opus-mt-bg-en-tv

Bulgarian → English translation for reality-TV transcripts, made in a BUas ADSAI student project (Year 2 Block A, group 1) for the Content Intelligence Agency.

Helsinki-NLP/opus-mt-bg-en fine-tuned with task7/machine_translation.py (seed 42, 1 epoch, learning rate 2e-5) on 100,000 OPUS-100 pairs and 8,679 lines from Bulgarian TV shows.

Evaluation and findings: task7/findings_translation.md

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

repo = "Shadowiee/opus-mt-bg-en-tv"
tokenizer = AutoTokenizer.from_pretrained(repo)
model = AutoModelForSeq2SeqLM.from_pretrained(repo)

batch = tokenizer(["Добро утро на всички!"], return_tensors="pt", padding=True, truncation=True, max_length=128)
print(tokenizer.batch_decode(model.generate(**batch, num_beams=4, max_new_tokens=128), skip_special_tokens=True))