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 |
|---|---|---|---|---|---|---|---|---|---|
260,299 | 27.06.2019 14:13:20 | -3,600 | 31fa0412ea1fd4d1c60319c6e48ba69d02673ad5 | Used shaped aval for custom_transforms jaxpr | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -984,8 +984,7 @@ class CustomTransformsFunction(object):\ndef __call__(self, *args, **kwargs):\ndef pv_like(x):\n- aval = x.aval if hasattr(x, 'aval') else xla.abstractify(x)\n- return pe.PartialVal((aval, cor... | Python | Apache License 2.0 | google/jax | Used shaped aval for custom_transforms jaxpr |
260,299 | 27.06.2019 15:35:12 | -3,600 | 323d9f51dc0b7d67f8ee427882be8607c81aedc5 | Raise error when differentiating w.r.t. outer variable with defjvp_all | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -1118,7 +1118,13 @@ def defjvp_all(fun, custom_jvp):\n_check_custom_transforms_type(\"defjvp_all\", fun)\ndef custom_transforms_jvp(primals, tangents, **params):\nconsts, jax_kwargs, jax_args = primals[0], pri... | Python | Apache License 2.0 | google/jax | Raise error when differentiating w.r.t. outer variable with defjvp_all |
260,299 | 27.06.2019 17:35:34 | -3,600 | f76a1c963960edd11662b4cf70ae64410e3f26b4 | Add out of scope error to defvjp | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -1263,7 +1263,13 @@ def defvjp_all(fun, custom_vjp):\n(4.0, 3.0)\n\"\"\"\n_check_custom_transforms_type(\"defvjp_all\", fun)\n- def custom_transforms_vjp(consts, jax_kwargs, *jax_args, **params):\n+ def custom... | Python | Apache License 2.0 | google/jax | Add out of scope error to defvjp |
260,299 | 27.06.2019 17:39:42 | -3,600 | ec3fb89d1f7bf19dfc4069fd9601ac50693c6070 | Test for defvjp closure error | [
{
"change_type": "MODIFY",
"old_path": "tests/api_test.py",
"new_path": "tests/api_test.py",
"diff": "@@ -576,6 +576,20 @@ class APITest(jtu.JaxTestCase):\n\"the scope of <jax.custom_transforms function bar>, but defjvp and \"\n\"defjvp_all only support differentiation w.r.t. positional arguments.\"... | Python | Apache License 2.0 | google/jax | Test for defvjp closure error |
260,314 | 27.06.2019 17:28:36 | 14,400 | dc91d00afd48e25cd9c8f27e38e171fe715d635d | use split 2 instead of split 1 | [
{
"change_type": "MODIFY",
"old_path": "jax/random.py",
"new_path": "jax/random.py",
"diff": "@@ -545,12 +545,12 @@ def _gamma_one(key, alpha):\nsqueeze_const = _constant_like(alpha, 0.0331)\ndtype = lax.dtype(alpha)\n+ key, subkey = split(key)\n# for alpha < 1, we boost alpha to alpha + 1 and get a... | Python | Apache License 2.0 | google/jax | use split 2 instead of split 1 |
260,489 | 01.07.2019 11:02:26 | -32,400 | b7cca2bbc96c96518e3f0553afb654c745d46fd6 | Add Laplace CDF. | [
{
"change_type": "MODIFY",
"old_path": "jax/scipy/stats/laplace.py",
"new_path": "jax/scipy/stats/laplace.py",
"diff": "@@ -33,3 +33,16 @@ def logpdf(x, loc=0, scale=1):\n@_wraps(osp_stats.laplace.pdf)\ndef pdf(x, loc=0, scale=1):\nreturn lax.exp(logpdf(x, loc, scale))\n+\n+@_wraps(osp_stats.laplace... | Python | Apache License 2.0 | google/jax | Add Laplace CDF. |
260,489 | 02.07.2019 10:10:27 | -32,400 | 6dbd0abffd12f8a37f5dc24690cc1b3397636489 | Add tests for Laplace CDF. | [
{
"change_type": "MODIFY",
"old_path": "jax/scipy/stats/laplace.py",
"new_path": "jax/scipy/stats/laplace.py",
"diff": "@@ -41,8 +41,6 @@ def cdf(x, loc=0, scale=1):\none = _constant_like(x, 1)\nzero = _constant_like(x, 0)\ndiff = lax.div(lax.sub(x, loc), scale)\n- case1 = lax.mul(lax.mul(half, lax.... | Python | Apache License 2.0 | google/jax | Add tests for Laplace CDF. |
260,299 | 02.07.2019 13:47:59 | -3,600 | 03d6d0a5bc4e4cccd9bd8e2ca101710cfb856089 | Add note to custom_transorms docstring | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -1008,6 +1008,11 @@ def custom_transforms(fun):\nspecified manually. The default behavior is retained for any non-overridden\nrules.\n+ The function ``fun`` must satisfy the same constraints required for jit\n... | Python | Apache License 2.0 | google/jax | Add note to custom_transorms docstring |
260,335 | 02.07.2019 12:38:16 | 25,200 | 9790890b7d1e571d60675e65992b8e5cc45e8755 | reduce travis sampled test cases from 100 to 25
This change is just to speed up travis, which has ballooned from ~10
minutes to ~40 minutes. | [
{
"change_type": "MODIFY",
"old_path": ".travis.yml",
"new_path": ".travis.yml",
"diff": "@@ -7,8 +7,8 @@ python:\n- \"2.7\"\n- \"3.6\"\nenv:\n- - JAX_ENABLE_X64=0 JAX_NUM_GENERATED_CASES=100\n- - JAX_ENABLE_X64=1 JAX_NUM_GENERATED_CASES=100\n+ - JAX_ENABLE_X64=0 JAX_NUM_GENERATED_CASES=25\n+ - JAX_... | Python | Apache License 2.0 | google/jax | reduce travis sampled test cases from 100 to 25
This change is just to speed up travis, which has ballooned from ~10
minutes to ~40 minutes. |
260,299 | 03.07.2019 08:00:00 | -3,600 | ffa43b895b75f4533d17b1b85a01a290c4fe4025 | Update signature fo lower_fun | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -1122,7 +1122,7 @@ 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 | Update signature fo lower_fun |
260,299 | 03.07.2019 08:13:34 | -3,600 | bf367d3a56ddb28b9ba0ebb068dd6cba75ecf6cf | Set instantiate=True in custom_transforms translation rule | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -1122,7 +1122,7 @@ 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 | Set instantiate=True in custom_transforms translation rule |
260,285 | 03.07.2019 17:33:10 | -3,600 | 94889ae18fe296029211e6a11e36fb221beb4326 | Support scipy.stats.norm.ppf | [
{
"change_type": "MODIFY",
"old_path": "jax/scipy/stats/norm.py",
"new_path": "jax/scipy/stats/norm.py",
"diff": "@@ -48,3 +48,8 @@ def cdf(x, loc=0, scale=1):\ndef logcdf(x, loc=0, scale=1):\nx, loc, scale = _promote_args_like(osp_stats.norm.logcdf, x, loc, scale)\nreturn special.log_ndtr(lax.div(l... | Python | Apache License 2.0 | google/jax | Support scipy.stats.norm.ppf |
260,335 | 03.07.2019 21:15:52 | 25,200 | c2e7336b02cf183f4a2d9618dbb211a6dc234f36 | fix ShardedDeviceValue.block_until_ready()
cf. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/pxla.py",
"new_path": "jax/interpreters/pxla.py",
"diff": "@@ -433,7 +433,18 @@ pe.custom_partial_eval_rules[axis_index_p] = _axis_index_partial_eval\n### lazy device-memory persistence and result handling\n-class ShardedDeviceTuple(xla.Device... | Python | Apache License 2.0 | google/jax | fix ShardedDeviceValue.block_until_ready()
cf. #973 |
