jakevdp
Follow
Pinned
1,769 contributions in the last year
Contribution activity
March 2023
Created 64 commits in 2 repositories
Created a pull request in google/jax that received 2 comments
Add repr for UnspecifiedValue
This helps make low-level errors with jaxpr dumps more readible. Before:
ValueError: Cannot lower jaxpr with verifier errors: #loc = loc(unknown)
.…
+2
−1
•
2
comments
Opened 68 other pull requests in 2 repositories
google/jax
6
open
35
merged
1
closed
- Use ml_dtypes definition for jnp.finfo
- [sparse] fix coo efficiency warning
- jax.typing: recommend instance check in Python 3.10 or newer
- jax.scipy.linalg.expm: support batched inputs
- lax_numpy: move quantile-based functions to reductions.py
- Fix mypy issue in jax/experimental/jet.py
- jax.random: clean up handling of dtypes
- internal: refactor array methods into separate private submodule
- DOC: add formulae for distributions in jax.random
- jax.random: remove scale from wald function
- Document ShapeDtypeStruct
- jnp.mean: fix incorrect return value for large arrays
- Improve error for indexing with string
- Sharp bits: refer to ndarray.at in out-of-bound indexing discussion
- DOC: remove jax 0.4.1 banner from index page
- Improve error for tolist() and tobytes() on tracer objects
- jnp.einsum: make signature match documentation.
- Add regression test for #4780
- DOC: document array methods
- jax.numpy reductions: validate axis for scalar input
- pytree doc: add discussion of children vs aux_data
- README: improve Colab TPU installation discussion
- [sparse] add BCOO lowering for div
- [typing] add type annotations to index_update code
- Remove leading underscores in jax._src.numpy.util
- Some pull requests not shown.
jax-ml/ml_dtypes
23
merged
3
closed
- Bump version to 0.0.4
- Add py.typed file to distribution
- ml_dtypes.finfo: extension of np.finfo for new types.
- Bump version to 0.0.3
- Add license section to readme
- bump version to v0.0.2
- build: ensure numpy is compatible with python version
- build: pin numpy build version correctly
- CI: switch release job to public PyPI
- Remove obsolete CI configuration
- CI: trigger PyPI publishing on github release
- Append Eigen license to package license
- Add test of pickleability
- README: update description and add brief example
- CI: build OSX universal wheels
- Add pypi-test release trigger
- Test PyPI auto-release
- CI: add automated wheel build for windows
- Fix Windows build & add CI tests
- CI: add buildwheel action
- Add ml_dtypes.__version__
- actions: fix pytest_and_autopublish
- Test: ignore warnings
- Add simple pytest github action
- README: add local installation instructions
- Some pull requests not shown.
Reviewed 45 pull requests in 2 repositories
google/jax
25 pull requests
- Use ml_dtypes definition for jnp.finfo
- Update user_guides.rst
- Fix hessian llnk
- Suppress mypy warnings about missing imports.
-
hilbert transform similar to that in
osp.signal - hilbert transform
- Adds the trailing-whitespace pre-commit hook
-
Fix
locandscaleparameters inscipy.logistic.*. Add CDF and SF for several distributions - [NVIDIA GPU] Add missing file for the nightly CI.
- WAR the dependency issue in the nightly CI container.
-
Add
walddistribution torandom.py - Document ShapeDtypeStruct
- explicitly convert jnp.var scalar normalizer to float (from int)
- Support extrapolation in jnp.interp
- Typo fix in ResizeMethod docstring, scale.py
- Fix typing for register_pytree_with_keys
- DOC: document array methods
- jnp.arange: better validation of inputs
-
Add
rayleighdistribution torandom.py - Correct register_pytree_with_keys annotation
- update docs to remove stale reference to laziness optimization
- Correct multidimensional Rosenbrock function
- jnp.argmin/max: correctly handle out argument
- Add xla garbage collection to gc.callback
- Fix jnp.sort & jnp.vdot in no-jit mode
- Some pull request reviews not shown.
Created an issue in google/jax that received 3 comments
device_put leaks references on numpy inputs
First reported in #14876; here's a minimal repro that reproduces on CPU and GPU: import sys import jax import numpy as np x = np.arange(1000, dtype='…
3
comments
Opened 2 other issues in 2 repositories
jax-ml/ml_dtypes
1
open
google/jax
1
open
Answered 18 discussions in 1 repository
google/jax
google/jax
-
How to sort rows of a sparse matrix / make sure sparse matrix-vector multiplication is working optimally
This contribution was made on Mar 27
-
Since when is this form of dynamic indexing possible under jit?
This contribution was made on Mar 26
-
Convert General PyTree to builtins
This contribution was made on Mar 24
-
Trouble getting correct inverse-gamma samples from `jax.random.gamma`
This contribution was made on Mar 23
-
Element-wise multiplication of BCOO sparse matrices reports out of memory error
This contribution was made on Mar 22
-
pmap_test::PythonPmapTest::testRule30
This contribution was made on Mar 21
-
TypeError: Using a non-tuple sequence for multidimensional indexing is not allowed
This contribution was made on Mar 20
-
Sparsify and gradient calculations
This contribution was made on Mar 17
-
jax.grad with multiple return values
This contribution was made on Mar 16
-
How does `jit` compute cache keys and hits for generic PyTrees?
This contribution was made on Mar 15
-
What is the "JAX" way of defining elementwise vector functions
This contribution was made on Mar 13
-
Create Ragged Arrays via Vmap
This contribution was made on Mar 11
-
jnp.where leads to infinite loop?
This contribution was made on Mar 10
-
composing - piping operations
This contribution was made on Mar 9
-
Does the BCCO sparse matrix not support the operation power?
This contribution was made on Mar 9
-
Anything faster than `cond` for splicing together functions?
This contribution was made on Mar 8
-
How to measure computational cost in JAX?
This contribution was made on Mar 8
-
Speed up computations with dynamic input shapes
This contribution was made on Mar 7