape rule | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax_control_flow.py",
"new_path": "jax/lax/lax_control_flow.py",
"diff": "@@ -1474,13 +1474,6 @@ def _scan_batching_rule(args, dims, reverse, length, jaxpr, num_consts,\nys_bdims = [1 if b else batching.not_mapped for b in ys_batched]\nreturn outs, car... | Python | Apache License 2.0 | google/jax | Remove unused scan shape rule |
260,519 | 11.06.2020 02:57:35 | -36,000 | 832431db9eea95f2212370b47bdd0e6e891fcd0a | Add np.triu_indices_from function | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.numpy.rst",
"new_path": "docs/jax.numpy.rst",
"diff": "@@ -268,8 +268,10 @@ Not every function in NumPy is implemented; contributions are welcome!\ntri\ntril\ntril_indices\n+ tril_indices_from\ntriu\ntriu_indices\n+ triu_indices_from\ntrue_divide\ntru... | Python | Apache License 2.0 | google/jax | Add np.triu_indices_from function (#3346) |
260,285 | 11.06.2020 07:08:12 | -7,200 | dfd2d8c564b304343090b85643c1f2abcf2515fa | Fix dtype, minor | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/masking.py",
"new_path": "jax/interpreters/masking.py",
"diff": "@@ -23,7 +23,7 @@ from typing import Callable, Dict, Sequence, Union\nimport numpy as onp\nfrom .. import abstract_arrays\n-from .. import core\n+from .. import core, dtypes\nfro... | Python | Apache License 2.0 | google/jax | Fix dtype, minor |
260,335 | 11.06.2020 14:21:02 | 25,200 | 1a433620a3c98ec6f49a9f39eb71730987b9003e | make jnp.array(x, copy=True) copies device buffers
fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -43,7 +43,7 @@ from .. import dtypes\nfrom ..abstract_arrays import UnshapedArray, ShapedArray, ConcreteArray, canonicalize_shape\nfrom ..config import flags\nfrom ..interpreters.xla im... | Python | Apache License 2.0 | google/jax | make jnp.array(x, copy=True) copies device buffers
fixes #3412 |
260,314 | 12.06.2020 01:42:25 | 14,400 | b680c994ae0f07f4fdfb482177ef82857ee97a49 | allow scalar input in poisson sampler | [
{
"change_type": "MODIFY",
"old_path": "jax/random.py",
"new_path": "jax/random.py",
"diff": "@@ -1186,7 +1186,7 @@ def poisson(key, lam, shape=(), dtype=np.int64):\nshape = abstract_arrays.canonicalize_shape(shape)\nif np.shape(lam) != shape:\nlam = jnp.broadcast_to(lam, shape)\n- lam = lam.astype(... | Python | Apache License 2.0 | google/jax | allow scalar input in poisson sampler |
260,287 | 12.06.2020 15:03:26 | -7,200 | 2dac55a0bfca6911237b43bd8882c9a4bf4ed258 | Skip known invars and outvars in JaxprTrace.process_call | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/masking.py",
"new_path": "jax/interpreters/masking.py",
"diff": "@@ -425,6 +425,9 @@ class MaskTrace(Trace):\n# Make padded_env hashable\npadded_env = (env_keys, padded_env_vals)\nf, shapes_out = mask_subtrace(f, self.master, shapes, padded_en... | Python | Apache License 2.0 | google/jax | Skip known invars and outvars in JaxprTrace.process_call (#3242) |
260,335 | 12.06.2020 15:41:07 | 25,200 | ae9df752de75344d231c7ec973fdd562a91d8913 | add docstring to ravel_pytree | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.rst",
"new_path": "docs/jax.rst",
"diff": "@@ -17,6 +17,7 @@ Subpackages\njax.ops\njax.random\njax.tree_util\n+ jax.flatten_util\njax.dlpack\njax.profiler\n"
},
{
"change_type": "MODIFY",
"old_path": "jax/flatten_util.py",
"new_path": ... | Python | Apache License 2.0 | google/jax | add docstring to ravel_pytree |
260,335 | 14.06.2020 13:56:53 | 25,200 | 482067640578a40f088e5556a702090d12c26d5a | add jax.numpy.concatenate(..., axis=None) support
fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -1995,6 +1995,8 @@ def concatenate(arrays, axis=0):\nraise ValueError(\"Need at least one array to concatenate.\")\nif ndim(arrays[0]) == 0:\nraise ValueError(\"Zero-dimensional arrays ... | Python | Apache License 2.0 | google/jax | add jax.numpy.concatenate(..., axis=None) support
