v0.3.0 — Ukrainian text optimizer for LLMs

dormouse v0.3.0

Ukrainian text optimizer for LLMs — fewer tokens, better comprehension.

Key metrics

• 73% token savings (cloud squeeze with seq2seq)
• 88% lexicon coverage on 53K text corpus
• 98.2% exact match seq2seq expression translation
• 150% quality preservation — GPT understands squeezed better than original UA

Assets

Download these files automatically on first use, or manually place in ~/.cache/dormouse/v0.3.0/:

• lexicon.db — 47K entry UA→EN lexicon
• expr_seq2seq.pt — GRU encoder-decoder for expression translation
• expr_vocab_src.json / expr_vocab_tgt.json — source/target vocabularies
• expr_config.json — model configuration

Install

pip install dormouse