Sidd Karamcheti
Sidd Karamcheti
Hey @TariqAHassan - sorry I'm late to the party! This is awesome work, and minimally, I'd love to help with figuring out how to reproduce the CIFAR numbers in your...
Hey @TariqAHassan - sounds good and take your time! Maybe ping me directly in the PR (comments) if there are things that are particularly strange, otherwise, I'll go through it...
What version of python are you using?
Right -> This repo was written for Python 2.7. I'm not sure that this code is compatible with 3.6.
Yeah - I agree. I'm considering adding dropout, and seeing how that affects results.
Consider a seq2seq program generation task (e.g. image2latex) -- I'd like to evaluate on exact match of final generate program, not on perplexity or token-sensitive metrics.
Yeah - theoretically our models are general enough to support a ton of the new models others are releasing out of the box (including the one linked above, the original...
Thanks so much @amyeroberts — just to clarify, I can register these models with the same `XXXForConditionalGeneration` API used by models like LLaVA right? Ideally want things like HF Trainer,...
Any update on this? Running into the same issue when I have unused parameters for a given forward pass!
Bumping this - is there a reason to pin `timm` to 0.4.12?