fixes #3419 |
260,335 | 14.06.2020 14:45:29 | 25,200 | 29fa935ca56dd6aa8fc688a30f860c300fe93bd6 | fix vmap-of-pmap mapped_invars logic bug
fixes
This crept in via but more importantly it shows we don't have
good test coverage here! | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/batching.py",
"new_path": "jax/interpreters/batching.py",
"diff": "@@ -161,10 +161,12 @@ class BatchTrace(Trace):\nif all(dim is not_mapped for dim in dims):\nreturn map_primitive.bind(f, *vals, **params)\nelse:\n+ mapped_invars = params['mapp... | Python | Apache License 2.0 | google/jax | fix vmap-of-pmap mapped_invars logic bug
fixes #3399
This crept in via #1959, but more importantly it shows we don't have
good test coverage here! |
260,411 | 15.06.2020 10:05:23 | -10,800 | 0e804296766384763bfbb8cd6e2758b623d919ef | Added jax2tf test about primitive coverage | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -23,15 +23,21 @@ import jax\nfrom jax import abstract_arrays\nfrom jax import ad_util\nfrom jax import core\n+from jax import custom_derivatives\nfrom jax import l... | Python | Apache License 2.0 | google/jax | Added jax2tf test about primitive coverage (#3420) |
260,411 | 15.06.2020 10:59:46 | -10,800 | 1440ccf4f9ecd9a487d0ce55e5aa725ecaf0e961 | Fixed bug in argument type promotion for dynamic_update_slice
Also added tests for lax.slice, lax.dynamic_slice and
lax.dynamic_update_slice. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -292,8 +292,8 @@ def promote_types(*values):\ndef wrap_binary_op(func):\n- def wrapped_func(lhs, rhs, *args, **kwargs):\n- return func(*promote_types(lhs, rhs), *a... | Python | Apache License 2.0 | google/jax | Fixed bug in argument type promotion for dynamic_update_slice (#3427)
Also added tests for lax.slice, lax.dynamic_slice and
lax.dynamic_update_slice. |
260,411 | 15.06.2020 12:14:09 | -10,800 | 2e3d4393c218a5cd51e88e9e1fca0b47e587ddaa | [jax2tf] fix the too-early use of tf.constant
If I leave it at top-level, I get test failures about missing platform Host | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -57,14 +57,13 @@ TfVal = Any\n# Whenever we are in a JAX tracing context we must use `core.unit` values\n# in those places. However, when we move to TF we have to ... | Python | Apache License 2.0 | google/jax | [jax2tf] fix the too-early use of tf.constant (#3446)
If I leave it at top-level, I get test failures about missing platform Host |
260,422 | 13.06.2020 11:39:01 | -3,600 | 3c78605bb8c1133bba6b59bb6a973b046018717a | Propagate raw __name__ and __doc__ of functions wrapped by jit and sharded_jit. | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -166,9 +166,6 @@ def jit(fun: Callable, static_argnums: Union[int, Iterable[int]] = (),\nout = xla.xla_call(flat_fun, *args_flat, device=device, backend=backend,\nname=flat_fun.__name__, donated_invars=donated... | Python | Apache License 2.0 | google/jax | Propagate raw __name__ and __doc__ of functions wrapped by jit and sharded_jit. |
260,335 | 15.06.2020 07:32:42 | 25,200 | 12ce6e376872d6e1bdacc0e88984f21f92dc0cc8 | roll back of while we debug internal fails | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/batching.py",
"new_path": "jax/interpreters/batching.py",
"diff": "@@ -161,12 +161,10 @@ class BatchTrace(Trace):\nif all(dim is not_mapped for dim in dims):\nreturn map_primitive.bind(f, *vals, **params)\nelse:\n- mapped_invars = params['mapp... | Python | Apache License 2.0 | google/jax | roll back of #3439 while we debug internal fails |
260,335 | 15.06.2020 09:10:40 | 25,200 | fcfcffe334351d8e79670c48e2aa08cc86426341 | add systematic tests for vmap-of-pmap
fixes
Also re-applies the fix in (i.e it rolls-back the rollback PR because we're now confident it's correct (and some internal tests are buggy). | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/batching.py",
