uer

/

sbert-base-chinese-nli

like 101

I try to reproduce the model result as the way u description.
The base model and the code is same with u .
the comand is:
python3 finetune/run_classifier_siamese.py --pretrained_model_path chinese_roberta/pytorch_model.bin --vocab_path models/google_zh_vocab.txt --config_path models/sbert/base_config.json --train_path datasets/ChineseTextualInference/train.tsv --dev_path datasets/ChineseTextualInference/dev.tsv --learning_rate 5e-5 --epochs_num 5 --batch_size 64
But the result model i got have a high similarity score(upper 90%) for all datasets(not matter labels 0,1 or 2) .