Views
No views yet
facebook/mbart-large-50-many-to-many-mmten_XX ↔ te_INpeft), transformers, datasetsenglish: Source texttelugu: Target translation1from transformers import MBartForConditionalGeneration, MBart50TokenizerFast
2from peft import PeftModel
3
4# Load base model & tokenizer
5base_model = MBartForConditionalGeneration.from_pretrained("facebook/mbart-large-50-many-to-many-mmt")
6tokenizer = MBart50TokenizerFast.from_pretrained("your-username/lora-mbart-en-te")
7
8# Load LoRA adapter
9model = PeftModel.from_pretrained(base_model, "your-username/lora-mbart-en-te")
10
11# Set source and target languages
12tokenizer.src_lang = "en_XX"
13tokenizer.tgt_lang = "te_IN"
14
15# Prepare input
16inputs = tokenizer("Hello, how are you?", return_tensors="pt")
17generated_ids = model.generate(**inputs, forced_bos_token_id=tokenizer.lang_code_to_id["te_IN"])
18translation = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
19print(translation)| Setting | Value |
|---|---|
| Base Model | mBART-50 |
| LoRA r | 8 |
| LoRA Alpha | 32 |
| Dropout | 0.1 |
| Optimizer | AdamW |
| Batch Size | 8 |
| Epochs | 3 |
| Mixed Precision | fp16 |
en_XX and te_IN (Telugu) at this stage1@inproceedings{liu2020mbart,
2 title={Multilingual Denoising Pre-training for Neural Machine Translation},
3 author={Liu, Yinhan and others},
4 booktitle={ACL},
5 year={2020}
6}