| d8c75da | Matthew Johnson | 13 April 2020, 02:30:05 UTC | update version and changelog for pypi | 13 April 2020, 02:30:05 UTC |
| 2dc81fb | Peter Hawkins | 12 April 2020, 19:35:35 UTC | Make pytest run over JAX tests warning clean, and error on warnings. (#2674) * Make pytest run over JAX tests warning clean, and error on warnings. Remove global warning suppression in travis.yml. Instead add a pytest.ini that converts warnings to errors, with the exception of a whitelist. Either fix or locally suppress warnings in tests. Also fix crashes on Mac related to a preexisting linear algebra bug. * Fix some type errors in the FFT transpose rules revealed by the convert_element_type transpose rule change. | 12 April 2020, 19:35:35 UTC |
| 453dc5f | George Necula | 12 April 2020, 17:05:04 UTC | Relax tolerance for LaxBackedNumpyTests.testConvolutions | 12 April 2020, 17:07:41 UTC |
| 989228a | Jake VanderPlas | 12 April 2020, 14:37:02 UTC | Add implementation of np.rollaxis | 12 April 2020, 15:53:34 UTC |
| f610867 | George Necula | 11 April 2020, 15:27:14 UTC | Improve error message for indexing into an empty dimension It seems that JAX (and XLA) are a lot more forgiving for index-out-of-bounds, except for the case when the dimension has size 0. In this case give a similar IndexError message as numpy, instead of the current one: RuntimeError: Invalid argument: Slice size at index 0 in gather op is out of range, must be within [0, 1), got 1.: This is a bug in JAX's shape-checking rules; please report it! Issue: #2671 | 11 April 2020, 16:30:44 UTC |
| c4dd1cf | George Necula | 11 April 2020, 11:38:29 UTC | Relax tolerance for new test on TPU | 11 April 2020, 11:40:14 UTC |
| ee2cd44 | George Necula | 11 April 2020, 10:09:05 UTC | Added missing import warnings Also fixed a couple of unnecessary f-strings (Cider was flagging them) | 11 April 2020, 10:59:16 UTC |
| 8a95ff0 | Matthew Johnson | 10 April 2020, 22:50:40 UTC | Merge pull request #2675 from jakevdp/convolve Add numpy and scipy convolution wrappers | 10 April 2020, 22:50:40 UTC |
| 89c9c43 | Jake VanderPlas | 10 April 2020, 21:11:16 UTC | Add support for mode=same in convolve2d & correlate2d | 10 April 2020, 21:11:16 UTC |
| 414e54a | Peter Hawkins | 10 April 2020, 19:40:57 UTC | Fix crash for NaN eigenvalues. (#2669) * Fix crash for NaN eigenvalues. * Incorporate review comments. | 10 April 2020, 19:40:57 UTC |
| 656c3a9 | Roman Ring | 10 April 2020, 19:30:01 UTC | fix a typo in docs notebook (#2672) | 10 April 2020, 19:30:01 UTC |
| bad3445 | Matthew Johnson | 10 April 2020, 19:20:29 UTC | Merge pull request #2673 from google/issue2657 add missing instantiate_zeros in custom_jvp | 10 April 2020, 19:20:29 UTC |
| edda69e | Jake VanderPlas | 10 April 2020, 18:54:10 UTC | Add implementations of scipy.signal.convolve & correlate, 1d & 2d | 10 April 2020, 18:54:10 UTC |
| 7750a16 | Matthew Johnson | 10 April 2020, 18:45:33 UTC | add test for #2657 | 10 April 2020, 18:45:33 UTC |
| 61abdc1 | Matthew Johnson | 10 April 2020, 18:41:35 UTC | add missing instantiate_zeros in custom_jvp fixes #2657 | 10 April 2020, 18:41:35 UTC |
| a3cc9a7 | Peter Hawkins | 10 April 2020, 15:45:27 UTC | Remove a workaround for a long-fixed XLA type conversion bug. (#2670) | 10 April 2020, 15:45:27 UTC |
| e904780 | Peter Hawkins | 10 April 2020, 12:37:32 UTC | Update XLA. (#2668) | 10 April 2020, 12:37:32 UTC |
