thomas chaton
thomas chaton
Yes, LitData encodes each leaf of the pytree as a single object and therefore, it doesn't know this is a single sample. You can convert it to numpy or torch...
Hey @gluonfield Litdata supports parquet files now. So u don t need to optimise it anymore
Any updates ? This is quite important for us. I have a solution to make it work, but it is racy. I can also work on contributing this.
Hey @thaJeztah @GordonTheTurtle Any updates ?
Thanks @vvoland . Do you have docs to create a custom snapshotter ?
Hey @vvoland Maybe this would work: https://github.com/containerd/containerd/blob/c673fcc9af3ec4a44ac112bd1e8e6479862a10af/docs/snapshotters/README.md?plain=1#L42C8-L45C27
Hey @sanggusti. Would you mind adding a bit more info to the description. Also, it seems the tests are failing.
We can adapt: https://github.com/Lightning-AI/pytorch-lightning/blob/master/.github/CONTRIBUTING.md
Nice @deependujha !
Hey @ygtxr1997. You can override the distributed env on the dataset. It is inferred automatically from torch. What is your use case ?