Please add me to the communishift group. Project Thoth wants to work on hosting a public instance of Thoth to the python and AI/ML community.
How is this used in Fedora?
We offer a CLI tool called thamos that could be used during application development to lock down the Python dependencies. The lockdown via thamos will incorporate the observations/findings we have made. This CLI is also integrated S2I so it could be used during OpenShift builds.
We would like to explore with the ML-SIG how to integrate thamos in Python-based Fedora projects.
So the answer is more like '(hopefully) it will be used in Fedora' rather than 'it is used', deploying this service is a pre requisite to explore its usage/value to Fedora.
Hi Christoph! Can you expand a little bit on how this service will integrate with / benefit the Fedora community?
With Thoth we create observations on software stacks (right now focused on AI/ML application stacks). These observations are wrt but not limited to build failures, the performance of the whole software stack, but also of individual packages.
These observations could be use to...
a) steer focus of Fedora's engineers: what packages limit the performance of ML applications? Where do we need to invest in tuning parameters? Can we help a specific framework wrt manylinux2014 compliance?
b) guide new developers by giving advise what versions of his dependencies are know to work well together
c) identify ABI glitches in a software stack at build time, and not at run time
Metadata Update from @kevin:
- Issue priority set to: Waiting on Assignee (was: Needs Review)
d) we could integrate Thoth advises with Packit (short term) and Ansible roles used by Fedora-infra team (long term)
to comment on this ticket.