Olivier Grisel

Here is an archive of a particularly bad recent run: https://gist.github.com/ogrisel/371405ac68b5c7a977b907c843cdf798 Observed on #385 but probably unrelated to the change in this PR.

`torch.linalg.svd` is significantly slower than expected on a Max Series GPU

1

### Describe the bug I tried `torch.linalg.svd` a Max Series GPU using the Intel Devcloud and packages from the `intel` conda channel, and while I cannot reproduce the segfault, the...

Update requirements.txt to add expecttest

1

It is apparently a test dependency: #416.

Support reproducible builds to automate security auditing of binary artifacts

2

### Your question: Has the conda-forge project any plans to make (some of) the builds [byte-for-byte reproducible](https://reproducible-builds.org)? And if so (or if they already are in some cases) has the...

Feature request: memmap support when loading models

6

Make it possible to unzip and memmap the numpy arrays in a `.skops` file to make it possible to load models with their large parameter arrays in shared memory between...

Find a way to skip host key check for the "starcluster sshmaster mycluster" command

4

Most of the time the EC2 host for the master node is newly provisioned, hence we get an annoying message such as the following: ``` The authenticity of host 'ec2-23-22-250-167.compute-1.amazonaws.com...

This might be fixed by calling `ldconfig` once all the packages are installed.

See: https://github.com/scikit-learn/sklearn-pypi-package/blob/main/README.md

Disable public IP access by default

3

For security reasons, I think that the default configuration should not map the jupyter / scheduler services on a public IP address (even if jupyter notebook asks for a password,...

Add default value for StoreBackendBase.compress

1

Fixes #1127. Ideally we would need update the tests but the bare try / except in `StoreBackendBase.dump_item` makes this difficult and looks like an anti-pattern anyway. However I am not...