| bd70db7 | Mathis Gerdes | 10 April 2020, 06:16:53 UTC | Port np.roots (#2250) * Implement np.roots. * Expose jit-compatible variant of np.roots. General np.roots implementation has a value dependent output shape. If the input coefficients are guaranteed to have no leading zeros, output shape is independent of values. Skip checking for leading zeros by setting a keyword argument. * Fix typo. * Make roots jit-argument keyword only. Co-Authored-By: Stephan Hoyer <shoyer@google.com> * Format docstring to enable parsing. Co-Authored-By: Stephan Hoyer <shoyer@google.com> * Add np.roots function to documentation. * Add more tests for np.roots function. - Include length 0 polynomial coefficients - Test strip_zeros=False argument - Test jit compiled version (only on cpu due to eigvals) - Confirm that adding leading zeros while skipping check for them results in nan's (expected behavior) * Fix bug in np.roots test. The polynomial with coefficents [0] never fails because the number of roots is 0. * Avoid bug in eigvals and adjust test accuracy. The parameters of the test that was changed are non-essential since they test for how the code behaves given invalid inputs. The accuracy in comparing to the numpy result is changed because the algorithm in those cases is slightly changed with respect to the original numpy algorithm (to allow jit). Co-authored-by: Stephan Hoyer <shoyer@google.com> | 10 April 2020, 06:16:53 UTC |
| e0ce254 | Jake VanderPlas | 10 April 2020, 05:50:10 UTC | Add implementations of np.convolve & np.correlate | 10 April 2020, 05:50:10 UTC |
| 65692db | Matthew Johnson | 10 April 2020, 05:02:47 UTC | Merge pull request #2666 from djdongjin/colab-tpu Remove unnecessary pip code | 10 April 2020, 05:02:47 UTC |
| 1b721b8 | Matthew Johnson | 10 April 2020, 05:02:18 UTC | Merge pull request #2665 from sschoenholz/master Add "callback_transform", two examples, and tests. | 10 April 2020, 05:02:18 UTC |
| f5908cb | Sam Schoenholz | 09 April 2020, 22:20:41 UTC | Fixed copyright, added type annotations, disabled mypy on one line. | 09 April 2020, 22:20:41 UTC |
| 60d856a | Jin Dong | 09 April 2020, 22:10:49 UTC | remove from __future__ code | 09 April 2020, 22:16:47 UTC |
| a23a6c2 | Jin Dong | 09 April 2020, 22:07:47 UTC | remove pip install | 09 April 2020, 22:16:38 UTC |
| 940640d | Sam Schoenholz | 09 April 2020, 20:52:14 UTC | Added tests of callback_transform(jit(fn)). | 09 April 2020, 20:52:14 UTC |
| 06b388c | Sam Schoenholz | 09 April 2020, 20:28:45 UTC | Removed temporary file. | 09 April 2020, 20:28:45 UTC |
| 25d797b | Sam Schoenholz | 09 April 2020, 20:27:16 UTC | Added callback_transform, examples, and tests. | 09 April 2020, 20:27:16 UTC |
| 1bb6763 | Peter Hawkins | 09 April 2020, 20:21:30 UTC | Add batch_group_count to conv_general_dilated. (#2635) * Add batch_group_count to conv_general_dilated. * Use batch_group_count for RHS grouped convolution transpose rule. * Implement lhs/rhs transpose and batching rules for batch_group_count convolution. | 09 April 2020, 20:21:30 UTC |
| 1694a56 | Andy Jones | 09 April 2020, 19:36:09 UTC | Fixes a fallthrough in the tensordot axes verification logic (#2658) | 09 April 2020, 19:36:09 UTC |
| dd92a03 | Stephan Hoyer | 09 April 2020, 17:18:07 UTC | Docstring for test_util.check_grads (#2656) Fixes https://github.com/google/jax/issues/2648 | 09 April 2020, 17:18:07 UTC |
| f7e7e08 | Matthew Johnson | 09 April 2020, 16:53:04 UTC | Merge pull request #2655 from shoyer/cg-complex Renable custom_linear_solve and cg with complex values | 09 April 2020, 16:53:04 UTC |
| c9c14c0 | Matthew Johnson | 09 April 2020, 14:31:23 UTC | Merge pull request #2602 from google/float-and-complex-builtins-error make float and complex builtins error on Tracers | 09 April 2020, 14:31:23 UTC |