260,547 | 04.07.2019 11:25:56 | -3,600 | 83fea8565f60ddf0ee842fdda15acb47b999b88c | Use zeros as initialization for avg_sq_grad in rmsprop
This seems more sensitive thing to do than ones. It is also consistent with Keras, PyTorch and Sonnet 2's implementation. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/optimizers.py",
"new_path": "jax/experimental/optimizers.py",
"diff": "@@ -275,7 +275,7 @@ def rmsprop(step_size, gamma=0.9, eps=1e-8):\n\"\"\"\nstep_size = make_schedule(step_size)\ndef init(x0):\n- avg_sq_grad = np.ones_like(x0)\n+ avg_sq_gr... | Python | Apache License 2.0 | google/jax | Use zeros as initialization for avg_sq_grad in rmsprop
This seems more sensitive thing to do than ones. It is also consistent with Keras, PyTorch and Sonnet 2's implementation. |
260,547 | 04.07.2019 18:29:35 | -3,600 | d04954d31f4807655166f2053691d5e005e62ac5 | Convert int input to static_argnums to a tuple
User could make mistake of passing an int to static_argnums, this helps to avoid unnecessary error. | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -110,6 +110,9 @@ def _jit(fun, static_argnums, device_values=True):\ndef f_jitted(*args, **kwargs):\nif _jit_is_disabled or config.read('jax_disable_jit'):\nreturn fun(*args, **kwargs)\n+ if isinstance(static_... | Python | Apache License 2.0 | google/jax | Convert int input to static_argnums to a tuple
User could make mistake of passing an int to static_argnums, this helps to avoid unnecessary error. |
260,335 | 05.07.2019 07:47:38 | 25,200 | 527fe14838eadc48fede68bd42ce4364e269fed3 | fix simple static_argnums bug | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -106,13 +106,13 @@ def jit(fun, static_argnums=()):\nreturn _jit(fun, static_argnums)\ndef _jit(fun, static_argnums, device_values=True):\n+ if isinstance(static_argnums, int):\n+ static_argnums = (static_argn... | Python | Apache License 2.0 | google/jax | fix simple static_argnums bug |
260,335 | 05.07.2019 14:32:04 | 25,200 | 93841df8221b76ffb95d02a88ddc6e7eb53e6013 | fix lax.imag jvp and enable test, fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -1608,7 +1608,7 @@ real_p = unop(_complex_basetype, _complex, 'real')\nad.deflinear(real_p, lambda t: [complex(t, onp.zeros((), _dtype(t)))])\nimag_p = unop(_complex_basetype, _complex, 'imag')\n-ad.de... | Python | Apache License 2.0 | google/jax | fix lax.imag jvp and enable test, fixes #979 |
260,335 | 05.07.2019 14:39:32 | 25,200 | db52d42597d0053c62383e5ee70f98a61b0e80f8 | also fix lax.complex jvp, enable test | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -1613,7 +1613,7 @@ ad.defjvp(imag_p, lambda g, _: real(mul(_const(g, -1j), g)))\n_complex_dtype = lambda dtype, *args: (onp.zeros((), dtype) + onp.zeros((), onp.complex64)).dtype\ncomplex_p = binop(_co... | Python | Apache License 2.0 | google/jax | also fix lax.complex jvp, enable test |
260,335 | 05.07.2019 16:55:11 | 25,200 | 0f1d913db3406aceb00bcba107e88ce14ef861c1 | add xla_computation to jax rst docs | [
{
"change_type": "MODIFY",
"old_path": "docs/jax.rst",
"new_path": "docs/jax.rst",
"diff": "@@ -19,6 +19,6 @@ Module contents\n---------------\n.. automodule:: jax\n- :members: jit, disable_jit, grad, value_and_grad, vmap, pmap, jacfwd, jacrev, hessian, jvp, linearize, vjp, make_jaxpr, eval_shape, c... | Python | Apache License 2.0 | google/jax | add xla_computation to jax rst docs |
260,335 | 06.07.2019 11:16:32 | 25,200 | ce2833367abfbf01a3b1896b717e09a7a9422117 | add jax.numpy.cov and tests (cf.
also add jax.numpy.array(..., ndmin=n) | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -1358,34 +1358,36 @@ def atleast_3d(*arys):\n@_wraps(onp.array)\ndef array(object, dtype=None, copy=True, order=\"K\", ndmin=0):\n- if ndmin != 0 or (order is not None and order != \"K\... | Python | Apache License 2.0 | google/jax | add jax.numpy.cov and tests (cf. #70)
also add jax.numpy.array(..., ndmin=n) |
260,335 | 17.06.2019 20:44:33 | 25,200 | ccb1760f499bbac0801fbef67f6bdede65509ddb | add a lot of systematic vmap tests | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -822,19 +822,20 @@ def _reduction_jaxpr(computation, init_value):\ndef _get_monoid_reducer(monoid_op, x):\naval = core.get_aval(x)\n+ dtype = _dtype(x)\nif (type(aval) is ConcreteArray) and aval.shape ... | Python | Apache License 2.0 | google/jax | add a lot of systematic vmap tests |
260,335 | 06.07.2019 11:47:50 | 25,200 | febad2d863bc01e3283b653277c370b61ebf6165 | fix broadcast_in_dim batching rule | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -2268,11 +2268,10 @@ def _broadcast_in_dim_batch_rule(batched_args, batch_dims, shape,\nbroadcast_dimensions):\noperand, = batched_args\nbdim, = batch_dims\n- new_shape = list(shape)\n- new_shape.inser... | Python | Apache License 2.0 | google/jax | fix broadcast_in_dim batching rule |
260,335 | 06.07.2019 11:52:24 | 25,200 | ddf7f69cad03e373ccfe1849dd6b5bf0977322a7 | fix seleect broadcasting rule | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -2569,8 +2569,7 @@ def _select_batch_rule(batched_args, batch_dims, **unused_kwargs):\nelif onp.ndim(pred) == 0 and ot_bdim is not None and of_bdim is not None:\nif ot_bdim == of_bdim:\nreturn select(p... | Python | Apache License 2.0 | google/jax | fix seleect broadcasting rule |
260,335 | 06.07.2019 11:58:33 | 25,200 | 79668ae4ed0ddb7aaebbb19f0077b0cafc8b8310 | fix reduce_window batching rule | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -3445,7 +3445,7 @@ def _reduce_window_batch_rule(\noperand = reduce_window(\noperand, window_dimensions, window_strides, padding)\n- return operand, 0\n+ return operand, bdim\nreduce_window_sum_p = sta... | Python | Apache License 2.0 | google/jax | fix reduce_window batching rule |
260,335 | 06.07.2019 12:17:00 | 25,200 | 968ad9b597c3f6e368a279cae7b559b259a03145 | disable failing reduce-min int64 extreme value tests | [
{
"change_type": "MODIFY",
"old_path": "tests/lax_test.py",
"new_path": "tests/lax_test.py",
"diff": "@@ -1032,8 +1032,9 @@ class LaxTest(jtu.JaxTestCase):\nself._CheckAgainstNumpy(op, numpy_op, args_maker)\n@parameterized.named_parameters(jtu.cases_from_list(\n- {\"testcase_name\": \"_op={}_inshape... | Python | Apache License 2.0 | google/jax | disable failing reduce-min int64 extreme value tests |
260,335 | 08.07.2019 18:33:09 | 25,200 | 30d0a84ba85f7593c1a8286845afd5433ec66d18 | JaxprTrace.process_map wasn't using env_tracers
fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/partial_eval.py",
