A.X Encoder

ModernBERT-based Korean encoder model (100M) with Flash Attention and 16K context. Used for data curation during A.X LLM training. Apache-2.0.