| 5965225 | George Necula | 09 April 2020, 12:10:52 UTC | Fix type hints for pytype * We have to be careful for return types to specify that we return Tuple[T, ...] instead of Sequence[T], at least in those places where the caller assumes that the result is a tuple. | 09 April 2020, 12:48:38 UTC |
| abbc70b | George Necula | 18 March 2020, 06:11:44 UTC | Added type annotations and comments related to partial evaluation. Introduced two new constructors for PartialVal: unknown and known. These should make it easier to read the code where we construct PartialVal: * instead of PartialVal((aval, core.unit) we use PartialVal.unknown(aval) * instead of PartialVal((None, pval)) we use PartialVal.known(pval) Also disabled some new tests in random_tests.py on Mac. They segfault, apparently due to the same issue #432. | 09 April 2020, 10:00:33 UTC |
| 9cc5e90 | Stephan Hoyer | 09 April 2020, 07:46:10 UTC | Renable custom_linear_solve and cg with complex values | 09 April 2020, 07:53:00 UTC |
| 7cf5a94 | Stephan Hoyer | 09 April 2020, 07:28:30 UTC | Use custom_jvp for lax.custom_root. (#2485) This gives us a batching rule, for free! | 09 April 2020, 07:28:30 UTC |
| 7ab6775 | Matthew Johnson | 04 April 2020, 04:33:32 UTC | make float and complex builtins error on Tracers cf. #2508 | 09 April 2020, 07:09:51 UTC |
| 0958f27 | Stephan Hoyer | 09 April 2020, 07:07:30 UTC | Disable gradients of custom_linear_solve for complex dtypes (#2573) * Failing test case for custom_linear_solve with complex dtype xref https://github.com/google/jax/issues/2572 * raise an error for complex values * Fix dtype check | 09 April 2020, 07:07:30 UTC |
| de82c4f | sschoenholz | 09 April 2020, 06:29:28 UTC | Merge pull request #1 from google/master Update branch to head. | 09 April 2020, 06:29:28 UTC |
| 9191843 | Matthew Johnson | 09 April 2020, 06:16:49 UTC | Merge pull request #2654 from google/pfix fix jaxpr invar avals | 09 April 2020, 06:16:49 UTC |
| 5f1f29e | Matthew Johnson | 09 April 2020, 05:29:07 UTC | fix jaxpr invar avals When an input argument is known to partial_eval, the corresponding jaxpr (which represents only the staged-out, unknown part of the computation) has a dummy placeholder input with type (i.e. aval) abstract_unit. However, the avals attached to the invars were recording the aval of the known value in that case. cf. #2299 | 09 April 2020, 05:29:07 UTC |
| f37f235 | Skye Wanderman-Milne | 08 April 2020, 18:29:02 UTC | Fix up previous jaxpr.rst commit. (#2647) | 08 April 2020, 18:29:02 UTC |
| 5174f6d | Peter Hawkins | 08 April 2020, 18:13:15 UTC | Add type annotations to user-facing functions in lax.py (#2644) * Add type annotations to user-facing functions in lax.py * Remove redundant comment. | 08 April 2020, 18:13:15 UTC |
| fcefc7e | Peter Hawkins | 08 April 2020, 17:02:58 UTC | Add a link to Google's codesearch for JAX. (#2646) | 08 April 2020, 17:02:58 UTC |
| e9910e0 | Matthew Johnson | 08 April 2020, 05:30:43 UTC | Merge pull request #2642 from jakevdp/trunc Add implementation of np.trunc | 08 April 2020, 05:30:43 UTC |
| 52e779c | Matthew Johnson | 08 April 2020, 04:01:19 UTC | Merge pull request #2637 from jacobjinkelly/morejets Add jet rules for expm1 and log1p | 08 April 2020, 04:01:19 UTC |
| aabc3ac | Matthew Johnson | 08 April 2020, 04:00:52 UTC | Merge pull request #2631 from jakevdp/copysign Implement np.copysign | 08 April 2020, 04:00:52 UTC |
| 46a8922 | Jake VanderPlas | 08 April 2020, 03:50:23 UTC | Add implementation of np.trunc | 08 April 2020, 03:50:23 UTC |
| f8dc650 | Skye Wanderman-Milne | 08 April 2020, 02:03:41 UTC | Update scan jaxpr documentation. (#2641) Closes #2640. | 08 April 2020, 02:03:41 UTC |
