genomic-llm-embeddings
Tokenization Strategy Selection
| Strategy | Context | Strength | Limitation | |---|---:|---|---| | Character (A/C/G/T/N) | Long | Max resolution | Long token sequences | | k-mer (k=3..6) | Short/medium | Fast baseline | Loses positional nuance | | DNABERT-2 (BPE) | ~512 bp windows | Strong short-window tasks | Limited context | | Nucleotide Transformer (6-mer) | kb-scale | Good transfer embeddings | Higher memory | | HyenaDNA | up to 1M bp | Long-range signals | He
[Description truncada. Veja o README completo no GitHub.]