integrate dev branch and update software environments #25
Conversation
btraven00
commented
May 13, 2025
btraven00
commented
- updated to use omnibenchmark dev branch
- update the 3 software environments
- make script executable - use /bin/sh instead of /bin/bash - add top-level Makefile to prepare env
|
Ideally this should be squashed into a single commit |
btraven00
force-pushed
the
integration-dev
branch
from
674c4b6 to
d673143
Compare
|
tests will need to be updated to reflect what is intended to run. I suggest any of the smoketest variants, since they should be short (around 1min) |
btraven00
force-pushed
the
integration-dev
branch
from
8a3187c to
f72d0ca
Compare
There was a problem hiding this comment.
Nice touch with the overrides, it goes into the direction of composable yamls that we've discussed. Then, my question, why don't we ignore the generated yamls, since they can be reproduced using the make commands.
Also, there needs to be more documentation about the yaml flavours.
.github/workflows/benchmark.yml
Outdated
| mamba install -y pip | ||
| pip install git+https://github.com/omnibenchmark/omnibenchmark.git@reduce_install_scope | ||
| pip install git+https://github.com/omnibenchmark/omnibenchmark.git@dev | ||
|
|
There was a problem hiding this comment.
We have to change this to track main
| # path: "{input}/{stage}/{module}/{params}/{dataset}.metrics.txt" | ||
| id: clustering_benchmark_apptainer_oras | ||
| description: Clustering benchmark on Gagolewski's. Using apptainer from omnibenchmark ORAS registry. | ||
| software_backend: apptainer |
There was a problem hiding this comment.
Why leave these at the end of the file? I suspect diffing tools
There was a problem hiding this comment.
this i don't like either; it's unfortunate. it depends on which has priority; and I had to change to base + override because in some of the cases we're changing some of the nested (the apptainer in software-envs)
envs/README.md
Outdated
| We distribute `Clustering.yml` runs with different backends. | ||
|
|
||
| - `Clustering_conda.yml`. Conda semi-reproducible (no pinning, pip) | ||
| - `Clustering_singularity.yml`. Singularity semi-reproducible, local SIF files. | ||
| - `Clustering_oras.yml`. Singularity semi-reproducible, prebuilt remote images. | ||
| - `Clustering_envmodules.yml`. Easybuilt with default optimization. | ||
| - `Clustering_envmodules.yml`. Easybuild backend with default optimization. | ||
| - `Clustering_apptainer.yml`. Apptainer, pinned, prebuilt remote images from [omnibenchmark's registry](https://quay.io/organization/omnibenchmark). | ||
| - `Clustering_apptainer_vanilla.yml`. Singularity, pinnned, from local SIF images. | ||
| - `Clustering_conda.yml`. Conda semi-reproducible (no pinning, using pip) |
There was a problem hiding this comment.
Can we document what _apptainer_short.yml and _apptainer_optimized.yml flavours are doing?
There was a problem hiding this comment.
done, have a look!
…hmark#25) * run from post-0.2.0 tag, main branch * docs: use public repo URIs * chore: add convenience target to build environments * add top-level Makefile to prepare env * feat: parametrize num of cores on the makefile * chore: ignore common temporary outputs and image build artifacts * update .eb files to easybuild 5.0 * remove remote storage * do not run artifacts if not in main repo * inject checksums to rmarkdown easyconfig * update sklearn singularity definition * feat: add microbenchmark for numpy operations * chore: bump clustering-benchmarks to 1.1.6 * feat: templatize the definitions * feat: mv output folders to timestamped names * feat: add --yes flag * docs: update README
