BERTose sequence encoder: five weighting variants

The initial submission remains preserved in the repository's older files.

WURCS 2.0 glycan embeddings. Variant 3 is the reference encoder used in the reported benchmark; the five exports represent objective-weight configurations from pretraining seed 42.

Start with the shared workbench notebook or the download, install and run instructions. All inference and training sources are maintained once in the shared code repository. Scientific weights retain their original versioned paths under releases/2026-09-28/.

Revised weights use Apache-2.0; see LICENSE and NOTICE. Original shared code uses Apache-2.0 with retained third-party exceptions. Author-generated data use CC-BY-4.0; inherited inputs and external features retain their source-specific terms and attribution. Upstream ESM-C weights are not included.

Downloads last month
10
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support