Выпуск
Training a $10 Model for Hebrew Speaker Gender
My Weird Prompts · 18 ноября 2026 г. · 29 мин
Hebrew bakes a speaker's gender into first-person verbs and adjectives, sometimes changing the letters and sometimes only the unwritten vowels. This episode examines a $10 work-in-progress pipeline built to solve that gap for text-to-speech. Using 17,725 sentences generated by Claude Sonnet and filtered by Dicta's morphological analyser, a LoRA fine-tune on Gemma 4 E4B was trained in 32 minutes on Modal. The version history reveals how a data-filter bug caused early failures on female speakers, and why version three still trails prompting on that half of the job. Key sources: • Hugging Face model card, v3, created/modified 2026-11-18, retrieved 2026-11-18. — https://huggingface.co/Hebrish/Hebrew-Gendered-Sentence-Creator • GitHub README, status dated 2026-11-18, retrieved 2026-11-18. — https://github.com/Hebrish-ASR/Hebrew-Gendered-Sentence-Creator • Hugging Face model repo, 305M params, CC-BY-4.0, created 2026-11-18; card currently the inherited DictaBERT-large-char-menaked card (SOTA claim dated 2025-03), retrieved 2026-11-18. — https://huggingface.co/Hebrish/Hebrew-Gendered-Nikud Episode 6664 — myweirdprompts.com/n/6664 (permanent short link: myweirdprompts.com/194371) Listen on Android: MWP FM Player — myweirdprompts.com/apk