mclanorjeff/twi-english-nllb-lora

Twi to English NLLB LoRA Adapter

This repository contains a LoRA adapter fine-tuned for Twi to English translation. It was trained from facebook/nllb-200-distilled-600M using the local Ghana NLP Twi-English parallel text dataset included in the Sukuupath project.

Project Collaboration

This model was built and first published by McLanor Jeff under the personal repository mclanorjeff/twi-english-nllb-lora.

The shared collaboration version for the project team is available under the Lanor-and-Nick organization:

• —Lanor-and-Nick/twi-english-nllb-lora

That organization repository is used for collaborative access and team-facing updates while this personal repository remains the original author copy.

Model Details

• —Task: Twi to English translation
• —Base model: facebook/nllb-200-distilled-600M
• —Adapter type: LoRA / PEFT
• —Source language tag: twi_Latn
• —Target language tag: eng_Latn
• —Training hardware: NVIDIA RTX 2060, 6 GB VRAM
• —Trainable parameters: 2,359,296 of 617,433,088 total parameters

Training Data

The adapter was fine-tuned on:

• —TWI_ENGLISH_PARALLEL_TEXT
• —3,888 training examples
• —431 validation examples

The original CSV used text as the Twi source column and label as the English target/reference column.

Evaluation

Best validation checkpoint: checkpoint-900

Translation does not use ordinary classification accuracy because many different English translations can be valid for the same Twi sentence.

Usage

from peft import PeftConfig, PeftModel
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import torch

model_id = "mclanorjeff/twi-english-nllb-lora"
source_lang = "twi_Latn"
target_lang = "eng_Latn"

def normalize_twi_keyboard_text(text: str) -> str:
    return text.replace("C", "Ɔ").replace("c", "ɔ").replace("3", "ɛ")

config = PeftConfig.from_pretrained(model_id)
tokenizer = AutoTokenizer.from_pretrained(model_id, src_lang=source_lang, tgt_lang=target_lang)
base_model = AutoModelForSeq2SeqLM.from_pretrained(config.base_model_name_or_path)
model = PeftModel.from_pretrained(base_model, model_id)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
model.eval()

text = normalize_twi_keyboard_text("3he na wowc? M'ani gye w'as3m ho")
inputs = tokenizer(text, return_tensors="pt").to(device)
forced_bos_token_id = tokenizer.convert_tokens_to_ids(target_lang)

with torch.no_grad():
    output = model.generate(
        **inputs,
        forced_bos_token_id=forced_bos_token_id,
        max_new_tokens=192,
        num_beams=5,
        no_repeat_ngram_size=3,
    )

print(tokenizer.batch_decode(output, skip_special_tokens=True)[0])

Sample Outputs

Limitations

This is a useful fine-tuned model, not a perfect translator. It may struggle with slang, idioms, informal Twi, spelling variation, names, or sentences far outside the training distribution. Human review is recommended for high-stakes use.