Kiran Karra
Kiran Karra
@sdesrozis Thanks for your quick reply! Sorry if my initial question was unclear. As an example: ``` m = model() m_dp = nn.DataParallel(m) m_ddp = nn.DistributedDataParallel(m) x = input #...
I'd like to call `loss.backward()`
@sdesrozis Thanks - I will investigate based on your link and report back.
Hi @vfdev-5, sure. We are using the [Supervised Contrastive loss](https://arxiv.org/pdf/2004.11362.pdf) to train an embedding. In Eq. 2 of the paper, we see that the loss depends on the number of...
Thanks for this contribution. I'm confused as to what the bug was before, can you describe it in more detail please?
Link for storing data > sector size is broken: This is original link in the webpage that is broken: https://docs.filecoin.io/how-to/store/prepare-data/#for-files-bigger-than-a-sector
> I might be missing something obvious but introducing a state machine here seems like unnecessary complexity and state, e.g. I don't disagree with anything you've stated, but this is...
I've updated the description and removed the pseudocode. Is it clearer now?
> I tried to view the attached PDF, but could only see 9 blank pages. Can someone else check this? @kkarrancsu do you want to try sending me the file...
I've added a link to the draft implementation in the [builtin-actors](https://github.com/filecoin-project/builtin-actors/pull/1557) repository. Is it OK to move this FIP forward for consideration, while we continue to refine the draft implementation?