core/nn/bert library

HuggingFace BERT-style encoder. Post-LN, learned absolute positions, WordPiece vocab, token-type embeddings folded into a single bias since sentence-transformer inference always uses type id 0. Matches the layout of bert-base-*, sentence-transformers/ all-MiniLM-L6-v2, etc.

Input is a 1D [seqLen] tensor of token indices (as floats, since this repo doesn't have an int tensor). Output is [seqLen, embedDim] — feed it into a SentenceEncoder (or a CrossEncoder) with mean pooling + L2 norm to reproduce the sentence-BERT recipe.

Classes

BertConfig

BertEmbeddings

BertLayer

BertModel