voxreality/src_ctx_aware_nllb_1.3B

The model and the tokenizer are based on facebook/nllb-200-1.3B.

We trained the model to use one sentence of context. The context is prepended to the input sentence with the sep_token in between. We used a subset of the OpenSubtitles2018 dataset for training. We trained on the interleaved dataset for all directions between the following languages: English, German, Dutch, Spanish, Italian, and Greek.

The tokenizer of the base model was not changed. For the language codes, see the base model.

Use this code for translation:

model_name = 'voxreality/src_ctx_aware_nllb_1.3B'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

max_length = 100
src_lang = 'eng_Latn'
tgt_lang = 'deu_Latn'
context_text = 'This is an optional context sentence.'
sentence_text = 'Text to be translated.'

# if the context is provided use the following:
input_text = f'{context_text} {tokenizer.sep_token} {sentence_text}'
# if no context is provided use the following:
# input_text = sentence_text

tokenizer.src_lang = src_lang
inputs = tokenizer(input_text, return_tensors='pt').to(model.device)
model_output = model.generate(**inputs,
                              forced_bos_token_id=tokenizer.lang_code_to_id[tgt_lang],
                              max_length=max_length)
output_text = tokenizer.batch_decode(model_output, skip_special_tokens=True)[0]

print(output_text)

You can also use the pipeline

from transformers import pipeline

model_name = 'voxreality/src_ctx_aware_nllb_1.3B'
translation_pipeline = pipeline("translation", model=model_name)
src_lang = 'eng_Latn'
tgt_lang = 'deu_Latn'
context_text = 'This is an optional context sentence.'
sentence_text = 'Text to be translated.'

# if the context is provided use the following:
input_texts = [f'{context_text} {tokenizer.sep_token} {sentence_text}']
# if no context is provided use the following:
# input_texts = [sentence_text]


pipeline_output = translation_pipeline(input_texts, src_lang=src_lang, tgt_lang=tgt_lang)

print(pipeline_output[0]['translation_text'])