ClearML - Auto-Magical CI/CD to streamline your AI workload. Experiment Management, Data Management, Pipeline, Orchestration, Scheduling & Serving in one MLOps/LLMOps solution
It would be nice to have the option of using uv to manage dependencies instead of pip/conda/poetry.
The simplest support would be to just use uv pip as a drop-in replacement for pip when installing from a task's requirements in the clearml agent.
Motivation
Much faster environment build times on slow file systems.
On my university cluster all storage on most GPU nodes is network-attached with high throughput but very slow random access, and building my environment using pip hits a massive performance cliff and can easily take over an hour to complete even with all downloads already cached, compared to uv which can do this in about a minute.
I've resorted to handrolling my own clearml agent replacement for now because environment building takes so long with pip (and environments are rebuilt from scratch with every change)
Related Discussion
If this continues a slack thread, please provide a link to the original slack thread.
Proposal Summary
It would be nice to have the option of using uv to manage dependencies instead of pip/conda/poetry.
The simplest support would be to just use
uv pip
as a drop-in replacement for pip when installing from a task's requirements in the clearml agent.Motivation
Much faster environment build times on slow file systems.
On my university cluster all storage on most GPU nodes is network-attached with high throughput but very slow random access, and building my environment using pip hits a massive performance cliff and can easily take over an hour to complete even with all downloads already cached, compared to uv which can do this in about a minute.
I've resorted to handrolling my own clearml agent replacement for now because environment building takes so long with pip (and environments are rebuilt from scratch with every change)
Related Discussion
If this continues a slack thread, please provide a link to the original slack thread.