Skip to main content

Traduction Francais <-> Ewe avec NLLB-200 fine-tune

Project description

ewe-nllb

Traduction Francais <-> Ewe avec NLLB-200 fine-tune.

Modele NLLB-200-distilled-600M de Meta, fine-tune avec LoRA sur ~40 000 paires de phrases paralleles francais-ewe.

Installation

pip install ewe-nllb

Pour la synthese vocale (TTS) :

pip install ewe-nllb[tts]

Utilisation

Python

from ewe_nllb import translate

# Francais -> Ewe
translate("Bonjour, comment allez-vous ?")
# => "Ŋdi, aleke wòle?"

# Ewe -> Francais
translate("Ŋdi", src="ee", tgt="fr")
# => "Bonjour"

Utilisation avancee

from ewe_nllb import EweTranslator

t = EweTranslator()

# Traduction simple
t.translate("Le Ghana est un beau pays.", src="fr", tgt="ee")

# Traduction par lot
t.translate_batch(["Bonjour", "Merci", "Au revoir"], src="fr", tgt="ee")

Synthese vocale (TTS)

from ewe_nllb import EweTTS

tts = EweTTS()
waveform = tts.synthesize("Ŋdi, aleke wòle?")
tts.save_audio(waveform, "output.wav")

Ligne de commande

# Francais -> Ewe
ewe-translate "Bonjour, comment allez-vous ?"

# Ewe -> Francais
ewe-translate "Ŋdi" --src ee --tgt fr

# Avec synthese vocale
ewe-translate "Bonjour" --tts --output audio.wav

Donnees d'entrainement

Le modele a ete fine-tune sur des donnees provenant de :

Source Paires Type
Masakhane MAFAND ~5 000 Corpus academique
eBible.org ~31 000 Versets bibliques
JW.org ~4 750 Bible + articles

Dataset complet : cnss-ewe-project/library

Modele

  • Base : facebook/nllb-200-distilled-600M (600M parametres)
  • Fine-tuning : LoRA (rank=16, alpha=32) sur les couches d'attention
  • Parametres entraines : 2.4M / 1404M (0.2%)
  • Eval loss : 2.02

Modele sur HuggingFace : cnss-ewe-project/nllb-ewe-finetuned

Langues supportees

Code Langue
fr Francais
ee Ewe (Eʋegbe)

L'ewe est parle par environ 7 millions de personnes au Ghana et au Togo.

Licence

MIT

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

ewe_nllb-0.1.0.tar.gz (6.1 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

ewe_nllb-0.1.0-py3-none-any.whl (7.3 kB view details)

Uploaded Python 3

File details

Details for the file ewe_nllb-0.1.0.tar.gz.

File metadata

  • Download URL: ewe_nllb-0.1.0.tar.gz
  • Upload date:
  • Size: 6.1 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.2

File hashes

Hashes for ewe_nllb-0.1.0.tar.gz
Algorithm Hash digest
SHA256 3f0eb113461aa9de8b3413f4e23e5eb401a72010d952025eaeb1b995815777d2
MD5 ba4acc82abcadb2d9290b4fcfe8c36ac
BLAKE2b-256 8efc13d072e23cd921be7e6e4f094fbc42337f9177fdc8c99fd028b1da052959

See more details on using hashes here.

File details

Details for the file ewe_nllb-0.1.0-py3-none-any.whl.

File metadata

  • Download URL: ewe_nllb-0.1.0-py3-none-any.whl
  • Upload date:
  • Size: 7.3 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.2

File hashes

Hashes for ewe_nllb-0.1.0-py3-none-any.whl
Algorithm Hash digest
SHA256 9575d3fe8184a9ca7b035a76bab949bf9d6282586d6f96a181f8ff57500ce5f9
MD5 fdbba4afd2d0ff4d2fa6b2fc5e7fc1cd
BLAKE2b-256 dd192642a57dc971ae2e435393a1272cc4075104bad380ae0b56de50aea237a1

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page