模型:
nickprock/mmarco-bert-base-italian-uncased
这是一个 sentence-transformers 模型:它将句子和段落映射到一个768维的密集向量空间,并可用于聚类或语义搜索等任务。
安装了 sentence-transformers 后,使用该模型变得很容易:
pip install -U sentence-transformers
然后您可以像这样使用模型:
from sentence_transformers import SentenceTransformer, util query = "Quante persone vivono a Londra?" docs = ["A Londra vivono circa 9 milioni di persone", "Londra è conosciuta per il suo quartiere finanziario"] #Load the model model = SentenceTransformer('nickprock/mmarco-bert-base-italian-uncased') #Encode query and documents query_emb = model.encode(query) doc_emb = model.encode(docs) #Compute dot score between query and all document embeddings scores = util.dot_score(query_emb, doc_emb)[0].cpu().tolist() #Combine docs & scores doc_score_pairs = list(zip(docs, scores)) #Sort by decreasing score doc_score_pairs = sorted(doc_score_pairs, key=lambda x: x[1], reverse=True) #Output passages & scores for doc, score in doc_score_pairs: print(score, doc)
如果没有 sentence-transformers ,您可以像这样使用模型:首先,将输入通过变换器模型,然后必须在上下文化的词嵌入之上应用正确的汇集操作。
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.last_hidden_state 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) #Encode text def encode(texts): # Tokenize sentences encoded_input = tokenizer(texts, padding=True, truncation=True, return_tensors='pt') # Compute token embeddings with torch.no_grad(): model_output = model(**encoded_input, return_dict=True) # Perform pooling embeddings = mean_pooling(model_output, encoded_input['attention_mask']) return embeddings # Sentences we want sentence embeddings for query = "Quante persone vivono a Londra?" docs = ["A Londra vivono circa 9 milioni di persone", "Londra è conosciuta per il suo quartiere finanziario"] # Load model from HuggingFace Hub tokenizer = AutoTokenizer.from_pretrained("nickprock/mmarco-bert-base-italian-uncased") model = AutoModel.from_pretrained("nickprock/mmarco-bert-base-italian-uncased") #Encode query and docs query_emb = encode(query) doc_emb = encode(docs) #Compute dot score between query and all document embeddings scores = torch.mm(query_emb, doc_emb.transpose(0, 1))[0].cpu().tolist() #Combine docs & scores doc_score_pairs = list(zip(docs, scores)) #Sort by decreasing score doc_score_pairs = sorted(doc_score_pairs, key=lambda x: x[1], reverse=True) #Output passages & scores print("Query:", query) for doc, score in doc_score_pairs: print(score, doc)
有关此模型的自动评估,请参见句子嵌入基准 https://seb.sbert.net :
该模型是使用以下参数进行训练的:
DataLoader:
torch.utils.data.dataloader.DataLoader,长度为6250,具有以下参数:
{'batch_size': 16, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
损失:
sentence_transformers.losses.TripletLoss.TripletLoss,具有以下参数:
{'distance_metric': 'TripletDistanceMetric.EUCLIDEAN', 'triplet_margin': 5}
fit()方法的参数:
{ "epochs": 10, "evaluation_steps": 500, "evaluator": "sentence_transformers.evaluation.TripletEvaluator.TripletEvaluator", "max_grad_norm": 1, "optimizer_class": "<class 'torch.optim.adamw.AdamW'>", "optimizer_params": { "lr": 2e-05 }, "scheduler": "WarmupLinear", "steps_per_epoch": 1500, "warmup_steps": 6250, "weight_decay": 0.01 }
SentenceTransformer( (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel (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}) )