这个模型是在 run_t5_mlm_flax.py 的略微改进代码上进行训练的。如果想了解训练细节或评估结果,请参考 SlovakT5_report.pdf 。对于评估,您也可以运行 SlovakT5_eval.ipynb 。
SlovakT5-small 可以用于许多不同的下游任务的微调。例如,NER(命名实体识别):
from transformers import AutoTokenizer, T5ForConditionalGeneration tokenizer = AutoTokenizer.from_pretrained("ApoTro/slovak-t5-small") model = T5ForConditionalGeneration.from_pretrained("ApoTro/slovak-t5-small") input_ids = tokenizer("ner veta: Do druhého kola postúpili Robert Fico a Andrej Kiska s rozdielom 4,0%.", return_tensors="pt").input_ids labels = tokenizer("per: Robert Fico | per: Andrej Kiska", return_tensors="pt").input_ids # the forward function automatically creates the correct decoder_input_ids loss = model(input_ids=input_ids, labels=labels).loss loss.item()