heat

heat copied to clipboard

Distributed Compressed Sparse Column Matrix

17

## Due Diligence - General: - [x] **title** of the PR is suitable to appear in the [Release Notes](https://github.com/helmholtz-analytics/heat/releases/latest) - Implementation: - [x] unit tests: all split configurations tested -...

Mystic-Slice

Declutter automated release notes

1

With the many dependency updates, our [Draft release notes](https://github.com/helmholtz-analytics/heat/releases) are getting a bit cluttered. I think it's important to keep track of the CI changes, not sure about all the...

ClaudiaComito

Resolved PyTorch meshgrid warning and indexing issue in `heat.meshgrid`

9

## Due Diligence - General: - [X] **base branch** must be `main` for new features, latest release branch (e.g. `release/1.3.x`) for bug fixes - [X] **title** of the PR is...

samadpls

Refactoring of QR: stabilized Gram-Schmidt for split=1 and TS-QR for split=0

24

## Description ### Changes in the implementation of QR: * `split=1`: column-block-wise stabilized/modified Gram Schmidt; thus any shape allowed. * `split=0`: TS-QR ("tall skinny QR"); thus input must be tall-skinny...

mrfh92

[Feature-384 Sparse] Allow `DNDarray` operands in arithmetic operations with `DCSR_matrix`

1

**Related** Pending work from https://github.com/helmholtz-analytics/heat/pull/1028 **Feature functionality** The sparse binary operator allows only `DCSR_matrix` and scalars. Expand the operator to allow `DNDarray`s too. **Additional context** Add any other context or...

Mystic-Slice

[Feature-384 Sparse] Array broadcasting for sparse arithmetic operations

1

**Related** Pending work from https://github.com/helmholtz-analytics/heat/pull/1028 **Feature functionality** The sparse binary operator expects both operands to be of the same shape. Implement array broadcasting for sparse arithmetic operations to allow operands...

Mystic-Slice

Support Apple's MPS backend

6

Since version 1.12, PyTorch supports GPU acceleration via Apple's Metal Performance Shaders (MPS) backend. https://pytorch.org/blog/introducing-accelerated-pytorch-training-on-mac/ **Feature functionality** `device="gpu"` should be a valid argument for `ht.factories` if the MPS backend is...

ClaudiaComito

[Bug]: parallel write-out to netcdf fails in tests

3

### What happened? Our own tests (`test_io.test_save`) fail when parallel netCDF is enabled. Tested on HDFML cluster with Stage 2022 stack and `netcdf4-python` module. ``` ml Stages/2022 ml GCC ml...

ClaudiaComito

Improve Alltoallv documentation of `sendbuf`, `recvbuf`

5

**Feature functionality** Currently, our documentation is pretty much identical for `communication.Alltoall` and `communication.Alltoallv`. Improve `communication.Alltoallv` docs to indicate different construction of `sendbuf`, `recvbuf`. **Additional context** [OpenMPI Alltoallv](https://www.open-mpi.org/doc/v4.0/man3/MPI_Alltoallv.3.php) vs. [Alltoall](https://www.open-mpi.org/doc/v4.0/man3/MPI_Alltoall.3.php)

ClaudiaComito

Support Apple MPS acceleration

14

## Description **RESEARCH FEATURE** This is a first attempt at extending PyTorch's support of Apple MPS acceleration to Heat users. **NOTE: distributed operations between MPS have not been tested!** This...

ClaudiaComito