| f0ab673 | Skye Wanderman-Milne | 08 April 2020, 01:21:04 UTC | Fully stage out scan under jit. (#2638) This prevents creating large arrays from the results of partial_eval. | 08 April 2020, 01:21:04 UTC |
| ab090e2 | Jacob Kelly | 07 April 2020, 21:55:43 UTC | Merge branch 'master' of https://github.com/google/jax into morejets | 07 April 2020, 21:55:43 UTC |
| 4d7b63c | Jacob Kelly | 07 April 2020, 21:55:07 UTC | add expm1 and log1p | 07 April 2020, 21:55:07 UTC |
| 2512ec6 | Peter Hawkins | 07 April 2020, 17:25:50 UTC | Glob over subdirectories in top-level BUILD file. (#2636) Makes BUILD file more robust to directory structure changes. | 07 April 2020, 17:25:50 UTC |
| 49c6c22 | Jake VanderPlas | 07 April 2020, 16:50:40 UTC | Set check_dtypes=False in copysign | 07 April 2020, 16:50:40 UTC |
| f20a1c9 | Peter Hawkins | 07 April 2020, 15:21:39 UTC | Enable some GPU and TPU tests that now pass. (#2633) | 07 April 2020, 15:21:39 UTC |
| bd00e59 | Jake VanderPlas | 07 April 2020, 13:44:44 UTC | Implement np.copysign | 07 April 2020, 13:44:44 UTC |
| fa383b4 | Peter Hawkins | 07 April 2020, 13:38:10 UTC | Mark primitive parameters as keyword-only arguments in rules in lax.py. (#2625) * Mark primitive parameters as keyword-only arguments in rules in lax.py. * Fix dynamic update slice batching rule. * Fix dynamic slice batching rule. | 07 April 2020, 13:38:10 UTC |
| 5c74c39 | Jake Vanderplas | 07 April 2020, 12:56:23 UTC | Make median() more efficient & handle infs correctly (#2626) | 07 April 2020, 12:56:23 UTC |
| f3a5765 | Jake VanderPlas | 07 April 2020, 04:29:24 UTC | jit: raise TypeError if called on generator function | 07 April 2020, 10:55:35 UTC |
| 44e761b | Peter Hawkins | 07 April 2020, 01:45:10 UTC | Enable a disabled convolution test. (#2624) | 07 April 2020, 01:45:10 UTC |
| bbed6f8 | Jake Vanderplas | 07 April 2020, 01:40:10 UTC | Add support for interpolation options in quantile() and percentile() (#2610) | 07 April 2020, 01:40:10 UTC |
| 6213f8b | Jin Dong | 07 April 2020, 00:26:51 UTC | Remove unnecessary code in colabs (#2623) * fix misspell in autodiff_cookbook[modify colab directly] * remove unnecessary from __future__ code[modify colab directly] * change tf&tfds-nightly to stable version | 07 April 2020, 00:26:51 UTC |
| 3fe8bd0 | Skye Wanderman-Milne | 06 April 2020, 23:38:34 UTC | Adjust pmap_bechmark.py values to be more realistic. (#2622) | 06 April 2020, 23:38:34 UTC |
| e8f989e | Stephan Hoyer | 06 April 2020, 21:45:02 UTC | Add import from scipy.sparse (#2621) * Add import from scipy.sparse * Fix formatting in cg docstring | 06 April 2020, 21:45:02 UTC |
| 1472eb3 | Stephan Hoyer | 06 April 2020, 19:49:11 UTC | DOC: note how derivatives are computed for CG (#2619) | 06 April 2020, 19:49:11 UTC |
| cf4dd84 | Peter Hawkins | 06 April 2020, 19:14:22 UTC | cumsum is linear, so its gradient can be linear also. (#2618) * cumsum is linear, so its gradient can be linear also. * Rename _impl functions to _prefix_scan. | 06 April 2020, 19:14:22 UTC |
| fc23e07 | Igor Babuschkin | 06 April 2020, 18:48:42 UTC | Also enable nan checking for complex numbers (#2616) | 06 April 2020, 18:48:42 UTC |
| 7629c5a | Peter Hawkins | 06 April 2020, 16:39:28 UTC | Add some missing functions to documents. (#2615) | 06 April 2020, 16:39:28 UTC |
| 36c529d | Peter Hawkins | 06 April 2020, 16:33:55 UTC | Handle n==0 case in TPU cumsum/cumprod. (#2617) | 06 April 2020, 16:33:55 UTC |
| 329321b | Peter Hawkins | 06 April 2020, 15:22:01 UTC | Add backend-specific lowering for cumsum/cumprod on TPU. (#2614) * Add backend-specific lowering for cumsum/cumprod on TPU. Make cumsum/cumprod primitives so they can have backend-specific lowerings. * Disable cumulative reduction gradient test on TPU. | 06 April 2020, 15:22:01 UTC |
