Independent developer
Jibby1729
Independent
Builds complete Kashmiri systems: KashViT for word-level OCR, and a two-stage English→Kashmiri pipeline of a Sarvam-Translate fine-tune and a character-level diacritic restorer that reports the highest KATHE score among the systems its author evaluated.
- 3
- Models
Focus
- Diacritization
- Machine Translation
- OCR
Models 3
Koshur-JibDiacritizer-152M — Kashmiri Diacritic Restorer
Character-level Transformer trained from scratch on 244K Kashmiri lines (BPCC and IndicVoices transcripts) that adds vowel marks to unmarked Perso-Arabic Kashmiri without changing letters. Built as the second stage after Koshur-Sarvam-4B; the author reports it beating the HNM ByT5 diacritizers on the same drafts.
Koshur-Sarvam-4B — English→Kashmiri Translator
Sarvam-Translate 4B fully fine-tuned on 108K human English–Kashmiri pairs (BPCC seed, NLLB-Seed, BPCC-Daily), paired with a separate diacritizer. Reports the highest KATHE score among the systems its author evaluated (mean 14.91 vs 12.90 for Sarvam-Translate), scored by the author on IN22, SMOL and FLORES. English→Kashmiri only; also licensed under the Gemma Terms of Use; 4-bit version runs in about 3 GB.
KashViT — Kashmiri Word OCR
Two Vision Transformer models for recognising single Kashmiri words, with training, evaluation and inference scripts and a small web app to compare them.
Every work above is a record in the Kashmiri NLP resource hub, credited as published; profile details come from huggingface.co, github.com. Something missing or wrong? Submit it.