bbpe16k
Datasets
All datasets matching “bbpe16k”babylm-bbpe16k-512-structured-priors
BabyLM BBPE16K 512 Structured Priors
Structured syntax/lexical prior arrays for the CausalFocus direct-sum ablations.
encoded corpus: miguelcsx/babylm-bbpe16k-512-encoded
tokenizer: miguelcsx/causal-focus-bbpe16k-tokenizer
syntax neighbors: directional immediate-context sketch
lexical neighbors: symmetric context-window sketch
k: 8
clusters: 512
babylm-bbpe16k-512-encoded
BabyLM BBPE16K 512 Encoded Corpus
Encoded NumPy arrays for CausalFocus-GPTBERT-D256-BBPE16K.
source corpus: miguelcsx/babylm-corpus-cleaned
tokenizer: miguelcsx/causal-focus-bbpe16k-tokenizer
sequence length target: 512
splits: train, validation