| f5f35c5 | George Necula | 02 April 2020, 12:44:36 UTC | Adapted vmap out_axes check to sum_match | 06 April 2020, 09:11:29 UTC |
| 7a5d1bc | George Necula | 28 March 2020, 15:50:31 UTC | Expand docstring for vmap with details about out_axes, and improve error checking The newly added test cases used to raise the following kinds of exceptions: AttributeError: 'float' object has no attribute 'shape' ValueError: (0, None) ValueError: vmap got inconsistent sizes for array axes to be mapped: arg 0 has shape (2,) and axis None is to be mapped so TypeError: only integer scalar arrays can be converted to a scalar index. | 06 April 2020, 09:11:29 UTC |
| 99944d1 | Sharad Vikram | 05 April 2020, 06:19:39 UTC | Fix lax.broadcast_shapes returning numpy ints in shape tuple (#2471) * Fix lax.broadcast_shapes returning numpy ints in shape tuple * Use _canonicalize_dimension and add test | 05 April 2020, 06:19:39 UTC |
| 1cf708e | Stephan Hoyer | 04 April 2020, 22:55:46 UTC | Support pytrees in jax.scipy.linalg.cg (#2600) * Support pytrees in jax.scipy.linalg.cg Ideally there would be an easier way to write this, but for now this will do. * Fixup test | 04 April 2020, 22:55:46 UTC |
| 2c4ced2 | Matthew Johnson | 04 April 2020, 03:18:21 UTC | Merge pull request #2601 from adarob/fix-err Fix error handling when an attempt is made to pmap a scalar. | 04 April 2020, 03:18:21 UTC |
| 48eb524 | Adam Roberts | 04 April 2020, 01:03:42 UTC | Fix error handling when an attempt is made to pmap a scalar. | 04 April 2020, 01:03:42 UTC |
| c2f56fb | Matthew Johnson | 03 April 2020, 23:21:38 UTC | add notes to changelog | 03 April 2020, 23:21:38 UTC |
| 60de46a | Matthew Johnson | 03 April 2020, 22:47:41 UTC | Merge pull request #2591 from google/tracer-printing make tracers tree-pretty-print their contents | 03 April 2020, 22:47:41 UTC |
| b67250e | Matthew Johnson | 03 April 2020, 22:21:45 UTC | Merge pull request #2599 from sharadmv/logit-fix Fix grad(logit) to use defjvps and enable it in tests | 03 April 2020, 22:21:45 UTC |
| 3c9fb35 | Sharad Vikram | 03 April 2020, 21:37:29 UTC | Fix dtype error | 03 April 2020, 21:37:29 UTC |
| 1b93bb5 | Stephan Hoyer | 03 April 2020, 20:37:11 UTC | Implement scipy.sparse.linalg.cg (second try) (#2566) * super minimal starter code * Update optimizers.py * implement flip with axis = None * Create sparse.py * fix some imports * Update sparse.py * add partial function & test * Update lax_scipy_sparse_test.py * Update lax_scipy_sparse_test.py * add a test case for sparse pd matrix & add bigger dim * address comments * fix info return & create matrix with rng_factory * Update lax_scipy_sparse_test.py * Update lax_scipy_sparse_test.py * Update sparse.py * Update sparse.py * Update sparse.py * Update lax_scipy_sparse_test.py * Update lax_scipy_sparse_test.py * cast jax arrays into numpy array for scipy compatibility * Update sparse.py * Update sparse.py * fix None issue, but algo is not working * fix return of build_and_solve and output of while_loop * fix condition func of while loop * clearer variable names * mismatch error * Update lax_scipy_sparse_test.py * Fixes to jax.experimental.sparse.cg * Fix tests for gradients * Add support for preconditioners to cg * Move cg into scipy, update docs * doc tweak Co-authored-by: Tuan Nguyen <anhtuan277@gmail.com> | 03 April 2020, 20:37:11 UTC |
| 72783bb | Sharad Vikram | 03 April 2020, 20:27:02 UTC | Fix grad(logit) to use defjvps and enable it in tests | 03 April 2020, 20:27:02 UTC |
