MT trained on crawled data still sucks

from transformers import M2M100ForConditionalGeneration, M2M100Tokenizer

model = M2M100ForConditionalGeneration.from_pretrained("facebook/m2m100_418M")
tokenizer = M2M100Tokenizer.from_pretrained("facebook/m2m100_418M")

def xlate(text, src, trg):
    tokenizer.src_lang = src
    encoded = tokenizer(text, return_tensors="pt")
    generated_tokens = model.generate(**encoded, forced_bos_token_id=tokenizer.get_lang_id(trg))
    out = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)
    return out[0]

xlate("Litwo! Ojczyzno moja! ty jesteś jak zdrowie; ile cię trzeba cenić, ten tylko się dowie, Kto cię stracił.", "pl", "ga")

'Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers, Brazzers'

xlate("Hello, how are you?", "en", "ga")

'Brazzers físeán catagóir Inexperienced, Déagóir Inexperienced, Déagóir Inexperienced'