This model is a fine-tuned version of facebook/nllb-200-3.3B on the EasyProject dataset.
Framework versions
Transformers 4.29.2
Pytorch 1.11.0+cu113
Datasets 2.8.0
Tokenizers 0.13.2
Paper link: Frustratingly Easy Label Projection for Cross-lingual Transfer
Github link: https://github.com/edchengg/easyproject
Please use the transformers==4.29.2 library as Huggingface recently fixed a bug in NLLB tokenizer
Code
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import torch
tokenizer = AutoTokenizer.from_pretrained(
"facebook/nllb-200-distilled-600M", src_lang="eng_Latn")
print("Loading model")
model = AutoModelForSeq2SeqLM.from_pretrained("ychenNLP/nllb-200-3.3b-easyproject")
model.cuda()
input_chunks = ["A translator always risks inadvertently introducing source-language words, grammar, or syntax into the target-language rendering."]
print("Start translation...")
output_result = []
batch_size = 1
for idx in tqdm(range(0, len(input_chunks), batch_size)):
start_idx = idx
end_idx = idx + batch_size
inputs = tokenizer(input_chunks[start_idx: end_idx], padding=True, truncation=True, max_length=128, return_tensors="pt").to('cuda')
with torch.no_grad():
translated_tokens = model.generate(**inputs, forced_bos_token_id=tokenizer.lang_code_to_id["zho_Hans"],
max_length=128, num_beams=5, num_return_sequences=1, early_stopping=True)
output = tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)
output_result.extend(output)
print(output_result)
Citation
@inproceedings{chen2023easyproject,
title={Frustratingly Easy Label Projection for Cross-lingual Transfer},
author={Chen, Yang and Jiang, Chao and Ritter, Alan and Xu, Wei},
booktitle={Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Findings)},
year={2023}
}
- Downloads last month
- 94