"new_path": "jax/interpreters/batching.py",
"diff": "@@ -161,10 +161,12 @@ class BatchTrace(Trace):\nif all(dim is not_mapped for dim in dims):\nreturn map_primitive.bind(f, *vals, **params)\nelse:\n+ mapped_invars = params['mapp... | Python | Apache License 2.0 | google/jax | add systematic tests for vmap-of-pmap
fixes #3440
Also re-applies the fix in #3439 (i.e it rolls-back the rollback PR #3448) because we're now confident it's correct (and some internal tests are buggy). |
260,700 | 11.05.2020 23:44:32 | 14,400 | 3cf6b1de542bb2b8770af99f236af698eae3db96 | add erf inv rule
erf_inv rule not working
works up to order 2
erf inv rule
use np for now
actually use np for now | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jet.py",
"new_path": "jax/experimental/jet.py",
"diff": "@@ -249,6 +249,49 @@ def def_comp(prim, comp):\ndef_comp(lax.erfc_p, lambda x: 1 - lax.erf(x))\n+\n+def _erf_inv_rule(primals_in, series_in):\n+ x, = primals_in\n+ series, = series_in\n+... | Python | Apache License 2.0 | google/jax | add erf inv rule
erf_inv rule not working
works up to order 2
erf inv rule
use np for now
actually use np for now |
260,700 | 15.06.2020 17:23:57 | 14,400 | f463598f19549cd37386e8d7e7fe9036f52d6f02 | add int pow rule | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jet.py",
"new_path": "jax/experimental/jet.py",
"diff": "@@ -298,10 +298,18 @@ jet_rules[lax.pow_p] = _pow_taylor\ndef _integer_pow_taylor(primals_in, series_in, *, y):\nif y == 2:\n- fn = lambda x: x * x\n- else:\n- fn = lambda x: lax.pow(x, ... | Python | Apache License 2.0 | google/jax | add int pow rule |
260,335 | 12.06.2020 16:10:45 | 25,200 | d4c6cb62abce5b7b5a28a4297ccda27d0a8405d8 | print warning when doing jit-of-pmap | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -169,8 +169,6 @@ def jit(fun: Callable, static_argnums: Union[int, Iterable[int]] = (),\nname=flat_fun.__name__, donated_invars=donated_invars)\nreturn tree_unflatten(out_tree(), out)\n- jitted_name = \"jit({}... | Python | Apache License 2.0 | google/jax | print warning when doing jit-of-pmap |
260,335 | 16.06.2020 11:46:37 | 25,200 | 005958e13ec0d30cd45192673e761e886622db6e | added reviewer suggestion | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/invertible_ad.py",
"new_path": "jax/interpreters/invertible_ad.py",
"diff": "@@ -54,7 +54,7 @@ def _invertible_call_make_output_tracers(trace, in_tracers, out_tracers, params)\nnew_in_tracers = (*in_tracers, *out_consts)\n# Append dummy output... | Python | Apache License 2.0 | google/jax | added reviewer suggestion |
260,335 | 16.06.2020 15:46:51 | 25,200 | 20f4ec649c9cccdac6822302a4c4695192fa9473 | fix a bug from | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/partial_eval.py",
"new_path": "jax/interpreters/partial_eval.py",
"diff": "@@ -704,6 +704,7 @@ def _remat_partial_eval(process_out, trace, _, f, tracers, params):\n# Using the instantiated tracers, run call_bind like JaxprTrace.process_call.\n... | Python | Apache License 2.0 | google/jax | fix a bug from #3459 (#3466) |
260,700 | 16.06.2020 22:48:25 | 14,400 | 575216e094ac55866320ae09dff4b05ef47fb7ea | add jet primitives, refactor tests | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jet.py",
"new_path": "jax/experimental/jet.py",
"diff": "@@ -18,6 +18,7 @@ from functools import partial\nimport numpy as np\nimport jax\n+import jax.numpy as jnp\nfrom jax import core\nfrom jax.util import unzip2\nfrom jax import ad_util\n@@ ... | Python | Apache License 2.0 | google/jax | add jet primitives, refactor tests (#3468)
Co-authored-by: Jesse Bettencourt <jessebett@cs.toronto.edu> |
260,411 | 17.06.2020 11:57:21 | -10,800 | 4f21b9351cd6f4be6d192ddda45c260db38b4380 | [jax2tf] Added special case for tf.pad.
