embedder-100p
This is a ms-marco bi-encoder from sentence-transformers model: It maps sentences & paragraphs to a 768 dimensional dense vector space and can be used for tasks like clustering or semantic search. It is trained on more than 20GiB of german text. It used the knowledge distillation to be a bi-language embedding model (English and German).
Usage (Sentence-Transformers)
Using this model becomes easy when you have sentence-transformers installed:
pip install -U sentence-transformers
Then you can use the model like this:
from sentence_transformers import SentenceTransformer
sentences = ["This is an example sentence", "Each sentence is converted"]
model = SentenceTransformer('embedder-100p')
embeddings = model.encode(sentences)
print(embeddings)
Usage (HuggingFace Transformers)
Without sentence-transformers, you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.
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 = ['This is an example sentence', 'Each sentence is converted']
# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('embedder-100p')
model = AutoModel.from_pretrained('embedder-100p')
# 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 Results
The evaluation on MTEB
Training
The model was trained with the parameters:
DataLoader:
torch.utils.data.dataloader.DataLoader
of length 231230 with parameters:
{'batch_size': 16, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
Loss:
sentence_transformers.losses.MSELoss.MSELoss
Parameters of the fit()-Method:
{
"epochs": 20,
"evaluation_steps": 1000,
"evaluator": "sentence_transformers.evaluation.SequentialEvaluator.SequentialEvaluator",
"max_grad_norm": 1,
"optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
"optimizer_params": {
"eps": 1e-06,
"lr": 7e-06
},
"scheduler": "WarmupLinear",
"steps_per_epoch": null,
"warmup_steps": 5000,
"weight_decay": 0.01
}
Full Model Architecture
SentenceTransformer(
(0): Transformer({'max_seq_length': 384, 'do_lower_case': False}) with Transformer model: XLMRobertaModel
(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})
)
By
- Downloads last month
- 4,432
Evaluation results
- accuracy on MTEB AmazonCounterfactualClassification (en)test set self-reported67.060
- ap on MTEB AmazonCounterfactualClassification (en)test set self-reported30.376
- f1 on MTEB AmazonCounterfactualClassification (en)test set self-reported61.305
- accuracy on MTEB AmazonPolarityClassificationtest set self-reported70.409
- ap on MTEB AmazonPolarityClassificationtest set self-reported64.616
- f1 on MTEB AmazonPolarityClassificationtest set self-reported70.281
- accuracy on MTEB AmazonReviewsClassification (en)test set self-reported33.214
- f1 on MTEB AmazonReviewsClassification (en)test set self-reported33.123
- map_at_1 on MTEB ArguAnatest set self-reported27.312
- map_at_10 on MTEB ArguAnatest set self-reported42.761
- map_at_100 on MTEB ArguAnatest set self-reported43.691
- map_at_1000 on MTEB ArguAnatest set self-reported43.698
- map_at_3 on MTEB ArguAnatest set self-reported37.091
- map_at_5 on MTEB ArguAnatest set self-reported40.398
- mrr_at_1 on MTEB ArguAnatest set self-reported28.165
- mrr_at_10 on MTEB ArguAnatest set self-reported43.050
- mrr_at_100 on MTEB ArguAnatest set self-reported43.994
- mrr_at_1000 on MTEB ArguAnatest set self-reported44.000
- mrr_at_3 on MTEB ArguAnatest set self-reported37.376
- mrr_at_5 on MTEB ArguAnatest set self-reported40.665
- ndcg_at_1 on MTEB ArguAnatest set self-reported27.312
- ndcg_at_10 on MTEB ArguAnatest set self-reported52.035
- ndcg_at_100 on MTEB ArguAnatest set self-reported55.891
- ndcg_at_1000 on MTEB ArguAnatest set self-reported56.043
- ndcg_at_3 on MTEB ArguAnatest set self-reported40.380
- ndcg_at_5 on MTEB ArguAnatest set self-reported46.364
- precision_at_1 on MTEB ArguAnatest set self-reported27.312
- precision_at_10 on MTEB ArguAnatest set self-reported8.193
- precision_at_100 on MTEB ArguAnatest set self-reported0.985
- precision_at_1000 on MTEB ArguAnatest set self-reported0.100
- precision_at_3 on MTEB ArguAnatest set self-reported16.643
- precision_at_5 on MTEB ArguAnatest set self-reported12.902
- recall_at_1 on MTEB ArguAnatest set self-reported27.312
- recall_at_10 on MTEB ArguAnatest set self-reported81.935
- recall_at_100 on MTEB ArguAnatest set self-reported98.506
- recall_at_1000 on MTEB ArguAnatest set self-reported99.644
- recall_at_3 on MTEB ArguAnatest set self-reported49.929
- recall_at_5 on MTEB ArguAnatest set self-reported64.509
- v_measure on MTEB ArxivClusteringP2Ptest set self-reported42.899
- v_measure on MTEB ArxivClusteringS2Stest set self-reported32.449
- map on MTEB AskUbuntuDupQuestionstest set self-reported61.051
- mrr on MTEB AskUbuntuDupQuestionstest set self-reported73.872
- cos_sim_pearson on MTEB BIOSSEStest set self-reported80.068
- cos_sim_spearman on MTEB BIOSSEStest set self-reported75.318
- euclidean_pearson on MTEB BIOSSEStest set self-reported77.705
- euclidean_spearman on MTEB BIOSSEStest set self-reported74.076
- manhattan_pearson on MTEB BIOSSEStest set self-reported77.180
- manhattan_spearman on MTEB BIOSSEStest set self-reported74.106
- accuracy on MTEB Banking77Classificationtest set self-reported82.731
- f1 on MTEB Banking77Classificationtest set self-reported82.620
- v_measure on MTEB BiorxivClusteringP2Ptest set self-reported37.236