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 6% 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：

python

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-v2-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：

python

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-v2-distill')
model = AutoModel.from_pretrained('DMetaSoul/sbert-chinese-general-v2-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	38ms	20ms	-47%
Throughput	418 sentence/s	791 sentence/s	1.9x

精度：

	csts_dev	csts_test	afqmc	lcqmc	bqcorpus	pawsx	xiaobu	Avg
Teacher	77.19%	72.59%	36.79%	76.91%	49.62%	16.24%	63.15%	56.07%
Student	76.49%	73.33%	26.46%	64.26%	46.02%	11.83%	52.45%	50.12%
Gap (abs.)	-	-	-	-	-	-	-	-5.95%

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