Fixed lax_reference.pad to handle lax.pad with negative edge padding. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -677,6 +677,10 @@ def _pad_shape(operand, padding_value, padding_config):\ndef _pad(operand, padding_value, padding_config):\nlow, high, interior = util.unzip3(pad... | Python | Apache License 2.0 | google/jax | [jax2tf] Added special case for tf.pad. (#3462)
Fixed lax_reference.pad to handle lax.pad with negative edge padding. |
260,519 | 18.06.2020 02:43:50 | -36,000 | 5ee936fdcf44711a2fcd8b1f448a6134c14294a4 | Add polyder numpy function | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.numpy.rst",
"new_path": "docs/jax.numpy.rst",
"diff": "@@ -212,6 +212,7 @@ Not every function in NumPy is implemented; contributions are welcome!\npad\npercentile\npolyadd\n+ polyder\npolymul\npolysub\npolyval\n"
},
{
"change_type": "MODIFY",
... | Python | Apache License 2.0 | google/jax | Add polyder numpy function (#3403) |
260,335 | 17.06.2020 10:05:28 | 25,200 | 5344ec5364bbe6b08bfe4f9e4c0f581b807f0cfe | use original numpy for shape calculations
cf. | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -2253,7 +2253,7 @@ def arange(start, stop=None, step=None, dtype=None):\nlax._check_user_dtype_supported(dtype, \"arange\")\nif stop is None and step is None:\ndtype = dtype or _dtype(s... | Python | Apache License 2.0 | google/jax | use original numpy for shape calculations (#3474)
cf. #3453 |
260,335 | 22.06.2020 08:12:41 | 25,200 | 3fff83790742a1b0a468032ecff6f8809fde57d9 | pin numpy version in setup.py to avoid warnings | [
{
"change_type": "MODIFY",
"old_path": "setup.py",
"new_path": "setup.py",
"diff": "@@ -29,7 +29,7 @@ setup(\npackages=find_packages(exclude=[\"examples\"]),\npython_requires='>=3.6',\ninstall_requires=[\n- 'numpy>=1.12', 'absl-py', 'opt_einsum'\n+ 'numpy >=1.12, <1.19', 'absl-py', 'opt_einsum'\n],\... | Python | Apache License 2.0 | google/jax | pin numpy version in setup.py to avoid warnings (#3509) |
260,335 | 22.06.2020 10:27:44 | 25,200 | a81e732d3d5398ef5cd74eaf518b6cb2d58877db | fix jet typo: jnp not np | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jet.py",
"new_path": "jax/experimental/jet.py",
"diff": "@@ -363,7 +363,7 @@ def _integer_pow_taylor(primals_in, series_in, *, y):\nfor k in range(1, len(v)):\nvu = sum(_scale(k, j) * v[k-j] * u[j] for j in range(1, k + 1))\nuv = sum(_scale(k,... | Python | Apache License 2.0 | google/jax | fix jet typo: jnp not np (#3507) |
260,705 | 22.06.2020 16:31:08 | 10,800 | e5d4ca31a8b59fb93229d3629035007e5aa329cc | Fix typo understanding jaxprs page on readthedocs | [
{
"change_type": "MODIFY",
"old_path": "docs/jaxpr.rst",
"new_path": "docs/jaxpr.rst",
"diff": "@@ -169,7 +169,7 @@ before (with two input vars, one for each element of the input tuple)\nConstant Vars\n--------------\n-ConstVars arise when the computation ontains array constants, either\n+ConstVars ... | Python | Apache License 2.0 | google/jax | Fix typo understanding jaxprs page on readthedocs (#3513) |
260,335 | 22.06.2020 17:50:33 | 25,200 | 2f7108f78ba4d7fd12a8ad6232685d0d10b28c01 | remove the lower_fun default multiple_results=True | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -1853,7 +1853,7 @@ def custom_transforms(fun):\nfun_p.def_abstract_eval(fun_abstract_eval)\ndef fun_translation(c, *xla_args, **params):\n- return xla.lower_fun(fun_impl)(c, *xla_args, **params)\n+ return xla.... | Python | Apache License 2.0 | google/jax | remove the lower_fun default multiple_results=True (#3524) |