| 2b3beff | Peter Hawkins | 03 April 2020, 20:09:48 UTC | Make reduce_prod differentiable to arbitrary order. (#2597) * Make reduce_prod differentiable to arbitrary order. The previous strategy for computing the JVP of reduce_prod used a pair of reduce_window operations to form left and right products for each position. This PR instead builds an explicit reduction tree and differentiates through it, which while not as efficient as using XLA's built-in reductions, has the advantage of being differentiable to arbitrary order. . * Return the tree-reduction primals instead of returning the original primals in JVP rule. | 03 April 2020, 20:09:48 UTC |
| 824ac86 | Peter Hawkins | 03 April 2020, 19:39:56 UTC | Reimplement np.cumsum and np.cumprod in terms of a parallel prefix scan. (#2596) * Reimplement np.cumsum and np.cumprod in terms of a parallel prefix scan. Unlike the existing implementation based on lax.reduce_window, this implementation is O(n log n) instead of O(n^2) and is arbitrarily differentiable. Fixes #1212, #2418, #2542. May help with issue #2380. * Relax gradient test tolerance. | 03 April 2020, 19:39:56 UTC |
| 192e908 | Matthew Johnson | 03 April 2020, 06:53:39 UTC | Merge pull request #2561 from madisonmay/batch-norm-no-scale-or-center FIX: batch norm w/ no scale, center | 03 April 2020, 06:53:39 UTC |
| 6402419 | Matthew Johnson | 03 April 2020, 06:52:40 UTC | Merge pull request #2563 from google/callable-typechecks add callable typechecks to more api.py functions | 03 April 2020, 06:52:40 UTC |
| 25aeb80 | Matthew Johnson | 03 April 2020, 06:42:31 UTC | Merge pull request #2592 from google/travis-mypy add trace state check tearDown to JaxTestCase | 03 April 2020, 06:42:31 UTC |
| 3dee689 | Matthew Johnson | 03 April 2020, 06:28:57 UTC | Merge pull request #2593 from google/issue2578 add full lower to custom_jvp/vjp call bind | 03 April 2020, 06:28:57 UTC |
| ba8225f | Matthew Johnson | 03 April 2020, 06:11:55 UTC | skip all parallelize tests (abandonware right now) | 03 April 2020, 06:11:55 UTC |
| 0e49133 | Matthew Johnson | 03 April 2020, 05:52:07 UTC | add full lower to custom_jvp/vjp call bind fixes #2578 | 03 April 2020, 05:54:29 UTC |
| f2de1bf | Matthew Johnson | 03 April 2020, 05:01:43 UTC | add trace state check tearDown to JaxTestCase | 03 April 2020, 05:01:43 UTC |
| 297c902 | Matthew Johnson | 03 April 2020, 04:04:12 UTC | make tracers tree-pretty-print their contents | 03 April 2020, 04:04:12 UTC |
| 64a7d17 | Matthew Johnson | 03 April 2020, 04:02:13 UTC | Merge pull request #2587 from google/travis-mypy re-enable travis mypy testing (typo broke it) | 03 April 2020, 04:02:13 UTC |
| 5d3f1bd | Matthew Johnson | 03 April 2020, 03:14:12 UTC | tell mypy: using __init__ to reinitialize is OK | 03 April 2020, 03:14:12 UTC |
| 6d4987c | Matthew Johnson | 03 April 2020, 01:19:44 UTC | make core.trace_state resetting be thread-local | 03 April 2020, 01:19:44 UTC |
| b78b7a0 | Matthew Johnson | 03 April 2020, 01:03:58 UTC | add global trace state checks to more tests | 03 April 2020, 01:03:58 UTC |
| ab0a005 | Matthew Johnson | 03 April 2020, 00:18:47 UTC | check sublevel is reset in loops_test.py | 03 April 2020, 00:18:47 UTC |
| c72abf6 | Matthew Johnson | 02 April 2020, 22:52:01 UTC | re-enable travis mypy testing (typo broke it) | 02 April 2020, 22:52:01 UTC |
| 5c0ac40 | Skye Wanderman-Milne | 02 April 2020, 19:54:58 UTC | Revert jax.numpy.matmul to pre-#2512 version. (#2584) https://github.com/google/jax/pull/2512 was causing some Google-internal tests to take longer. | 02 April 2020, 19:54:58 UTC |
| 0bdd0f6 | Matthew Johnson | 02 April 2020, 16:03:12 UTC | Merge pull request #2581 from google/jet-process-call post process call of jet! | 02 April 2020, 16:03:12 UTC |
