distilbert-base-multilingual-cased-sentiment

模型描述

需补充更多信息

预期用途与局限性

需补充更多信息

训练与评估数据

需补充更多信息

训练过程

训练超参数

训练过程中使用了以下超参数：

learning_rate: 5e-05
train_batch_size: 16
eval_batch_size: 16
seed: 33
distributed_type: sagemaker_data_parallel
num_devices: 8
total_train_batch_size: 128
total_eval_batch_size: 128
optimizer: Adam，betas=(0.9,0.999)，epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 5
mixed_precision_training: Native AMP

训练结果

训练损失	轮次	步数	验证损失	准确率	F1
0.6405	0.53	5000	0.5826	0.7498	0.7498
0.5698	1.07	10000	0.5686	0.7612	0.7612
0.5286	1.6	15000	0.5593	0.7636	0.7636
0.5141	2.13	20000	0.5842	0.7648	0.7648
0.4763	2.67	25000	0.5736	0.7637	0.7637
0.4549	3.2	30000	0.6027	0.7593	0.7593
0.4231	3.73	35000	0.6017	0.7552	0.7552
0.3965	4.27	40000	0.6489	0.7551	0.7551
0.3744	4.8	45000	0.6426	0.7534	0.7534

框架版本

Transformers 4.12.3
Pytorch 1.9.1
Datasets 1.15.1
Tokenizers 0.10.3

使用方法

from openmind import pipeline, AutoTokenizer, is_torch_npu_available
import argparse

def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--model_name_or_path",
        type=str,
        help="Path to model",
        default="ChongqingAscend/distilbert-base-multilingual-cased-sentiment",
    )
    args = parser.parse_args()
    return args

def main():
    args = parse_args()
    model_path = args.model_name_or_path

    if is_torch_npu_available():
        device = "npu:0"
    else:
        device = "cpu"

    #tokenizer = AutoTokenizer.from_pretrained(model_path)
    pipe = pipeline("token-classification", model=model_path, device=device)
    out = pipe("I love you")

    print(out)
    
if __name__ == "__main__":
    main()

distilbert-base-multilingual-cased-sentiment

模型描述

需补充更多信息

预期用途与局限性

需补充更多信息

训练与评估数据

需补充更多信息

训练过程

训练超参数

训练过程中使用了以下超参数：

learning_rate: 5e-05
train_batch_size: 16
eval_batch_size: 16
seed: 33
distributed_type: sagemaker_data_parallel
num_devices: 8
total_train_batch_size: 128
total_eval_batch_size: 128
optimizer: Adam，betas=(0.9,0.999)，epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 5
mixed_precision_training: Native AMP

训练结果

训练损失	轮次	步数	验证损失	准确率	F1
0.6405	0.53	5000	0.5826	0.7498	0.7498
0.5698	1.07	10000	0.5686	0.7612	0.7612
0.5286	1.6	15000	0.5593	0.7636	0.7636
0.5141	2.13	20000	0.5842	0.7648	0.7648
0.4763	2.67	25000	0.5736	0.7637	0.7637
0.4549	3.2	30000	0.6027	0.7593	0.7593
0.4231	3.73	35000	0.6017	0.7552	0.7552
0.3965	4.27	40000	0.6489	0.7551	0.7551
0.3744	4.8	45000	0.6426	0.7534	0.7534

框架版本

Transformers 4.12.3
Pytorch 1.9.1
Datasets 1.15.1
Tokenizers 0.10.3

使用方法

from openmind import pipeline, AutoTokenizer, is_torch_npu_available
import argparse

def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--model_name_or_path",
        type=str,
        help="Path to model",
        default="ChongqingAscend/distilbert-base-multilingual-cased-sentiment",
    )
    args = parser.parse_args()
    return args

def main():
    args = parse_args()
    model_path = args.model_name_or_path

    if is_torch_npu_available():
        device = "npu:0"
    else:
        device = "cpu"

    #tokenizer = AutoTokenizer.from_pretrained(model_path)
    pipe = pipeline("token-classification", model=model_path, device=device)
    out = pipe("I love you")

    print(out)
    
if __name__ == "__main__":
    main()