"new_path": "jax/interpreters/partial_eval.py",
"diff": "@@ -116,8 +116,7 @@ class JaxprTrace(Trace):\nout_pv_const, consts = call_primitive.bind(fun, *in_consts, **params)\nout_pv, jaxpr, env = aux()\nconst_tracers = map(sel... | Python | Apache License 2.0 | google/jax | JaxprTrace.process_map wasn't using env_tracers
fixes #1000 |
260,335 | 15.07.2019 23:04:50 | -3,600 | 861e939324d5ca3c0751e078b78021838363a077 | import random in jax/__init__.py | [
{
"change_type": "MODIFY",
"old_path": "jax/__init__.py",
"new_path": "jax/__init__.py",
"diff": "@@ -17,4 +17,5 @@ os.environ.setdefault('TF_CPP_MIN_LOG_LEVEL', '1')\nfrom jax.version import __version__\nfrom jax.api import *\n+from jax import random\nimport jax.numpy as np # side-effecting import ... | Python | Apache License 2.0 | google/jax | import random in jax/__init__.py |
260,335 | 17.07.2019 23:25:55 | 25,200 | 4c34541c00c02fa750b63a6ea9149909e6c4078f | raise error when vmap used with kwargs | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -574,6 +574,10 @@ def vmap(fun, in_axes=0, out_axes=0):\n@wraps(fun, docstr=docstr)\ndef batched_fun(*args, **kwargs):\n+ if kwargs:\n+ msg = (\"kwargs not yet supported for functions output by vmap. Please \"... | Python | Apache License 2.0 | google/jax | raise error when vmap used with kwargs (#912) |
260,713 | 19.07.2019 12:04:33 | 25,200 | 8985d684a1b6dee1ec6e0a7a29683a3e0f9d979d | remove static argnums from random.fold_in | [
{
"change_type": "MODIFY",
"old_path": "jax/random.py",
"new_path": "jax/random.py",
"diff": "@@ -190,7 +190,7 @@ def fold_in(key, data):\n\"\"\"\nreturn _fold_in(key, data)\n-@partial(jit, static_argnums=(1,))\n+@jit\ndef _fold_in(key, data):\nkey2 = lax.tie_in(key, PRNGKey(data))\nreturn threefry_... | Python | Apache License 2.0 | google/jax | remove static argnums from random.fold_in |
260,335 | 23.07.2019 12:21:28 | -10,800 | ec456a181b818b0cc07ce14b756b4e525beb6126 | Update random.py
Update `jax.random.fold_in` docstring to specify `data` is treated as a 32bit integer. | [
{
"change_type": "MODIFY",
"old_path": "jax/random.py",
"new_path": "jax/random.py",
"diff": "@@ -182,7 +182,7 @@ def fold_in(key, data):\nArgs:\nkey: a PRNGKey (an array with shape (2,) and dtype uint32).\n- data: an integer representing data to be folded in to the key.\n+ data: a 32bit integer rep... | Python | Apache License 2.0 | google/jax | Update random.py
Update `jax.random.fold_in` docstring to specify `data` is treated as a 32bit integer. |
260,335 | 23.07.2019 02:48:53 | 25,200 | c42665c5e926aeb2331c34aeb73c043715eb1a4c | first cut at jit device_assignment api
make execute_primitive put args on correct device | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -68,7 +68,7 @@ flags.DEFINE_bool(\"jax_disable_jit\",\n\"Disable JIT compilation and just call original Python.\")\n-def jit(fun, static_argnums=()):\n+def jit(fun, static_argnums=(), device_assignment=None):\... | Python | Apache License 2.0 | google/jax | first cut at jit device_assignment api
make execute_primitive put args on correct device |
260,335 | 24.07.2019 20:43:34 | -10,800 | 75150bf335c21abc16342b3f8f5b87d51cf9a9f1 | document int device_assignment argument of jit
fix int/long bug | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -90,6 +90,9 @@ def jit(fun, static_argnums=(), device_assignment=None):\ndifferent values for these constants will trigger recompilation. If the\njitted function is called with fewer positional arguments than ... | Python | Apache License 2.0 | google/jax | document int device_assignment argument of jit
fix int/long bug |
260,335 | 24.07.2019 21:45:56 | -10,800 | 3f9c001c3342d8e33745037024c09b0a07170d7f | add ShardedDeviceTuple constant handler, fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/pxla.py",
"new_path": "jax/interpreters/pxla.py",
"diff": "@@ -399,6 +399,8 @@ batching.pytype_aval_mappings[ShardedDeviceTuple] = op.attrgetter('aval')\nxla.canonicalize_dtype_handlers[ShardedDeviceTuple] = \\\nxla.canonicalize_dtype_handlers... | Python | Apache License 2.0 | google/jax | add ShardedDeviceTuple constant handler, fixes #1062 |
260,335 | 24.07.2019 22:03:19 | -10,800 | dbb907c8bc2ea189318a5b500cf4a87e6f6d507b | add warning that device_assignment api is unstable | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -90,9 +90,10 @@ def jit(fun, static_argnums=(), device_assignment=None):\ndifferent values for these constants will trigger recompilation. If the\njitted function is called with fewer positional arguments than... | Python | Apache License 2.0 | google/jax | add warning that device_assignment api is unstable |
260,335 | 25.07.2019 12:41:11 | 25,200 | 0546c9499223ee1a359833dac8adb9cf81015caa | speed up pmap axis-size getting | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -35,6 +35,7 @@ from warnings import warn\nimport numpy as onp\nfrom contextlib import contextmanager\nfrom distutils.util import strtobool\n+import six\nfrom six.moves import reduce\nfrom . import core\n@@ -43... | Python | Apache License 2.0 | google/jax | speed up pmap axis-size getting
Co-authored-by: Peter Hawkins <phawkins@google.com> |
260,316 | 25.07.2019 19:26:43 | 14,400 | ab20adea468546f24b804a03af2bdd9b51ed86a4 | Add example of an adaptive-step ODE solver and the adjoint sensitivities method. | [
{
"change_type": "ADD",
"old_path": null,
"new_path": "ode.py",
"diff": "+# Copyright 2018 Google LLC\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may obtain a copy of the License at\n+#\n+# htt... | Python | Apache License 2.0 | google/jax | Add example of an adaptive-step ODE solver and the adjoint sensitivities method. |
260,335 | 25.07.2019 18:11:44 | 25,200 | aa190ef86daf3531a1886934e7a3e6318f3a7f8d | fix num_replicas counts for initial-style control
fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/xla.py",
"new_path": "jax/interpreters/xla.py",
"diff": "@@ -348,12 +348,19 @@ def _axis_groups(nrep, mesh_spec, mesh_axes):\nreturn tuple(map(tuple, groups.T))\ndef jaxpr_replicas(jaxpr):\n- nums = (eqn_replicas(eqn) for eqn in jaxpr.eqns if ... | Python | Apache License 2.0 | google/jax | fix num_replicas counts for initial-style control
fixes #1065 |
260,335 | 26.07.2019 17:21:11 | 25,200 | f34df7060551dd38143a9f7d255f858764d64645 | fix triangular_solve_transpose_rule comment | [
{
"change_type": "MODIFY",
"old_path": "jax/lax_linalg.py",
"new_path": "jax/lax_linalg.py",
