Modelpublic

mehii123/kathe-2026-nllb-kashmiri

sourceHugging Facecc-by-nc-4.0updated 1mo agoView on Hugging Face

0likes23downloads

Model Card

KATHE 2026: Fine-Tuned NLLB for English to Kashmiri Translation

This model is a fine-tuned version of facebook/nllb-200-distilled-600M trained on the BPCC dataset for English (eng_Latn) to Kashmiri Perso-Arabic (kas_Arab) machine translation for the KATHE 2026 competition.

Model Details

—Base Model: facebook/nllb-200-distilled-600M
—Task: Translation (eng_Latn -> kas_Arab)
—Evaluation Score: 11.03 BLEU

Inference Code Example

python

import torch
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
MODEL_ID = "mehii123/kathe-2026-nllb-kashmiri"

tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_ID).to(DEVICE)

text = "She was a true visionary."
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True).to(DEVICE)

target_lang_id = tokenizer.lang_code_to_id["kas_Arab"]

outputs = model.generate(
    **inputs, 
    forced_bos_token_id=target_lang_id, 
    max_new_tokens=100,
    num_beams=2,
    length_penalty=1.0,
    early_stopping=True
)

translation = tokenizer.decode(outputs[0], skip_special_tokens=True)
print("Kashmiri Translation:", translation)