Kyle Daruwalla
Kyle Daruwalla
Lux's docs are a great example of well-written and organized docs. But the build system has many moving pieces, and I'm not sure it really provides much in the way...
We discussed this off GH and decided to take the hit and switch to Documenter for now. I think this PR is nearly ready, there is just @theabhirath's comments that...
I can manually build the previous tagged doc versions and push them. ~~@Saransh-cpp you had posted a snippet for doing this on some repo, but I can't find it now.~~...
Thank you for this contribution! Unfortunately, we don't accept PRs for isolated utilities like this. If this function is useful for constructing a larger model, then we would need to...
Also, we'll help guide your PR so that it is ready to merge. You don't need to worry about it being partially incomplete.
Another case like recurrence where the cuDNN API doesn't match the cute semantics that we use to define the layer. One thing that the Metalhead PR made me consider was...
A toggle would be a good idea so that the paper variant is still easily available.
Looks great! Seems like this is related to #117, i.e. having a repo separate for "just the models" and then a "vision" repo. The latter can take on heavier deps....
Yeah a docstring seems more useful than number of parameters. Maybe a description of the configurations too?
If Flux is willing to take on NeuralAttentionlib as a dep, and we would need to rework it to accept more inputs. Currently, it only accepts 3D.