Request access

Describe your intended use below. Downloads unlock after approval.

Weights are released under CC BY-NC-SA 4.0 for non-commercial research and education. Credit NileAGI, keep adaptations under the same licence, and contact us before any commercial product or paid service. Do not use the model to harm people, break the law, or present outputs as human writing.

Log in or Sign Up to review the conditions and access this model content.

Runyambo MT

Authors: Zephania and Isack Odero

Swahili → Runyambo (Orunyambo) lexical / phrase machine translation. Fine-tuned from NLLB-200 distilled 600M.

This is not sentence bitext MT. Prefer short lemmas and dictionary-style phrases.

Quick example

Always set the source language to swh_Latn and force the Runyambo BOS token nym_Latn. If you skip that, output will not match this card.

Summary

Task Lexical / phrase MT
Direction Swahili → Runyambo
Language codes swh_Latn → nym_Latn
Base facebook/nllb-200-distilled-600M
Primary metric chrF2

Evaluation

Held-out test. Identity baseline (copy Swahili) ≈ chrF2 8.0.

System Test chrF2 BLEU exact %
This SKU (600M) 16.6 0.0 0.0
Lite (1.3B) 16.6 3.4 0.2
Tiny Aya LoRA 10.0 0.0 0.0

A mid-teens chrF2 means stems are often partially recovered; surfaces are not reliable dictionary lookups. Research preview only.

How to use

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

tok = AutoTokenizer.from_pretrained("nileagi/nileagi-nym-mt", use_fast=False)
model = AutoModelForSeq2SeqLM.from_pretrained("nileagi/nileagi-nym-mt")

tok.src_lang = "swh_Latn"
inputs = tok("habari", return_tensors="pt")
bos = tok.convert_tokens_to_ids("nym_Latn")
out = model.generate(**inputs, forced_bos_token_id=bos, max_new_tokens=64)
print(tok.decode(out[0], skip_special_tokens=True))

HTTP API

On the NileAGI host: port 8096 (nileagi-nym-mt-api).

Intended use

  • Lexicon lookup / short phrase Swahili → Runyambo
  • Research on low-resource Bantu MT

Out of scope

  • Full-sentence or document translation
  • Runyambo → Swahili (use the reverse SKU)
  • Direct English → Runyambo
  • Chat, legal, medical, news
  • Commercial products without a NileAGI licence

Limitations

  • Trained on dictionary-style phrases; sentence quality is weak
  • Decode can insert spaces inside Runyambo words; prefer chrF2 over BLEU
  • Research preview scores — not a substitute for a human lexicographer

Related

Access

Weights are gated under CC BY-NC-SA 4.0.

  1. Open Request access and describe your intended use.
  2. After approval, authenticate (hf auth login or HF_TOKEN) and load the model.

Attribute NileAGI. Adapted weights must stay CC BY-NC-SA 4.0. Commercial: hi@nileagi.com.

License

Weights: CC BY-NC-SA 4.0.
See LICENSE and NOTICE.md.

Citation

@misc{nileagi-nym-mt-2026,
  title        = {Runyambo machine translation (Swahili to Runyambo)},
  author       = {Zephania and Isack Odero},
  year         = {2026},
  howpublished = {Hugging Face},
  url          = {https://huggingface.co/nileagi/nileagi-nym-mt},
  note         = {NileAGI}
}
Downloads last month
-
Safetensors
Model size
0.6B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Collection including nileagi/nileagi-nym-mt