260,452 | 22.06.2020 22:43:25 | 14,400 | 046006e047543d5c24f75a930ab87ef56c247032 | Fix typo: np.bool -> np.bool_
Replaced np.bool (which is just bool) with np.bool_, which is numpy's
Boolean type. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/host_callback.py",
"new_path": "jax/experimental/host_callback.py",
"diff": "@@ -729,7 +729,7 @@ _CODE_TO_DTYPE = {\n5: np.dtype(np.uint16),\n6: np.dtype(np.uint32),\n7: np.dtype(np.uint64),\n- 8: np.dtype(np.bool),\n+ 8: np.dtype(np.bool_),\n... | Python | Apache License 2.0 | google/jax | Fix typo: np.bool -> np.bool_ (#3525)
Replaced np.bool (which is just bool) with np.bool_, which is numpy's
Boolean type. |
260,335 | 22.06.2020 20:04:07 | 25,200 | 02494924f9554fb2b4317043319b3731a13fccdd | fix an issue with newer versions of pytype | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/partial_eval.py",
"new_path": "jax/interpreters/partial_eval.py",
"diff": "@@ -838,7 +838,7 @@ def _reconstruct_pval(pval1: PartialVal, const2: core.Value):\nreturn pval1\nelse:\nif type(pv1) is ConcreteArray:\n- return PartialVal.known(pv1.va... | Python | Apache License 2.0 | google/jax | fix an issue with newer versions of pytype (#3526) |
260,597 | 23.06.2020 05:46:41 | -7,200 | 5ee6bc00340e07d1b4fd705bddc2b496cd21f25a | Remove unnecessary static_argnum in np.gradient | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -1006,7 +1006,7 @@ def ediff1d(ary, to_end=None, to_begin=None):\nreturn result\n-@partial(jit, static_argnums=(1, 2))\n+@partial(jit, static_argnums=2)\ndef _gradient(a, varargs, axis)... | Python | Apache License 2.0 | google/jax | Remove unnecessary static_argnum in np.gradient (#3512) |
260,322 | 23.06.2020 14:25:53 | -3,600 | 490c8533c889127200a9d4a7ed283cfbacc96586 | Adds boolean support for bitwise not and unittests for boolean support on logical operations. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -496,6 +496,24 @@ tf_impl[lax.shift_right_logical_p] = _shift_right_logical\ntf_impl[lax.shift_left_p] = tf.bitwise.left_shift\n+def _not(x):\n+ \"\"\"Computes bit... | Python | Apache License 2.0 | google/jax | Adds boolean support for bitwise not and unittests for boolean support on logical operations. (#3483)
Co-authored-by: Thomas Keck <thomaskeck@google.com> |
260,335 | 23.06.2020 12:08:12 | 25,200 | a45e28377f54c5bd3033b989f8eb9fe516a7583d | add back a full_lower, dropped in | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -36,7 +36,7 @@ from . import source_info_util\nfrom .util import safe_zip, safe_map, partial, curry, prod, partialmethod\nfrom .pprint_util import pp, vcat, PrettyPrint\n-# TODO(dougalm): the trace cache bre... | Python | Apache License 2.0 | google/jax | add back a full_lower, dropped in #3491 (#3530) |
260,335 | 23.06.2020 14:03:36 | 25,200 | 74ee2ef6eb522f3de93feacc6b15e2f9a93fb8a7 | avoid value-based error check in random.choice | [
{
"change_type": "MODIFY",
"old_path": "jax/random.py",
"new_path": "jax/random.py",
"diff": "@@ -574,8 +574,6 @@ def choice(key, a, shape=(), replace=True, p=None):\np = jnp.asarray(p)\nif p.shape != (n_inputs,):\nraise ValueError(\"p must be None or match the shape of a\")\n- if jnp.any(p < 0):\n-... | Python | Apache License 2.0 | google/jax | avoid value-based error check in random.choice (#3531) |
260,677 | 23.06.2020 23:36:45 | -3,600 | 9d173c622555898b3825a296f82d166da67b6f46 | Support `b` and `return_sign` in scipy.special.logsumexp | [
