We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
unnecessary, pytorch native softmax is numerically stable
zero init all branch output linear layers
properly mask for structure module
make sure outermean can be masked for padding in msa rows, and also f… …ix a bug with mask
fix quaternion update
use stable softmax in attention
release checkpointing for evoformer
bug