Experiment duplicates

Hi! I am trying to integrate experiments usage, but faced with some issues.

I have the following pipeline:

    cmd: python train.py data/dataset/ --config config.yaml --out data/model.pt
    - data/dataset/
    - train.py
    - data/model.pt

I run experiment and push:

dvc exp run -n v1.0 -S config.yaml:model.dim=64
dvc exp push origin v1.0

The I try on other server run the same exp:

dvc exp pull origin v1.0
dvc exp run -n v1.1 -S config.yaml:model.dim=64 -v —dry
train stage modified
git status
changes in config.yaml

i.e. dvc thinks that there are changes in config and runs pipeline, but experiment was pulled and in a cache

But if I apply this exp:

dvc exp apply v1.0
dvc chekout
dvc exp run -n v1.1 -S config.yaml:model.dim=64 -v —dry
no changes, get from cache

i.e. dvc sees that experiment is duplicate only if apply it

I expected that dvc allows not to run exps if it was run already. So, my questions are:

  • what am I doing wrong?
  • if smth is ok, why dvc can’t detect that new experiment was already run? Is this feature going to be implemented?