{
"change_type": "MODIFY",
"old_path": "jax/scipy/special.py",
"new_path": "jax/scipy/special.py",
"diff": "@@ -101,15 +101,27 @@ expit.defjvps(lambda g, ans, x: g * ans * (lax._const(ans, 1) - ans))\n@_wraps(osp_special.logsumexp)\ndef logsumexp(a, axis=None, b=None, keepdims=False, return_sign=Fal... | Python | Apache License 2.0 | google/jax | Support `b` and `return_sign` in scipy.special.logsumexp (#3488) |
260,597 | 24.06.2020 15:22:35 | -7,200 | a6e3f992a70775ef16c285e024cd2ac62ea32077 | Add np.unwrap | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/__init__.py",
"new_path": "jax/numpy/__init__.py",
"diff": "@@ -56,7 +56,7 @@ from .lax_numpy import (\nsquare, squeeze, stack, std, subtract, sum, swapaxes, take, take_along_axis,\ntan, tanh, tensordot, tile, trace, trapz, transpose, tri, tril, tril... | Python | Apache License 2.0 | google/jax | Add np.unwrap (#3527) |
260,280 | 24.06.2020 11:54:06 | 10,800 | 319eeaf5c97b076cbdc70fb9fa60260cb32416c0 | Future warning about lists and tuples | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -1521,6 +1521,10 @@ def _make_reduction(np_fun, op, init_val, preproc=None, bool_op=None,\nif out is not None:\nraise ValueError(\"reduction does not support the `out` argument.\")\n+ i... | Python | Apache License 2.0 | google/jax | Future warning about lists and tuples (#3369) |
260,519 | 24.06.2020 21:49:16 | -19,080 | a44bc0c2c05aa4a079eda3995379dab4a63182dc | Add np.diag_indices_from | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.numpy.rst",
"new_path": "docs/jax.numpy.rst",
"diff": "@@ -97,6 +97,7 @@ Not every function in NumPy is implemented; contributions are welcome!\ndegrees\ndiag\ndiag_indices\n+ diag_indices_from\ndiagflat\ndiagonal\ndigitize\n"
},
{
"change_typ... | Python | Apache License 2.0 | google/jax | Add np.diag_indices_from (#3500) |
260,705 | 24.06.2020 19:00:24 | 10,800 | 7e5407c734b22aaa73da2bfe9c75edff3d87d37e | Fix typos pytrees page on readthedocs | [
{
"change_type": "MODIFY",
"old_path": "docs/pytrees.rst",
"new_path": "docs/pytrees.rst",
"diff": "@@ -55,8 +55,8 @@ to only map over the dictionary argument, we can use::\n(None, 0) # equivalent to (None, {\"k1\": 0, \"k2\": 0})\n-Or, if want every argument to be mapped, we can simply write a sing... | Python | Apache License 2.0 | google/jax | Fix typos pytrees page on readthedocs (#3548) |
260,335 | 24.06.2020 16:11:26 | 25,200 | 696958d2bd97ecfb83feafc5d70bc9fc32cc8b6d | add remat docstring
* add remat docstring
first part of addressing | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -1792,6 +1792,87 @@ def eval_shape(fun: Callable, *args, **kwargs):\ndef checkpoint(fun: Callable, concrete: bool = False) -> Callable:\n+ \"\"\"Make ``fun`` recompute internal linearization points when differ... | Python | Apache License 2.0 | google/jax | add remat docstring (#3542)
* add remat docstring
first part of addressing #3314 |
260,299 | 25.06.2020 15:50:11 | -3,600 | c9670d50c5015f36d7569bdf204b0651c64e55fb | Fix lazy broadcast issue | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -2776,7 +2776,8 @@ ad.deflinear(broadcast_p, lambda t, sizes: [_reduce_sum(t, range(len(sizes)))])\nbatching.primitive_batchers[broadcast_p] = _broadcast_batch_rule\ndef _broadcast_in_dim_impl(operand,... | Python | Apache License 2.0 | google/jax | Fix lazy broadcast issue (#3536) |
