Bert Base 1024 Biencoder 64M Pairs

🚀 bert-base-1024-biencoder-64M-pairs

这是一个基于 MosaicML在1024序列长度上预训练的BERT 的长上下文双编码器模型。该模型可将句子和段落映射到768维的密集向量空间，适用于聚类或语义搜索等任务。

🚀 快速开始

📦 安装指南

下载模型和相关脚本

git clone https://huggingface.co/shreyansh26/bert-base-1024-biencoder-64M-pairs

💻 使用示例

基础用法

import torch
from torch import nn
from transformers import AutoModelForMaskedLM, AutoTokenizer, pipeline, AutoModel
from mosaic_bert import BertModel

# pip install triton==2.0.0.dev20221202 --no-deps if using Pytorch 2.0

class AutoModelForSentenceEmbedding(nn.Module):
    def __init__(self, model, tokenizer, normalize=True):
        super(AutoModelForSentenceEmbedding, self).__init__()

        self.model = model.to("cuda")
        self.normalize = normalize
        self.tokenizer = tokenizer

    def forward(self, **kwargs):
        model_output = self.model(**kwargs)
        embeddings = self.mean_pooling(model_output, kwargs['attention_mask'])
        if self.normalize:
            embeddings = torch.nn.functional.normalize(embeddings, p=2, dim=1)

        return embeddings

    def mean_pooling(self, 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)

model = AutoModel.from_pretrained("<path-to-model>", trust_remote_code=True).to("cuda")
model = AutoModelForSentenceEmbedding(model, tokenizer)
tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')

sentences = ["This is an example sentence", "Each sentence is converted"]

encoded_input = tokenizer(sentences, padding=True, truncation=True, max_length=1024, return_tensors='pt').to("cuda")
embeddings = model(**encoded_input)

print(embeddings)
print(embeddings.shape)

📚 详细文档

训练

该模型在6400万个随机采样的句子/段落对上进行了训练，这些数据来自Sentence Transformers模型使用的同一训练集。训练集的详细信息可查看 此处。

训练（包括超参数）、推理和数据加载脚本均可在 这个GitHub仓库 中找到。

评估

我们在一些基于检索的基准测试（CQADupstackEnglishRetrieval、DBPedia、MSMARCO、QuoraRetrieval）上运行了该模型，结果可查看 此处。

🔧 技术细节