mlem
mlem copied to clipboard
Potential for a discrepancy in hash and size when artifact is store with dvc
In dvc branch both .mlem/model/rf.dvc and .mlem/model/rf.mlem have hash and size for the same artifact. This feels fragile. One might update one of this independently, say *.dvc file:
cd .mlem/model
dvc pull
<edit> rf
dvc add rf
dvc push
git add rf.dvc && git commit ...
What do you thing @aguschin?
I moved this issue from example-mlem-get-started repo here. The problem you mention @Suor definitely exists.
- I think we could issue a warning when MLEM loads a model, but binary have a hash different to what's in
.mlemfile (and maybe we already do, need to check this). - We could add
--check-hash-and-fail-if-it-changedflag to MLEM commands or something with a shorter name
Do you maybe have some ideas also?