"diff": "@@ -312,9 +312,8 @@ def triangular_solve_jvp_rule_a(\ndef triangular_solve_transpose_rule(\ncotangent, a, b, left_side, lower, transpose_a, conjugate_a,\nunit_diagonal):\n- # Triangular solve is lin... | Python | Apache License 2.0 | google/jax | fix triangular_solve_transpose_rule comment |
260,683 | 28.07.2019 15:27:16 | 14,400 | 07a90470f93924c114a1e73eadb0f9a102f424ae | Implementation of np.corrcoef | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -2625,6 +2625,30 @@ def cov(m, y=None, rowvar=True, bias=False, ddof=None, fweights=None,\nreturn true_divide(dot(X, X_T.conj()), f).squeeze()\n+@_wraps(onp.corrcoef)\n+def corrcoef(x, ... | Python | Apache License 2.0 | google/jax | Implementation of np.corrcoef |
260,698 | 29.07.2019 11:06:08 | 14,400 | d9b7c5fa394406a38d79e818ac831c7c601a598e | made changes to corrcoef | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -2440,7 +2440,6 @@ def cov(m, y=None, rowvar=True, bias=False, ddof=None, fweights=None,\nX_T = X.T if w is None else (X * w).T\nreturn true_divide(dot(X, X_T.conj()), f).squeeze()\n-\n... | Python | Apache License 2.0 | google/jax | made changes to corrcoef |
260,683 | 29.07.2019 11:53:40 | 14,400 | a06883d91f6e8bea2c263ab0fddf7dced2665cba | Made changes based on review. | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -2627,16 +2627,11 @@ def cov(m, y=None, rowvar=True, bias=False, ddof=None, fweights=None,\n@_wraps(onp.corrcoef)\ndef corrcoef(x, y=None, rowvar=True, bias=None, ddof=None):\n- msg = (... | Python | Apache License 2.0 | google/jax | Made changes based on review. |
260,683 | 29.07.2019 13:05:48 | 14,400 | 5487c784d62e345da7800c412eee7f3aab7ffde1 | added shape check | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -2628,7 +2628,7 @@ def cov(m, y=None, rowvar=True, bias=False, ddof=None, fweights=None,\n@_wraps(onp.corrcoef)\ndef corrcoef(x, y=None, rowvar=True, bias=None, ddof=None):\nc = cov(x, ... | Python | Apache License 2.0 | google/jax | added shape check |
260,683 | 29.07.2019 13:11:43 | 14,400 | b89e5a7ac05550925605545321d581aa897f3054 | shape_c variable taken out | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -2628,7 +2628,7 @@ def cov(m, y=None, rowvar=True, bias=False, ddof=None, fweights=None,\n@_wraps(onp.corrcoef)\ndef corrcoef(x, y=None, rowvar=True, bias=None, ddof=None):\nc = cov(x, ... | Python | Apache License 2.0 | google/jax | shape_c variable taken out |
260,683 | 29.07.2019 22:56:30 | 14,400 | 4dcae5debf5d8e93bb4ad7b947edd8ec8926d38d | Update lax_numpy.py | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -2628,7 +2628,7 @@ def cov(m, y=None, rowvar=True, bias=False, ddof=None, fweights=None,\n@_wraps(onp.corrcoef)\ndef corrcoef(x, y=None, rowvar=True, bias=None, ddof=None):\nc = cov(x, ... | Python | Apache License 2.0 | google/jax | Update lax_numpy.py |
260,335 | 31.07.2019 13:27:19 | 25,200 | 0600b738f44914df66d696094aec3e8a8380043a | fix symbolic zero handling in _pad_transpose
tested manually against example from | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -2386,8 +2386,11 @@ def _pad_shape_rule(operand, padding_value, padding_config):\nreturn tuple(out_shape)\ndef _pad_transpose(t, operand, padding_value, padding_config):\n- lo, hi, interior = zip(*padd... | Python | Apache License 2.0 | google/jax | fix symbolic zero handling in _pad_transpose
tested manually against example from @matthewdhoffman |
260,299 | 01.08.2019 15:44:23 | -3,600 | 47f9eedb60f72a83b50a35b44a8f71e65e09ed36 | Correct jax.numpy.pad signature | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -1259,7 +1259,7 @@ def _pad(array, pad_width, mode, constant_values):\nraise NotImplementedError(msg.format(mode))\n@_wraps(onp.pad)\n-def pad(array, pad_width, mode, constant_values=0)... | Python | Apache License 2.0 | google/jax | Correct jax.numpy.pad signature |
260,314 | 01.08.2019 12:39:33 | 14,400 | e1ee87b5598a687f691d9f81018df163982dd579 | add batching rule for lax.sort | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -3796,8 +3796,16 @@ def _sort_jvp_rule(g, operand, dimension):\n_, g_out = sort_key_val(operand, g, dimension)\nreturn g_out\n+def _sort_batch_rule(batched_args, batch_dims, dimension):\n+ operand, = b... | Python | Apache License 2.0 | google/jax | add batching rule for lax.sort |
260,314 | 01.08.2019 14:19:41 | 14,400 | 98152d9d07ea494bd2219d55b8429e1597de7aed | add numpy.median | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -2720,6 +2720,13 @@ def percentile(a, q, axis=None, out=None, overwrite_input=False,\nreturn quantile(a, q, axis=axis, out=out, overwrite_input=overwrite_input,\ninterpolation=interpola... | Python | Apache License 2.0 | google/jax | add numpy.median |
260,314 | 01.08.2019 16:20:08 | 14,400 | 45c5bd4fbace397d26fe37c1e417d0e158954421 | support ddof for var | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -1083,8 +1083,6 @@ def var(a, axis=None, dtype=None, out=None, ddof=0, keepdims=False):\nif out is not None:\nraise ValueError(\"var does not support the `out` argument.\")\n- if ddof !... | Python | Apache License 2.0 | google/jax | support ddof for var |
260,314 | 01.08.2019 16:39:08 | 14,400 | 2836d03b5e5cfccc4cd547e299c8ed945ab8382a | add some nonlinearity | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/stax.py",
"new_path": "jax/experimental/stax.py",
"diff": "@@ -42,6 +42,8 @@ import jax.numpy as np\ndef relu(x): return np.maximum(x, 0.)\ndef softplus(x): return np.logaddexp(x, 0.)\ndef sigmoid(x): return 1. / (1. + np.exp(-x))\n+def elu(x)... | Python | Apache License 2.0 | google/jax | add some nonlinearity |
260,314 | 01.08.2019 16:41:06 | 14,400 | 688d77f4326cc04f0a358f070a228b06b02efdb8 | better test str message | [
{
"change_type": "MODIFY",
"old_path": "tests/stax_test.py",
"new_path": "tests/stax_test.py",
"diff": "@@ -131,9 +131,9 @@ class StaxTest(jtu.JaxTestCase):\n.format(input_shape, nonlinear),\n\"input_shape\": input_shape, \"nonlinear\": nonlinear}\nfor input_shape in [(2, 3), (2, 3, 4)]\n- for nonli... | Python | Apache License 2.0 | google/jax | better test str message |
260,314 | 01.08.2019 17:11:31 | 14,400 | 5ffe2ae5dd8ef5625bbd0ad2c5c15fd3bc142964 | expose sigmoid too | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/stax.py",
"new_path": "jax/experimental/stax.py",
"diff": "@@ -198,6 +198,7 @@ Exp = elementwise(np.exp)\nLogSoftmax = elementwise(logsoftmax, axis=-1)\nSoftmax = elementwise(softmax, axis=-1)\nSoftplus = elementwise(softplus)\n+Sigmoid = elem... | Python | Apache License 2.0 | google/jax | expose sigmoid too |
