Part of the Windstorm Labs open model catalogue. Scores, licence and attribution for every model: https://windytranslate.com/models/translate-fr-tum
Canonical copy: https://huggingface.co/WindyTranslate/translate-fr-tum
Translates French → Tumbuka.
No quality score is published in this repository. Current screening scores, where measured, are on the catalogue page linked above.
Deployment formats in this repository (subfolders):
| Variant | Description |
|---|---|
lora/ |
WindyStandard — production baseline. Transformers format for GPU inference. |
lora-ct2-int8/ |
WindyStandard · CPU INT8 — CTranslate2 INT8 quantization of WindyStandard for CPU inference. |
Transformers (PyTorch):
from transformers import MarianMTModel, MarianTokenizer
tokenizer = MarianTokenizer.from_pretrained("WindstormLabs/translate-fr-tum", subfolder="lora")
model = MarianMTModel.from_pretrained("WindstormLabs/translate-fr-tum", subfolder="lora")
CTranslate2 (fast CPU inference):
import ctranslate2
translator = ctranslate2.Translator("path/to/translate-fr-tum/lora-ct2-int8")
The Windy Word apps are built on this model family.
Weights derived from Helsinki-NLP/opus-mt-fr-tum (OPUS-MT, Helsinki-NLP, University of Helsinki), licensed Apache-2.0. Windy variants are released under the same licence.
lora/model.safetensors 303 MBlora-ct2-int8/model.bin 78 MB