260,335 | 25.06.2020 17:36:17 | 25,200 | db80ca5dd87d27e1800834c2828298aa3f6d8992 | allow closures for odeint dynamics functions
* allow closures for odeint dynamics functions
fixes
* add tests for odeint dynamics closing over tracers | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/ode.py",
"new_path": "jax/experimental/ode.py",
"diff": "@@ -31,15 +31,39 @@ import jax.numpy as jnp\nfrom jax import core\nfrom jax import lax\nfrom jax import ops\n-from jax.util import safe_map, safe_zip\n+from jax.util import safe_map, saf... | Python | Apache License 2.0 | google/jax | allow closures for odeint dynamics functions (#3562)
* allow closures for odeint dynamics functions
fixes #2718, #3557
* add tests for odeint dynamics closing over tracers |
260,335 | 25.06.2020 19:17:24 | 25,200 | 062ce297ddf9056ca7743a2d262b0db070eb4553 | removed stale faq entries | [
{
"change_type": "MODIFY",
"old_path": "docs/faq.rst",
"new_path": "docs/faq.rst",
"diff": "-JAX Frequently Asked Questions\n-==============================\n+JAX Frequently Asked Questions (FAQ)\n+====================================\n.. comment RST primer for Sphinx: https://thomas-cokelaer.info/t... | Python | Apache License 2.0 | google/jax | removed stale faq entries (#3565) |
260,335 | 25.06.2020 20:57:34 | 25,200 | 26c6c3a457414577f7232699095877d6c86d032d | fix error when doing forward-mode of odeint
fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/custom_derivatives.py",
"new_path": "jax/custom_derivatives.py",
"diff": "@@ -590,3 +590,5 @@ batching.primitive_batchers[custom_vjp_call_jaxpr_p] = _custom_vjp_call_jaxpr_vm\nxla.initial_style_translations[custom_vjp_call_jaxpr_p] = \\\nxla.lower_fun_init... | Python | Apache License 2.0 | google/jax | fix error when doing forward-mode of odeint (#3566)
fixes #3558 |
260,676 | 26.06.2020 18:40:00 | -3,600 | 99a43f20db0c35f6da5d2dfc636bd57bd5bfad6e | Added missing is_stable argument to lax.sort | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -1196,8 +1196,8 @@ def cumprod(operand: Array, axis: int) -> Array:\n\"\"\"Computes a cumulative product along `axis`.\"\"\"\nreturn cumprod_p.bind(operand, axis=int(axis))\n-def sort(operand: Union[Ar... | Python | Apache License 2.0 | google/jax | Added missing is_stable argument to lax.sort (#3553) |
260,335 | 26.06.2020 11:44:16 | 25,200 | 11caa21eca1634a6c4cd7cd1b94a25110c7929aa | ensure lax.reduce monoid test uses original numpy | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -1054,17 +1054,17 @@ def _get_monoid_reducer(monoid_op: Callable, x: Array) -> Optional[Callable]:\ndtype = _dtype(x)\nif (type(aval) is ConcreteArray) and aval.shape == ():\nif monoid_op is add:\n- re... | Python | Apache License 2.0 | google/jax | ensure lax.reduce monoid test uses original numpy (#3573) |
260,444 | 27.06.2020 00:19:57 | -7,200 | 42cbe49ce648f909eea9b69e2f70ae33a59ea570 | Correction a typo of the period of the PRNG. | [
{
"change_type": "MODIFY",
"old_path": "docs/notebooks/Common_Gotchas_in_JAX.ipynb",
"new_path": "docs/notebooks/Common_Gotchas_in_JAX.ipynb",
"diff": "\"id\": \"ORMVVGZJgSVi\"\n},\n\"source\": [\n- \"Underneath the hood, numpy uses the [Mersenne Twister](https://en.wikipedia.org/wiki/Mersenne_Twist... | Python | Apache License 2.0 | google/jax | Correction a typo of the period of the PRNG. (#3578) |
260,411 | 27.06.2020 14:55:28 | -10,800 | 496cde6ebcce844f6b9321ce7d73c8fbdd96532d | [jax2tf] Add special case for translation of lax.gather to tf.gather.
