dhivehi-en-mt5-large-paragraph

mT5-large Dhivehi→English (paragraph-level)

mT5-large (1.2B, encoder-decoder) for Dhivehi→English translation, trained on the multi-granularity Dhivehi–English parallel corpus (gated).

mT5-based paragraph model — lighter and faster (1.2B). For higher translation quality, see the Qwen paragraph variant. Handles both single sentences and multi-sentence paragraphs.

Scores (chrF / chrF++ / BLEU)

chrF is the metric to trust for Thaana; BLEU is unreliable (word segmentation / morphology).

Example

Input (dv): އެއީ، މިދިޔަ އަހަރުގެ މި މުއްދަތާ ބަލާއިރު، 7.6 އިންސައްތައިގެ ކުރިއެރުމެއް ކަމަށްވާއިރު، ދުވާލަކަށް 7،778 ފަތުރުވެރިން ރާއްޖެ ޒިޔާރަތްކުރެއެވެ.

Output (en): While this is a 7.6 percent growth compared to the same period last year, 7,778 tourists visit the Maldives daily.

Multi-sentence input (dv): ނާޝިދާ މުޙައްމަދަކީ ދިވެހި ފިލްމީ ތަރިއެކެވެ. އޭނާވަނީ ދިވެހި ތަފާތު އެތައް ފިލްމްތަކެއްގައި ހަރަކާތްތެރި ވެފައެވެ. މީގެ އިތުރުން ތަފާތު ވީޑިއޯ ލަވަތަކާއި، ޓީވީ ޑްރާމާ ސިލްސިލާ ތަކުން ވެސް ނާއްކޮގެ ހުނަރު ބެލުންތެރިންނަށް ފެނިގެން ގޮސްފައި ވެއެވެ. އަމިއްލަ ދިރިއުޅުން ނާޝިދާ މުޙައްމަދު އަކީ ހދ. ކުޅުދުއްފުއްޓަށް އުފަން ބަތަލާއެކެވެ. މިހާރު ޒަވާޖީ ޙަޔާތެއް ވަނީ ފަށާފައެވެ. ފިލްމީ ދާއިރާއަށް ނިކުތުން ނާޝިދާ މުޙައްމަދު ފިލްމީ ދާއިރާ އަށް ނިކުތީ ފިލްމީ ތަރި އަޙުމަދު އާޞިމް އާއެކު ކުޅުނު ވީޑިއޯ ލަވަޔަކުންނެނެވެ. އެއަށްފަހު ތަފާތު ޑްރާމާ ، ފިލްމް، ސިލްސިލާ ތަކުގައި ހަރަކާތްތެރި ވެފައި ވެއެވެ.

Output (en): Nasheeda Mohamed is a Maldivian film star. She has performed in many different Maldivian films. In addition, Naakko's talent has been seen by audiences in various video songs and TV drama series. Personal Life: Nasheeda Mohamed is a lady born in HDh. Kulhudhuffushi. She has now begun a youth life. Entering the film industry: Nasheeda Mohamed entered the film industry through a video song she sang with film star Ahmed Asim. After that, she has performed in various dramas, films, and series.

Real held-out sample and this model's own output.

Usage

import torch
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

m = "Neobe/dhivehi-en-mt5-large-paragraph"
tok = AutoTokenizer.from_pretrained(m)
model = AutoModelForSeq2SeqLM.from_pretrained(m, torch_dtype=torch.float32).eval().cuda()  # fp32

src = "ދިވެހިރާއްޖޭގެ ރައީސް މިއަދު ކެބިނެޓާ ބައްދަލުކުރެއްވި އެވެ."
inp = tok(src, return_tensors="pt", truncation=True, max_length=1024).to("cuda")
out = model.generate(**inp, max_new_tokens=256, num_beams=4)
print(tok.decode(out[0], skip_special_tokens=True))

Run in fp32 — the T5 family produces garbage in bf16/fp16. ~5–6 GB VRAM.

Training

Base google/mt5-large; fp32; Adafactor; LR 1e-4 cosine; max_length 512; 1 epoch; effective batch ~32; gradient checkpointing.

Limitations

Domain = Maldivian news / press / Wikipedia; technical or informal English is out of distribution. Non-human references are machine-generated (distillation).

Citation

@misc{neobe_dhivehi_en_mt5_large_paragraph_2026,
  title  = {mT5-large Dhivehi→English (paragraph-level)},
  author = {Neobe},
  year   = {2026},
  howpublished = {\url{https://huggingface.co/Neobe/dhivehi-en-mt5-large-paragraph}}
}