260,314 | 01.08.2019 19:12:03 | 14,400 | 7f4dc87a4c33a0b56631632b9b8b12c96db71514 | add multigammaln, entr | [
{
"change_type": "MODIFY",
"old_path": "jax/scipy/special.py",
"new_path": "jax/scipy/special.py",
"diff": "@@ -99,6 +99,27 @@ def xlog1py(x, y):\nreturn lax._safe_mul(x, lax.log1p(y))\n+@_wraps(osp_special.entr)\n+def entr(x):\n+ x, = _promote_args_like(osp_special.entr, x)\n+ return lax.select(lax... | Python | Apache License 2.0 | google/jax | add multigammaln, entr |
260,335 | 02.08.2019 11:26:17 | 25,200 | 3168006f4add95224082515a3e4fc84b3c42a144 | fix np.var dtype bug | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -1097,9 +1097,9 @@ def var(a, axis=None, dtype=None, out=None, ddof=0, keepdims=False):\nnormalizer = onp.prod(onp.take(shape(a), axis))\nnormalizer = normalizer - ddof\n- return lax.di... | Python | Apache License 2.0 | google/jax | fix np.var dtype bug |
260,288 | 02.08.2019 16:42:17 | 25,200 | 444aced7912de6d4b7d91ed29ec164ebf7f52a76 | Fix pack_optimizer_state to correctly use tuples everywhere in the packed state and add a unit test to check round trip unpack/pack. | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/optimizers.py",
"new_path": "jax/experimental/optimizers.py",
"diff": "@@ -506,5 +506,6 @@ def pack_optimizer_state(marked_pytree):\nsentinels, tree_def = tree_flatten(marked_pytree)\nassert all(isinstance(s, JoinPoint) for s in sentinels)\nsu... | Python | Apache License 2.0 | google/jax | Fix pack_optimizer_state to correctly use tuples everywhere in the packed state and add a unit test to check round trip unpack/pack. |
260,389 | 03.08.2019 21:27:06 | 25,200 | d0c9f453497a9a647df3b7710b8dc732bac3058a | fix jax.numpy reduction init_val for bools | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax.py",
"new_path": "jax/lax/lax.py",
"diff": "@@ -3255,7 +3255,7 @@ def _reduction_computation(c, jaxpr, consts, init_value):\nreduce_p = standard_primitive(_reduce_shape_rule, _input_dtype, 'reduce',\n_reduce_translation_rule)\n-# batching.primitive... | Python | Apache License 2.0 | google/jax | fix jax.numpy reduction init_val for bools |
260,299 | 06.08.2019 11:55:59 | -3,600 | 45e52b8f3fcef719e692d275ac05c970ced06351 | Fix ks test and rm randint ks test | [
{
"change_type": "MODIFY",
"old_path": "tests/random_test.py",
"new_path": "tests/random_test.py",
"diff": "@@ -51,8 +51,7 @@ class LaxRandomTest(jtu.JaxTestCase):\ndef _CheckKolmogorovSmirnovCDF(self, samples, cdf):\nfail_prob = 0.01 # conservative bound on statistical fail prob by Kolmo CDF\n- sta... | Python | Apache License 2.0 | google/jax | Fix ks test and rm randint ks test |
260,299 | 06.08.2019 12:19:05 | -3,600 | f351dedbb7b3ade9252ec7096c7c8012cc536aba | Add logistic distribution to jax.random | [
{
"change_type": "MODIFY",
"old_path": "jax/random.py",
"new_path": "jax/random.py",
"diff": "@@ -35,6 +35,7 @@ from .api import custom_transforms, defjvp, jit, vmap\nfrom .numpy.lax_numpy import _constant_like, asarray\nfrom jax.lib import xla_bridge\nfrom jax import core\n+from jax.scipy.special i... | Python | Apache License 2.0 | google/jax | Add logistic distribution to jax.random |
260,335 | 08.08.2019 08:59:44 | 25,200 | 873d8e4001f29ad15214e35ce1209cc989b19476 | compiled while_loop impl cf. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/xla.py",
"new_path": "jax/interpreters/xla.py",
"diff": "@@ -79,11 +79,17 @@ def primitive_computation(prim, *shapes, **params):\nc = xb.make_computation_builder(\"primitive_computation\")\nplatform = xb.get_backend().platform\nxla_args = map(... | Python | Apache License 2.0 | google/jax | compiled while_loop impl cf. #1131 #1130 #852 |
260,344 | 08.08.2019 19:42:59 | -10,800 | 442ce9701c0e1edbdd42082da20b47ce9343edc6 | update notes and comments in tutorial | [
{
"change_type": "MODIFY",
"old_path": "notebooks/score_matching.ipynb",
"new_path": "notebooks/score_matching.ipynb",
"diff": "\" net_params = get_params(opt_state)\\n\",\n\" loss = compute_loss(net_params, batch)\\n\",\n\" grads = jax.grad(compute_loss, argnums=0)(net_params, batch)\\n\",\n- \" re... | Python | Apache License 2.0 | google/jax | update notes and comments in tutorial |
260,344 | 08.08.2019 22:56:51 | -10,800 | 2fbe6daacd462c04b80e7799bf0e0fa241e30ce1 | fix: typo in collab link | [
{
"change_type": "MODIFY",
"old_path": "notebooks/README.md",
"new_path": "notebooks/README.md",
"diff": "@@ -27,7 +27,7 @@ Use the links below to open any of these for interactive exploration in colab.\n[Common_Gotchas_in_JAX]:https://colab.sandbox.google.com/github/google/jax/blob/master/notebooks... | Python | Apache License 2.0 | google/jax | fix: typo in collab link |
260,344 | 08.08.2019 23:09:57 | -10,800 | 0b22d4db20e74161030b86494d6b7989e5a3c5cc | update colab link inside tutorial notebook | [
{
"change_type": "MODIFY",
"old_path": "notebooks/score_matching.ipynb",
"new_path": "notebooks/score_matching.ipynb",
"diff": "\"source\": [\n\"### Score Matching with JAX\\n\",\n\"\\n\",\n- \"[](https://colab.research.google.... | Python | Apache License 2.0 | google/jax | update colab link inside tutorial notebook |
260,565 | 08.08.2019 16:24:46 | 25,200 | d8271179b3372a6776cb792537a41cc14cdea7a6 | Fix documentation for jax.lax.ppermute | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax_parallel.py",
"new_path": "jax/lax/lax_parallel.py",
"diff": "@@ -107,11 +107,11 @@ def ppermute(x, axis_name, perm):\n``axis_name``. Any two pairs should not have the same source index or the\nsame destination index. For each index of the axis ``a... | Python | Apache License 2.0 | google/jax | Fix documentation for jax.lax.ppermute |
260,646 | 09.08.2019 20:31:43 | 0 | 5887fe3d413fed1ae58e5e601ed2bc57ebe349cc | Add adaptive-step ODE solver and VJP calculation with JAX control flow operators for fast compilation | [
{
"change_type": "ADD",
"old_path": null,
"new_path": "jax/experimental/odeint.py",
"diff": "+\"\"\"JAX-based Dormand-Prince ODE integration with adaptive stepsize.\n+\n+Integrate systems of ordinary differential equations (ODEs) using the JAX\n+autograd/diff library and the Dormand-Prince method fo... | Python | Apache License 2.0 | google/jax | Add adaptive-step ODE solver and VJP calculation with JAX control flow operators for fast compilation |
