Multi Sentence BERTino

🚀 多句子BERTino模型

這是一個用於句子相似度任務的模型，它能將句子和段落映射到768維的密集向量空間，可用於聚類或語義搜索等任務。該模型基於特定數據集訓練，能有效處理相關自然語言處理任務。

🚀 快速開始

本模型可通過不同方式使用，下面將分別介紹不同庫下的使用方法。

✨ 主要特性

• 基於 sentence-transformers 框架，能將句子和段落映射到768維的密集向量空間。
• 採用 indigo-ai/BERTino 進行訓練，使用了 mmarco italian (200K) 和 stsb italian 數據集。

📦 安裝指南

使用 sentence-transformers 庫

安裝 sentence-transformers 庫：

pip install -U sentence-transformers

使用 FastEmbed 庫

安裝 FastEmbed 庫：

pip install fastembed

💻 使用示例

使用 sentence-transformers 庫

from sentence_transformers import SentenceTransformer
sentences = ["Una ragazza si acconcia i capelli.", "Una ragazza si sta spazzolando i capelli."]

model = SentenceTransformer('nickprock/multi-sentence-BERTino')
embeddings = model.encode(sentences)
print(embeddings)

使用 FastEmbed 庫

from fastembed import TextEmbedding
from fastembed.common.model_description import PoolingType, ModelSource

sentences = ["Una ragazza si acconcia i capelli.", "Una ragazza si sta spazzolando i capelli."]

TextEmbedding.add_custom_model(
    model="nickprock/multi-sentence-BERTino",
    pooling=PoolingType.MEAN,
    normalization=True,
    sources=ModelSource(hf="nickprock/multi-sentence-BERTino"),  # can be used with an `url` to load files from a private storage
    dim=768,
    model_file="onnx/model_qint8_avx512_vnni.onnx",  # can be used to load an already supported model with another optimization or quantization, e.g. onnx/model_O4.onnx
)

model = TextEmbedding(model_name="nickprock/multi-sentence-BERTino")
embeddings = list(model.embed(sentences))
print(embeddings)

使用 HuggingFace Transformers 庫

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 = ['Una ragazza si acconcia i capelli.', 'Una ragazza si sta spazzolando i capelli.']

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('nickprock/multi-sentence-BERTino')
model = AutoModel.from_pretrained('nickprock/multi-sentence-BERTino')

# 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)

📚 詳細文檔

評估結果

要對該模型進行自動評估，請參考 Sentence Embeddings Benchmark: https://seb.sbert.net

訓練參數

該模型的訓練參數如下：

數據加載器與損失函數設置

• DataLoader 1: torch.utils.data.dataloader.DataLoader 長度為 31250，參數如下：

  {'batch_size': 16, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
  

  Loss 1: sentence_transformers.losses.TripletLoss.TripletLoss，參數為：

  {'distance_metric': 'TripletDistanceMetric.EUCLIDEAN', 'triplet_margin': 5}
  

• DataLoader 2: torch.utils.data.dataloader.DataLoader 長度為 360，參數如下：

  {'batch_size': 16, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
  

  Loss 2: sentence_transformers.losses.CosineSimilarityLoss.CosineSimilarityLoss
• DataLoader 3: torch.utils.data.dataloader.DataLoader 長度為 31250，參數如下：

  {'batch_size': 16, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
  

  Loss 3: sentence_transformers.losses.CachedMultipleNegativesRankingLoss.CachedMultipleNegativesRankingLoss，參數為：

  {'scale': 20.0, 'similarity_fct': 'cos_sim'}
  

fit() 方法參數

{
    "epochs": 10,
    "evaluation_steps": 500,
    "evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
    "optimizer_params": {
        "lr": 2e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": 1500,
    "warmup_steps": 360,
    "weight_decay": 0.01
}

完整模型架構

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: DistilBertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False})
)

📄 許可證

本模型採用 MIT 許可證。