英文

德语语义STS_V2

这是一个 sentence-transformers 模型:它将句子和段落映射到1024维的稠密向量空间,可用于聚类或语义搜索等任务。

特别感谢 deepset 提供的gBERT-large模型,以及 Philip May 对数据集的翻译和关于该主题的讨论。

经过微调后,模型得分最好,与这些模型相比:

Model Name Spearman
xlm-r-distilroberta-base-paraphrase-v1 0.8079
12310321 0.7877
xlm-r-bert-base-nli-stsb-mean-tokens 0.7877
12311321 0.6371
12312321 0.8529
12313321 0.8355
12314321 0.8550
aari1995/German_Semantic_STS_V2 0.8626

使用(Sentence-Transformers)

安装 sentence-transformers 后,使用这个模型非常简单:

pip install -U sentence-transformers

然后你可以像这样使用模型:

from sentence_transformers import SentenceTransformer
sentences = ["This is an example sentence", "Each sentence is converted"]

model = SentenceTransformer('aari1995/German_Semantic_STS_V2')
embeddings = model.encode(sentences)
print(embeddings)

使用(HuggingFace Transformers)

如果没有 sentence-transformers ,可以像这样使用模型:首先,将输入通过变换器模型,然后必须在情境化的词嵌入之上应用正确的汇聚操作。

from transformers import AutoTokenizer, AutoModel
import torch


#Mean Pooling - Take attention mask into account for correct averaging
def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)


# Sentences we want sentence embeddings for
sentences = ['This is an example sentence', 'Each sentence is converted']

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('aari1995/German_Semantic_STS_V2')
model = AutoModel.from_pretrained('aari1995/German_Semantic_STS_V2')

# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')

# Compute token embeddings
with torch.no_grad():
    model_output = model(**encoded_input)

# Perform pooling. In this case, mean pooling.
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])

print("Sentence embeddings:")
print(sentence_embeddings)

评估结果

要对该模型进行自动评估,请参阅句子嵌入基准: https://seb.sbert.net

训练

模型使用以下参数进行训练:

DataLoader:

torch.utils.data.dataloader.DataLoader,长度为1438,带有以下参数:

{'batch_size': 4, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}

Loss:

sentence_transformers.losses.ContrastiveLoss.ContrastiveLoss,带有以下参数:

{'distance_metric': 'SiameseDistanceMetric.COSINE_DISTANCE', 'margin': 0.5, 'size_average': True}

fit()方法的参数:

{
    "epochs": 4,
    "evaluation_steps": 500,
    "evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
    "optimizer_params": {
        "lr": 5e-06
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 576,
    "weight_decay": 0.01
}

完整模型架构

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
)

引用和作者

基本模型由deepset进行训练。数据集由Philip May发布/翻译。模型由Aaron Chibb进行了微调。