Also adds more tests for conversion for gather. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -902,19 +902,58 @@ def _gather_shape(operand, start_indices, dimension_numbers, slice_sizes):\nreturn out.shape\n+def _try_tf_gather(operand, start_indices, dimens... | Python | Apache License 2.0 | google/jax | [jax2tf] Add special case for translation of lax.gather to tf.gather. (#3486)
Also adds more tests for conversion for gather. |
260,530 | 28.06.2020 12:11:12 | 14,400 | 7b57dc8c8043163a5e649ba66143ccef880d7d58 | Issue1635 expm frechet
* Implement Frechet derivatives for expm.
* Update expm to use the current custom gradients API.
Make some stylistic fixes. | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.scipy.rst",
"new_path": "docs/jax.scipy.rst",
"diff": "@@ -16,6 +16,7 @@ jax.scipy.linalg\ndet\neigh\nexpm\n+ expm_frechet\ninv\nlu\nlu_factor\n"
},
{
"change_type": "MODIFY",
"old_path": "jax/scipy/linalg.py",
"new_path": "jax/scipy/l... | Python | Apache License 2.0 | google/jax | Issue1635 expm frechet (#2062)
* Implement Frechet derivatives for expm.
* Update expm to use the current custom gradients API.
Make some stylistic fixes.
Co-authored-by: Peter Hawkins <phawkins@google.com> |
260,411 | 29.06.2020 12:48:27 | -10,800 | 6147026b81bc1a2754cc40bd1de58961a8929c4a | [jax2tf] Add support for custom JVP/VJP
* [jax2tf] Add support for custom JVP/VJP
The custom VJP/JVP in initial style control-flow primitives make
use of special custom_jvp_call_jaxpr primitives
* Fix flake | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -80,13 +80,20 @@ def _tfval_remove_unit(args: Sequence[TfValOrUnit]) -> Sequence[TfVal]:\ndef _tfval_add_unit(vals: Sequence[TfValOrUnit],\navals: Sequence[core.Ab... | Python | Apache License 2.0 | google/jax | [jax2tf] Add support for custom JVP/VJP (#3581)
* [jax2tf] Add support for custom JVP/VJP
The custom VJP/JVP in initial style control-flow primitives make
use of special custom_jvp_call_jaxpr primitives
* Fix flake |
260,411 | 30.06.2020 10:12:22 | -10,800 | 59344028c85ed0c8fad20b052275b572c2d60785 | [jax2tf] Fix tf.gather handling of out-of-bounds indices
JAX and XLA clamp out-of-bounds indices, while tf.gather aborts. We
ensure that when we rewrite lax.gather to tf.gather, we use XLA. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/jax2tf/jax2tf.py",
"new_path": "jax/experimental/jax2tf/jax2tf.py",
"diff": "@@ -938,9 +938,14 @@ def _try_tf_gather(operand, start_indices, dimension_numbers, slice_sizes):\nreturn None\n# TODO: should we allow ourselves to add a reshape, or ... | Python | Apache License 2.0 | google/jax | [jax2tf] Fix tf.gather handling of out-of-bounds indices (#3604)
JAX and XLA clamp out-of-bounds indices, while tf.gather aborts. We
ensure that when we rewrite lax.gather to tf.gather, we use XLA. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.