260,646 | 09.08.2019 21:19:45 | 0 | 156b3c86e5c3d3be8c6698d3feaf1a844fde2e40 | Cleaning up variable & function names | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/odeint.py",
"new_path": "jax/experimental/odeint.py",
"diff": "@@ -276,14 +276,14 @@ def grad_odeint(ofunc, args):\nrev_aug_dynamics = lambda y, t: -aug_dynamics(y, -t)\n@jax.jit\n- def _fori_loop_func(i, val):\n+ def _fori_body_fun(i, val):\n... | Python | Apache License 2.0 | google/jax | Cleaning up variable & function names |
260,646 | 12.08.2019 17:56:46 | 0 | a7323ec5df415a7d245aad86901b058046339ecb | Restored license header + plotting example, added benchmark | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/ode.py",
"new_path": "jax/experimental/ode.py",
"diff": "+# Copyright 2018 Google LLC\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may ob... | Python | Apache License 2.0 | google/jax | Restored license header + plotting example, added benchmark |
260,646 | 12.08.2019 22:11:42 | 0 | 225c90b7e8e954e30aafbfa2f377e16061de6b7a | Convert tableaux & runge_kutta_step to JAX control flow, use np.where | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/ode.py",
"new_path": "jax/experimental/ode.py",
"diff": "@@ -25,6 +25,8 @@ from __future__ import division\nfrom __future__ import print_function\nimport functools\n+import time\n+\nimport jax\nfrom jax.config import config\nfrom jax.flatten_u... | Python | Apache License 2.0 | google/jax | Convert tableaux & runge_kutta_step to JAX control flow, use np.where |
260,335 | 12.08.2019 18:03:25 | 25,200 | 4a4304e08c103dc1adb20550cd9c1d55a6f784ac | make pmap read axis size from kwargs
fixes | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -711,7 +711,7 @@ def pmap(fun, axis_name=None):\n@wraps(fun)\ndef f_pmapped(*args, **kwargs):\n- axis_size = _pmap_axis_size(args)\n+ axis_size = _pmap_axis_size((args, kwargs))\nf = lu.wrap_init(fun)\nargs_fl... | Python | Apache License 2.0 | google/jax | make pmap read axis size from kwargs
fixes #1170 |
260,335 | 13.08.2019 07:20:09 | 25,200 | 275bf9da6d29bb6b39e05c2a4886a6123d4ecddc | improve prng compile times with loop rolling
cf. | [
{
"change_type": "MODIFY",
"old_path": "jax/random.py",
"new_path": "jax/random.py",
"diff": "@@ -123,34 +123,29 @@ def threefry_2x32(keypair, count):\nelse:\nx = list(np.split(count.ravel(), 2))\n- rotations = onp.uint32([13, 15, 26, 6, 17, 29, 16, 24])\n+ rotations = asarray([13, 15, 26, 6, 17, 29... | Python | Apache License 2.0 | google/jax | improve prng compile times with loop rolling
cf. #1172 |
260,335 | 13.08.2019 09:43:44 | 25,200 | d857d3f595f387495b016b6b60de91ed9636af79 | improve prng compile times with outer-loop rolling | [
{
"change_type": "MODIFY",
"old_path": "jax/random.py",
"new_path": "jax/random.py",
"diff": "@@ -123,33 +123,19 @@ def threefry_2x32(keypair, count):\nelse:\nx = list(np.split(count.ravel(), 2))\n- rotations = asarray([13, 15, 26, 6, 17, 29, 16, 24], dtype=\"uint32\")\n- ks = [key1, key2, key1 ^ ke... | Python | Apache License 2.0 | google/jax | improve prng compile times with outer-loop rolling |
260,335 | 13.08.2019 11:30:24 | 25,200 | 75bb38e7418aa5203badea7d0bcf6afc1272a177 | address reviewer comments, no op-by-op in threefry | [
{
"change_type": "MODIFY",
"old_path": "jax/random.py",
"new_path": "jax/random.py",
"diff": "@@ -32,7 +32,7 @@ from . import lax\nfrom . import numpy as np\nfrom . import tree_util\nfrom .api import custom_transforms, defjvp, jit, vmap\n-from .numpy.lax_numpy import _constant_like, asarray\n+from .... | Python | Apache License 2.0 | google/jax | address reviewer comments, no op-by-op in threefry |
260,329 | 13.08.2019 17:37:15 | 18,000 | 8718e30528e0e9626934ab3b3509ccc39640fafd | Make DeviceValue and subclasses weakref friendly | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/xla.py",
"new_path": "jax/interpreters/xla.py",
"diff": "@@ -434,7 +434,7 @@ for _t in array_types:\nclass DeviceValue(object):\n\"\"\"A DeviceValue represents a value backed by device memory.\"\"\"\n- __slots__ = [\"aval\", \"device_buffer\"]... | Python | Apache License 2.0 | google/jax | Make DeviceValue and subclasses weakref friendly
https://stackoverflow.com/questions/19526340/weakref-and-slots |
260,335 | 14.08.2019 07:13:42 | 25,200 | 2dec779b49709454e5ccdc7488d02ef4a26ed267 | remove author info in notebook (old, redundant) | [
{
"change_type": "MODIFY",
"old_path": "notebooks/XLA_in_Python.ipynb",
"new_path": "notebooks/XLA_in_Python.ipynb",
"diff": "\"\\n\",\n\"_Anselm Levskaya_ \\n\",\n\"\\n\",\n- \"_The Python XLA client was designed by Roy Frostig._\\n\",\n- \"\\n\",\n- \"_JAX was written by Dougal Maclaurin, Peter Ha... | Python | Apache License 2.0 | google/jax | remove author info in notebook (old, redundant) |
260,329 | 15.08.2019 11:26:25 | 18,000 | 4b693777aaa19ab54cf811acf3fa96f0b4534e5b | Ensure reps is a tuple (allows list or other iterable) | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -1297,7 +1297,7 @@ def tile(a, reps):\nif isinstance(reps, int):\nreps = (reps,)\na = reshape(a, (1,) * (len(reps) - ndim(a)) + shape(a))\n- reps = (1,) * (ndim(a) - len(reps)) + reps\n... | Python | Apache License 2.0 | google/jax | Ensure reps is a tuple (allows list or other iterable) |
260,329 | 16.08.2019 12:31:53 | 18,000 | 48f64ec86361968ca59131abbbcbb00b9bbb9be5 | Avoid a TypeError when reps is or contains a ndarray
Something along the lines of `TypeError: multiply only accepts scalar or ndarray, but got a list.` | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -1300,7 +1300,7 @@ def tile(a, reps):\na = reshape(a, (1,) * (len(reps) - ndim(a)) + shape(a))\nreps = (1,) * (ndim(a) - len(reps)) + tuple(reps)\nfor i, rep in enumerate(reps):\n- a = ... | Python | Apache License 2.0 | google/jax | Avoid a TypeError when reps is or contains a ndarray
Something along the lines of `TypeError: multiply only accepts scalar or ndarray, but got a list.` |
260,329 | 16.08.2019 17:18:44 | 18,000 | d07107af5a395b33ac0a67ab4201b301568c82c4 | Makes the `Tracer` object `weakref`-able | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -282,7 +282,7 @@ class Trace(object):\nclass Tracer(object):\n__array_priority__ = 1000\n- __slots__ = ['trace']\n+ __slots__ = ['trace', '__weakref__']\ndef __array__(self):\nraise Exception(\"Tracer can't ... | Python | Apache License 2.0 | google/jax | Makes the `Tracer` object `weakref`-able |
