TSjB/labse-qm

It maps sentences & paragraphs to a 768 dimensional dense vector space and can be used for tasks like clustering or semantic search.
Fine-tined by Bogdan Tewunalany
Based on LaBSE

Usage (Sentence-Transformers)

Python:

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", "Бу айтым юлгюдю"]

model = SentenceTransformer('TSjB/labse-qm')
embeddings = model.encode(sentences)
print(embeddings)

R language:

library(data.table)
library(reticulate)
library(ggplot2)
library(ggrepel)
library(Rtsne)

py_install("sentence-transformers", pip = TRUE)
st <- import("sentence_transformers")

english_sentences = base::c("dog", "Puppies are nice.", "I enjoy taking long walks along the beach with my dog.")
italian_sentences = base::c("cane", "I cuccioli sono carini.", "Mi piace fare lunghe passeggiate lungo la spiaggia con il mio cane.")
qarachay_sentences = base::c("ит", "Итле джагъымлыдыла.", "Джагъа юсю бла итим бла айланыргъа сюеме.")

model = st$SentenceTransformer('TSjB/labse-qm')

english_embeddings = model$encode(english_sentences)
italian_embeddings = model$encode(italian_sentences)
qarachay_embeddings = model$encode(qarachay_sentences)

m <- rbind(english_embeddings,
           italian_embeddings,
           qarachay_embeddings) %>% as.matrix

tsne <- Rtsne(m, perplexity = floor((nrow(m) - 1) / 3))


tSNE_df <- tsne$Y %>% 
  as.data.table() %>% 
  setnames(old = c("V1", "V2"), new = c("tSNE1", "tSNE2")) %>% 
  .[, `:=`(sentence = c(english_sentences, italian_sentences, qarachay_sentences),
           language = c(rep("english", length(english_sentences)),
                        rep("italian", length(italian_sentences)),
                        rep("qarachay", length(qarachay_sentences))))]


tSNE_df %>%
 ggplot(aes(x = tSNE1, 
            y = tSNE2,
            color = language,
            label = sentence             
             )
         )  + 
    geom_label_repel() +    
  geom_point()

Evaluation Results

For an automated evaluation of this model, see the Sentence Embeddings Benchmark: https://seb.sbert.net

Training

The model was trained with the parameters:

DataLoader:

torch.utils.data.dataloader.DataLoader of length 6439 with parameters:

{'batch_size': 8, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}

Loss:

sentence_transformers.losses.MultipleNegativesRankingLoss.MultipleNegativesRankingLoss with parameters:

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

Parameters of the fit()-Method:

{
    "epochs": 1,
    "evaluation_steps": 100,
    "evaluator": "__main__.ChainScoreEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
    "optimizer_params": {
        "lr": 2e-05
    },
    "scheduler": "warmupcosine",
    "steps_per_epoch": null,
    "warmup_steps": 1000,
    "weight_decay": 0.01
}

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 256, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False})
  (2): Dense({'in_features': 768, 'out_features': 768, 'bias': True, 'activation_function': 'torch.nn.modules.activation.Tanh'})
  (3): Normalize()
)