Instructions to use JosueG/adja-nmt-nllb-600m-reverse-r10ks4k-seed42 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use JosueG/adja-nmt-nllb-600m-reverse-r10ks4k-seed42 with Transformers:
# Use a pipeline as a high-level helper # Warning: Pipeline type "translation" is no longer supported in transformers v5. # You must load the model directly (see below) or downgrade to v4.x with: # 'pip install "transformers<5.0.0' from transformers import pipeline pipe = pipeline("translation", model="JosueG/adja-nmt-nllb-600m-reverse-r10ks4k-seed42")# Load model directly from transformers import AutoTokenizer, AutoModelForSeq2SeqLM tokenizer = AutoTokenizer.from_pretrained("JosueG/adja-nmt-nllb-600m-reverse-r10ks4k-seed42") model = AutoModelForSeq2SeqLM.from_pretrained("JosueG/adja-nmt-nllb-600m-reverse-r10ks4k-seed42", device_map="auto") - Notebooks
- Google Colab
- Kaggle
NLLB-600M Adja -> French
This is the thesis-final Adja-to-French machine translation checkpoint for the May 2026 CS thesis release. It is the reverse-direction companion to the French-to-Adja NLLB model.
Thesis Role
This model is the reverse MT artifact: Adja source text to French target text. It is used for Adja text interpretation, reverse translation checks, and the speech-pipeline experiments where ASR output is translated into French.
Model And Data
- Task: machine translation, Adja -> French
- Base model:
facebook/nllb-200-distilled-600M - Direction:
aj_Latn->fra_Latn - Training data: French-Adja thesis corpus lineage, with the public
canonical MT reference at
JosueG/french-adja-parallel-corpus - Release repo:
FrejusGdm/cs-thesis-may-2026
Headline Result
This seed-42 NLLB-600M checkpoint is the public reverse-direction MT weight location for the thesis. It is the intended model to pair with Adja ASR when building Adja speech -> French text pipelines.
See:
results/mt/summary/all_results.csvresults/adja-nmt/pipeline-comparison.md
Limitations
- The model should be treated as a research artifact, not a production translator.
- ASR noise compounds MT errors in speech pipelines.
- The model uses the release tokenizer/config; load this repo directly so the Adja language token and generation settings match the checkpoint.
Citation
If you use this model, cite:
Josue Godeme. 2026. CS Thesis May 2026: French-Adja MT and Adja Speech Experiments. https://github.com/FrejusGdm/cs-thesis-may-2026
- Downloads last month
- 8
Model tree for JosueG/adja-nmt-nllb-600m-reverse-r10ks4k-seed42
Base model
facebook/nllb-200-distilled-600M