Views
No views yet
xgboost library. It uses the embeddings generated by sentence-transformers/LaBSE as a feature vector.1from transformers import AutoTokenizer, AutoModel
2from xgboost import XGBClassifier
3import torch.nn.functional as F
4import torch
5
6def mean_pooling(model_output, attention_mask):
7 token_embeddings = model_output[0]
8 input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
9 return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
10
11tokenizer = AutoTokenizer.from_pretrained("sentence-transformers/LaBSE")
12embedding_model = AutoModel.from_pretrained("sentence-transformers/LaBSE")
13device = ("cuda" if torch.cuda.is_available() else "cpu")
14embedding_model.to(device)
15
16bst = XGBClassifier({'device': device})
17bst.load_model('/path/to/XGBClassifier-text-classifier.json')
18
19def score_text(text, model):
20
21 encoded_input = tokenizer(text, padding=True, truncation=True, return_tensors='pt').to(device)
22
23 with torch.no_grad():
24 model_output = embedding_model(**encoded_input)
25
26 sentence_embedding = mean_pooling(model_output, encoded_input['attention_mask'])
27
28 embedding = F.normalize(sentence_embedding, p=2, dim=1).numpy()
29 score = model.predict(embedding)[0]
30
31 return score
32
33score_text("Os tucanos são aves que correspondem à família Ramphastidae, vivem nas florestas tropicais da América Central e América do Sul. A família inclui cinco gêneros e mais de quarenta espécies diferentes. Possuem bicos notavelmente grandes e coloridos, que possuem a função de termorregulação para as muitas espécies que passam muito tempo na copa da floresta exposta ao sol tropical quente.", bst)1@misc{correa2024tucanoadvancingneuraltext,
2 title={{Tucano: Advancing Neural Text Generation for Portuguese}},
3 author={Corr{\^e}a, Nicholas Kluge and Sen, Aniket and Falk, Sophia and Fatimah, Shiza},
4 year={2024},
5 eprint={2411.07854},
6 archivePrefix={arXiv},
7 primaryClass={cs.CL},
8 url={https://arxiv.org/abs/2411.07854},
9}
10
11@article{correa2025tucanoadvancingneuraltext,
12 title={{Tucano: Advancing Neural Text Generation for Portuguese}},
13 author={Corr{\^e}a, Nicholas Kluge and Sen, Aniket and Falk, Sophia and Fatimah, Shiza},
14 journal={Patterns},
15 publisher={Elsevier},
16 year={2025},
17 doi={10.1016/j.patter.2025.101325},
18 url={https://doi.org/10.1016/j.patter.2025.101325},
19 issn={2666-3899}
20}