Use BLEURT models in native PyTorch with Transformers.
Install with:
pip install git+https://github.com/lucadiliello/bleurt-pytorch.gitNow load your favourite model with:
import torch
from bleurt_pytorch import BleurtConfig, BleurtForSequenceClassification, BleurtTokenizer
config = BleurtConfig.from_pretrained('lucadiliello/BLEURT-20-D12')
model = BleurtForSequenceClassification.from_pretrained('lucadiliello/BLEURT-20-D12')
tokenizer = BleurtTokenizer.from_pretrained('lucadiliello/BLEURT-20-D12')
references = ["a bird chirps by the window", "this is a random sentence"]
candidates = ["a bird chirps by the window", "this looks like a random sentence"]
model.eval()
with torch.no_grad():
inputs = tokenizer(references, candidates, padding='longest', return_tensors='pt')
res = model(**inputs).logits.flatten().tolist()
print(res)
# [0.9604414105415344, 0.8080050349235535]You can find all BLUERT models adapted for PyTorch here. The recommended model is lucadiliello/BLEURT-20, however this model is very large and may require too much resources. BLEURT-20-D12 is smaller but works well enough for most comparisons.
- Google original BLEURT implementation
- Transformers project
- Users of this issue, from which I took inspiration.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent