Wip/translation api - #211
Conversation
98a4959 to
3402f2a
Compare
gcampax
left a comment
gcampax
left a comment
There was a problem hiding this comment.
This looks good I think.
3402f2a to
a4212d9
Compare
a4212d9 to
6f61f48
Compare
instead change ids on the go to be unique, lifting the burden from user or dataset
6f61f48 to
122928f
Compare
| 'pathos==0.2.8', | ||
| # for kf: | ||
| 'kfserving>=0.5.0', | ||
| 'kfserving==0.5.1', |
There was a problem hiding this comment.
Please don't do this. kfserving depends on ray, but doesn't specify a version bound on it. ray has broken APIs before, breaking kfserving, and us transitively. If we pin kfserving, we remain broken even if kfserving releases a fixed version, until we change genienlp. This creates a cascade effect and takes a lot of work to fix.
(yes python packaging is the worst)
There was a problem hiding this comment.
what is the solution then? pinning "ray" instead?
There was a problem hiding this comment.
There is no good solution, but at least if we leave kfserving unpinned we'll get any fix they make as soon as they make it.
There was a problem hiding this comment.
but that will leave our tests broken till then. Apparently ray 1.7.0 was released 3 days ago and had major changes. I suggest pinning ray to 1.6.0 until they patch 1.7.0. what do you think?
ce9aa51 to
4f21e70
Compare
Recent release of ray (1.7.0) is breaking our tests due to some import errors. Will revert this change once a new patch is pushed.
4f21e70 to
934f77c
Compare
|
Ok @gcampax, I pinned ray for now until they patch it. I'm gonna merge this branch and rebase the other one. |
2 participants
Accommodating stanford-oval/genie-toolkit#806