260,403 | 19.08.2019 23:45:36 | 25,200 | cc87fb601332855c8a6d753ac461935a1c664d9f | WIP: experimental multibackend jit | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -83,7 +83,7 @@ class _ThreadLocalState(threading.local):\n_thread_local_state = _ThreadLocalState()\n-def jit(fun, static_argnums=(), device_assignment=None):\n+def jit(fun, static_argnums=(), device_assignmen... | Python | Apache License 2.0 | google/jax | WIP: experimental multibackend jit |
260,403 | 20.08.2019 00:24:30 | 25,200 | 29c6fe889969c866c9a9bd5f0e751ceaf2000697 | small fix for multibackend | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/xla.py",
"new_path": "jax/interpreters/xla.py",
"diff": "@@ -255,6 +255,7 @@ def _jaxpr_computation(jaxpr, backend, axis_env, const_vals, freevar_shapes, *ar\n_map(write, jaxpr.invars, map(c.ParameterWithShape, arg_shapes))\n_prefetch_jaxpr_li... | Python | Apache License 2.0 | google/jax | small fix for multibackend |
260,403 | 20.08.2019 01:00:34 | 25,200 | 7529815614ad09ccaad16d1fbfca714a8daa9051 | fix missing pop default | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/pxla.py",
"new_path": "jax/interpreters/pxla.py",
"diff": "@@ -519,7 +519,7 @@ xb.register_constant_handler(ChunkedDeviceArray,\ndef xla_pmap_impl(fun, *args, **params):\naxis_name = params.pop('axis_name')\naxis_size = params.pop('axis_size')... | Python | Apache License 2.0 | google/jax | fix missing pop default |
260,403 | 21.08.2019 00:22:53 | 25,200 | f01fc35ce5644586f4c4239ab3bc7ff0bbd2b0d5 | Make op-by-op work with all jit-returned devicearrays. | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -105,6 +105,8 @@ def jit(fun, static_argnums=(), device_assignment=None, backend=None):\nchange. Optional, an int specifying the device ordinal for which to compile the\nfunction. The default is inherited from... | Python | Apache License 2.0 | google/jax | Make op-by-op work with all jit-returned devicearrays. |
260,403 | 21.08.2019 01:06:04 | 25,200 | dbe4bdf944188e8de7d797aae39086aa1c1ff629 | add multibackend option to soft_pmap | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -753,16 +753,16 @@ def _axis_size(x):\nraise ValueError(\"could not get axis size of type: {}\".format(x))\n-def soft_pmap(fun, axis_name=None):\n+def soft_pmap(fun, axis_name=None, backend=None):\n_check_call... | Python | Apache License 2.0 | google/jax | add multibackend option to soft_pmap |
260,299 | 21.08.2019 14:27:23 | -3,600 | 9b19afd4d90956f10ed1a444bea52c3096db7812 | Implement Jaxpr __hash__
This means that primitives like scatter, which have a Jaxpr in their
**params, will get cache hits appropriately. | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -54,6 +54,12 @@ class Jaxpr(object):\ndef __repr__(self):\nreturn self.__str__()\n+ def __hash__(self):\n+ return hash(self.__str__())\n+\n+ def __eq__(self, other):\n+ return type(other) is type(self) and s... | Python | Apache License 2.0 | google/jax | Implement Jaxpr __hash__
This means that primitives like scatter, which have a Jaxpr in their
**params, will get cache hits appropriately. |
260,474 | 26.07.2019 18:01:38 | 14,400 | 20fad746a8407d3ba661edfcb1871557cf61cecf | De-dup equations with multiple lhs vars when creating a jaxpr | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -105,8 +105,12 @@ def jaxpr_as_fun(typed_jaxpr, *args):\nreturn out\n-JaxprEqn = namedtuple('JaxprEqn', ['invars', 'outvars', 'primitive',\n+def new_jaxpr_eqn(*args):\n+ return JaxprEqn(object(), *args)\n+\n... | Python | Apache License 2.0 | google/jax | De-dup equations with multiple lhs vars when creating a jaxpr
Co-authored-by: Matthew Johnson <mattjj@google.com> |
260,474 | 26.07.2019 23:17:21 | 14,400 | c53c8bbb438a01310d48a27a46087651e820bd07 | Some progress de-tupling ad.py | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -44,7 +44,7 @@ from . import core\nfrom . import linear_util as lu\nfrom . import ad_util\nfrom .core import eval_jaxpr\n-from .api_util import (wraps, flatten_fun, abstract_tuple_tree_leaves)\n+from .api_util... | Python | Apache License 2.0 | google/jax | Some progress de-tupling ad.py |
260,474 | 27.07.2019 10:43:40 | 14,400 | 3c37a3260ab96c73d17270cb8c198d4c51eb0536 | Update linearize to no-tuple version | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -928,7 +928,7 @@ def linearize(fun, *primals):\n(in_tree, out_tree), out_pvals)\nreturn out_primal_py, lifted_jvp\n-def lift_linearized(jaxpr, primal_avals, consts, io_tree, out_pval, *py_args):\n+def lift_lin... | Python | Apache License 2.0 | google/jax | Update linearize to no-tuple version |
260,329 | 21.08.2019 13:40:50 | 14,400 | 451ff2d694f497e94d1f68e281308a0c840e2e00 | Cope with old numpy lacking axis arg | [
{
"change_type": "MODIFY",
"old_path": "jax/numpy/lax_numpy.py",
"new_path": "jax/numpy/lax_numpy.py",
"diff": "@@ -1516,11 +1516,18 @@ def _wrap_numpy_nullary_function(f):\ndef linspace(start, stop, num=50, endpoint=True, retstep=False, dtype=None,\naxis=0):\n+ try:\nout = onp.linspace(start, stop,... | Python | Apache License 2.0 | google/jax | Cope with old numpy lacking axis arg |
260,335 | 21.08.2019 13:53:57 | 25,200 | f56312bbfff3973ac3d2ca813c0699a0bfce61ad | remove pat_fmap | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -172,9 +172,9 @@ def eval_jaxpr(jaxpr, consts, freevar_vals, *args):\nenv = {}\nwrite(unitvar, unit)\n- pat_fmap(write, jaxpr.constvars, consts)\n- pat_fmap(write, jaxpr.invars, args)\n- pat_fmap(write, jaxp... | Python | Apache License 2.0 | google/jax | remove pat_fmap |
260,335 | 21.08.2019 15:28:23 | 25,200 | 765af9774916cd8f75a842754a444a7b1822e634 | make stax.sigmoid call scipy.special.expit
improves numerical stability of differentiation, so that in particular
now jax.hessian(sigmoid)(-100.) doesn't produce a nan | [
{
"change_type": "MODIFY",
"old_path": "jax/experimental/stax.py",
"new_path": "jax/experimental/stax.py",
"diff": "@@ -30,7 +30,7 @@ from six.moves import reduce\nfrom jax import lax\nfrom jax import random\n-from jax.scipy.special import logsumexp\n+from jax.scipy.special import logsumexp, expit\n... | Python | Apache License 2.0 | google/jax | make stax.sigmoid call scipy.special.expit
improves numerical stability of differentiation, so that in particular
now jax.hessian(sigmoid)(-100.) doesn't produce a nan |
260,335 | 21.08.2019 16:39:32 | 25,200 | 06b5c402dcc38d3bdcb10fd45dd803134da36296 | update tree_util_tests w/ None not a pytree | [
{
"change_type": "MODIFY",
"old_path": "tests/tree_util_tests.py",
"new_path": "tests/tree_util_tests.py",
