sentence-transformers feature-extraction sentence-similarity transformers semantic-search chinese

DMetaSoul/sbert-chinese-general-v1-distill

此模型是之前开源通用语义匹配模型的蒸馏版本(仅4层 BERT),适用于通用语义匹配场景(此模型在 Chinese-STS 任务上效果较好,但在其它任务上效果并非最优,存在一定过拟合风险),比如文本特征抽取、文本向量聚类、文本语义搜索等业务场景。

离线训练好的大模型如果直接用于线上推理,对计算资源有苛刻的需求,而且难以满足业务环境对延迟、吞吐量等性能指标的要求,这里我们使用蒸馏手段来把大模型轻量化。从 12 层 BERT 蒸馏为 4 层后,模型参数量缩小到 44%,大概 latency 减半、throughput 翻倍、精度下降 3% 左右(具体结果详见下文评估小节)。

Usage

1. Sentence-Transformers

通过 sentence-transformers 框架来使用该模型,首先进行安装:

pip install -U sentence-transformers

然后使用下面的代码来载入该模型并进行文本表征向量的提取:

from sentence_transformers import SentenceTransformer
sentences = ["我的儿子!他猛然间喊道,我的儿子在哪儿?", "我的儿子呢!他突然喊道,我的儿子在哪里?"]

model = SentenceTransformer('DMetaSoul/sbert-chinese-general-v1-distill')
embeddings = model.encode(sentences)
print(embeddings)

2. HuggingFace Transformers

如果不想使用 sentence-transformers 的话,也可以通过 HuggingFace Transformers 来载入该模型并进行文本向量抽取:

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 = ["我的儿子!他猛然间喊道,我的儿子在哪儿?", "我的儿子呢!他突然喊道,我的儿子在哪里?"]

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('DMetaSoul/sbert-chinese-general-v1-distill')
model = AutoModel.from_pretrained('DMetaSoul/sbert-chinese-general-v1-distill')

# 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)

Evaluation

这里主要跟蒸馏前对应的 teacher 模型作了对比

性能:

Teacher Student Gap
Model BERT-12-layers (102M) BERT-4-layers (45M) 0.44x
Cost 23s 12s -47%
Latency 37ms 20ms -46%
Throughput 422 sentence/s 788 sentence/s 1.8x

精度:

csts_dev csts_test afqmc lcqmc bqcorpus pawsx xiaobu Avg
Teacher 84.54% 82.17% 23.80% 65.94% 45.52% 11.52% 48.51% 51.71%
Student 83.39% 79.96% 20.25% 63.39% 43.70% 7.54% 46.91% 49.28%
Gap (abs.) - - - - - - - -2.43%

基于1万条数据测试,GPU设备是V100,batch_size=16,max_seq_len=256

Citing & Authors

E-mail: xiaowenbin@dmetasoul.com