01
Gather.
179k parallel Mien sentences, cleaned, deduplicated, and aligned to the unified Latin orthography.
A free translator, a community dictionary, and a 20-level course for Iu Mien.
chrF scores how close a translation is to a human one: 0 unrelated, 100 identical. Even human translators cap near 80, since everyone phrases things differently.
Read the bars from short to long. Longer is better.
Our v7 model lands at 62: fluent enough to read for meaning, occasional small errors, clearly past every published baseline for Mien.
Trained on William & Mary's SciClone HPC cluster.
179k parallel Mien sentences, cleaned, deduplicated, and aligned to the unified Latin orthography.
Fine tuned Meta's NLLB 1.3B over seven iterations with LoRA on a single A30 GPU. Each pass corrected the weak spots from the last.
The v7 model scores chrF 62 ium → eng on a held-out test set.
Free forever. No ads. The goal is one centralized curriculum the whole community can lean on.