"diff": "@@ -57,10 +57,10 @@ PYTREES = [\n((),),\n(([()]),),\n((1, 2),),\n- (((1, \"foo\"), [\"bar\", (3, None, 7)]),),\n+ (((1, \"foo\"), [\"bar\", (3, (), 7)]),),\n([3],),\n- ([3, ATuple(foo=... | Python | Apache License 2.0 | google/jax | update tree_util_tests w/ None not a pytree |
260,335 | 21.08.2019 16:39:59 | 25,200 | 7323e1546e45c5e6820666bb7a3a55afa21cb41f | fix ShardedDeviceArray.__getitem__ | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/pxla.py",
"new_path": "jax/interpreters/pxla.py",
"diff": "@@ -344,8 +344,7 @@ class ShardedDeviceArray(ShardedDeviceValue, xla.DeviceArray):\nif self._npy_value is None and type(idx) is int:\nids = self._ids()\ndevice_buffer = self.device_buf... | Python | Apache License 2.0 | google/jax | fix ShardedDeviceArray.__getitem__ |
260,403 | 21.08.2019 18:07:14 | 25,200 | 88a1fbf68548aab5abc88d511784d274d73ea749 | Error on nested conflicting explicit jit backend specifications. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/xla.py",
"new_path": "jax/interpreters/xla.py",
"diff": "@@ -266,7 +266,15 @@ def _jaxpr_computation(jaxpr, backend, axis_env, const_vals, freevar_shapes, *ar\n_map(write, jaxpr.invars, map(c.ParameterWithShape, arg_shapes))\n_prefetch_jaxpr_l... | Python | Apache License 2.0 | google/jax | Error on nested conflicting explicit jit backend specifications. |
260,335 | 21.08.2019 20:36:47 | 25,200 | 8517997518d10aa1885be0b67ad60cc3fc7e28b8 | minor fixes from trax, revise eval_shape api | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -1660,15 +1660,33 @@ def _make_graphviz(fun):\nreturn graphviz_maker\n+class ShapeDtypeStruct(object):\n+ __slots__ = [\"shape\", \"dtype\"]\n+ def __init__(self, shape, dtype):\n+ self.shape = shape\n+ self.d... | Python | Apache License 2.0 | google/jax | minor fixes from trax, revise eval_shape api |
260,403 | 21.08.2019 20:59:18 | 25,200 | c839c6a602ed9eb9e97715c31b481a6ba2a2bd0b | Added basic behavior unit tests of multibackend jit. | [
{
"change_type": "ADD",
"old_path": null,
"new_path": "tests/multibackend_test.py",
"diff": "+# Copyright 2018 Google LLC\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may obtain a copy of the Li... | Python | Apache License 2.0 | google/jax | Added basic behavior unit tests of multibackend jit. |
260,335 | 22.08.2019 12:50:47 | 25,200 | de6ce2a55584ed05d371070e1767cd73ae7357d9 | allow vmap in_axes to be lists | [
{
"change_type": "MODIFY",
"old_path": "jax/api.py",
"new_path": "jax/api.py",
"diff": "@@ -616,7 +616,7 @@ def _flatten_axes(treedef, axis_tree):\n# TODO(mattjj): remove this when jaxlib is updated past 0.1.25\ndef _replace_nones(tuptree):\n- if type(tuptree) is tuple:\n+ if type(tuptree) in (list,... | Python | Apache License 2.0 | google/jax | allow vmap in_axes to be lists |
260,314 | 22.08.2019 22:53:31 | 14,400 | 9f876c485028d0f94ddd0cd60fa0fe8ec23ed9f8 | fix cond with true_consts | [
{
"change_type": "MODIFY",
"old_path": "jax/lax/lax_control_flow.py",
"new_path": "jax/lax/lax_control_flow.py",
"diff": "@@ -304,8 +304,9 @@ def cond(pred, true_operand, true_fun, false_operand, false_fun):\ndef _cond_impl(pred, *args, **kwargs):\ntrue_jaxpr, false_jaxpr, true_nconsts, false_nconst... | Python | Apache License 2.0 | google/jax | fix cond with true_consts |
260,403 | 22.08.2019 20:50:31 | 25,200 | 8bbe1b0c6a6455535bcd1ced1056c555e46625c9 | fix kwarg order bug for device constant put handler | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/xla.py",
"new_path": "jax/interpreters/xla.py",
"diff": "@@ -77,7 +77,7 @@ xla_result_handlers[ConcreteArray] = array_result_handler\ndef device_put(x, device_num=0, backend=None):\nx = canonicalize_dtype(x)\ntry:\n- return device_put_handlers... | Python | Apache License 2.0 | google/jax | fix kwarg order bug for device constant put handler |
260,403 | 22.08.2019 21:17:05 | 25,200 | 78c04fe3bd330cec4026ca9c99ba592c3a2af769 | fix default kwarg issue | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/xla.py",
"new_path": "jax/interpreters/xla.py",
"diff": "@@ -346,7 +346,7 @@ def eqn_replicas(eqn):\ndef _xla_call_impl(fun, *args, **params):\ndevice_assignment = params['device_assignment']\n- backend = params['backend']\n+ backend = params.... | Python | Apache License 2.0 | google/jax | fix default kwarg issue |
260,403 | 23.08.2019 01:20:39 | 25,200 | ceca1b7ae04b5502169fe089ddea06e7504e7664 | try not to sabotage all jax users, use normalized platform names to control device_put behavior. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/xla.py",
"new_path": "jax/interpreters/xla.py",
"diff": "@@ -629,9 +629,9 @@ def _device_array_constant_handler(c, val, canonicalize_types=True):\nxb.register_constant_handler(DeviceArray, _device_array_constant_handler)\ndef _device_put_devic... | Python | Apache License 2.0 | google/jax | try not to sabotage all jax users, use normalized platform names to control device_put behavior. |
260,335 | 23.08.2019 07:47:42 | 25,200 | 7539325f26f6527537013812abf7b531374bb414 | ensure DeviceArray.aval is ShapedArray
Fixes which was caused by DeviceArray.aval, and hence
xla.abstractify(device_array), being a ConcreteArray, leading to
compilation cache misses all the time because those are necessarily
cached on id. | [
{
"change_type": "MODIFY",
"old_path": "jax/interpreters/pxla.py",
"new_path": "jax/interpreters/pxla.py",
"diff": "@@ -308,12 +308,14 @@ class ShardedDeviceArray(ShardedDeviceValue, xla.DeviceArray):\n_collect = staticmethod(onp.stack)\ndef __init__(self, aval, device_buffers):\n- # aval must be a ... | Python | Apache License 2.0 | google/jax | ensure DeviceArray.aval is ShapedArray
Fixes #1239, which was caused by DeviceArray.aval, and hence
xla.abstractify(device_array), being a ConcreteArray, leading to
compilation cache misses all the time because those are necessarily
cached on id. |
260,335 | 23.08.2019 08:17:41 | 25,200 | 98c7567a0d1ba9ca4bc3e72d6043632f96dcddc7 | add flag for logging when jit performs compilation | [
{
"change_type": "MODIFY",
"old_path": "jax/core.py",
"new_path": "jax/core.py",
"diff": "@@ -476,6 +476,7 @@ bot = Bot()\nclass AbstractUnit(AbstractValue):\ndef join(self, other): return self\n+ def _eq(self, self_traced, other): return get_aval(other) is self\nabstract_unit = AbstractUnit()\n"
... | Python | Apache License 2.0 | google/jax | add flag for logging when jit performs compilation |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.