Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

add pull_request event #4

Open
wants to merge 26 commits into
base: main
Choose a base branch
from
Open

add pull_request event #4

wants to merge 26 commits into from

Conversation

skye
Copy link
Owner

@skye skye commented Oct 24, 2022

No description provided.

jakevdp and others added 26 commits October 21, 2022 11:59
This ensures all existing JAX buffer types have a `delete` method that can be used to free device buffer allocation eagerly.

User code sometimes have lingering python refs due to cyclic deps and other reasons, yet users may know for sure that certain arrays will no longer be used after a certain point. Calling `foo_array.delete()` for DeviceArray/ShardedDeviceArray/GlobalDeviceArray/Array allows users to force free the device side allocation to minimize device memory usage.

PiperOrigin-RevId: 482892157
…e in CUDA 11.1

PiperOrigin-RevId: 482897448
…ther than trivial computation.

PiperOrigin-RevId: 482919649
The shape function of DotGeneralOp can't be integrated into MHLO yet: the shape function only predicts return shape but not able to predict element type. However, the current python binding infra will generate the constructor __init__() without the `return` as the first arg, which assumes the shape function can provide a fully inferred type (including an accurate element type). This leads to "inferred type does not match actual result type" errors in JAX. This needs a future solution.

This CL is the corresponding change with openxla/stablehlo#269

Related Python __init__() interface changes (used by JAX):
batch_norm_grad:      not used by JAX
batch_norm_inference: not used by JAX
batch_norm_training:  not used by JAX
case:                 no change*
dot_general:          open new b/253644255 to track the issue
if:                   no change*
map:                  no change*
reduce:               no change*
reduce_window:        no change*
sort:                 no change*
triangular_solve:     updated in `linalg.py`
while:                no change*

no change*: the signature of __init()__ for the op is not changed because of existence of regions https://github.com/llvm/llvm-project/blob/main/mlir/tools/mlir-tblgen/OpPythonBindingGen.cpp#L577

PiperOrigin-RevId: 482951512
fix some shape and type issues

import into namespace

imports into non-_src library

working logpdf test

cleanup

working tests for cdf and sf after fixing select

relax need for x to be in (a, b)

ensure behavior with invalid input matches scipy

remove enforcing valid parameters in tests

added truncnorm to docs

whoops alphabetical

fix linter error

fix circular import issue
PiperOrigin-RevId: 483380789
No functional changes intended.

PiperOrigin-RevId: 483413031
PiperOrigin-RevId: 483425197
overriden -> overridden
@skye skye closed this Oct 24, 2022
@skye skye reopened this Oct 24, 2022
@skye skye closed this Oct 24, 2022
@skye skye reopened this Oct 24, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

8 participants