Divyansh Khanna
Divyansh Khanna
@TanmayPani Thanks for explaining! TIL about vmap today. Sounds like a real good use-case. I'm curious, when stacking M varied batches together to practically train N models concurrently, how do...
@TanmayPani thanks for the code. I will go through it, in the meantime I think for this feature we should think of all the features that would be needed (stacking...
@TanmayPani Let's create the PR! Also check out any recent nodes added for any patterns / unit testing etc. cc. @ramanishsingh
Thanks for the question. By data do you mean actual training data or code from Github which can be also used as data ? If it the former, then I'm...
Hey @mirceamironenco we are working through some PRs before cutting v0.12. We are looking at early next month. Thanks - stay tuned!
@ramanishsingh can clarify on how we don't "lose data" and we resume on correct data. But regarding saving actual data in ckpt, that is not feasible because data can be...
@pytorchmergebot rebase