This is a 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. The model was trained on Turkish machine translated versions of NLI and STS-b datasets, using example training scripts from sentence-transformers GitHub repository.

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:

python

from sentence_transformers import SentenceTransformer
sentences = ["Bu örnek bir cümle", "Her cümle vektöre çevriliyor"]

model = SentenceTransformer('emrecan/bert-base-turkish-cased-mean-nli-stsb-tr')
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.

python

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 = ["Bu örnek bir cümle", "Her cümle vektöre çevriliyor"]

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('emrecan/bert-base-turkish-cased-mean-nli-stsb-tr')
model = AutoModel.from_pretrained('emrecan/bert-base-turkish-cased-mean-nli-stsb-tr')

# 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

Evaluation results on test and development sets are given below:

Split	Epoch	cosine_pearson	cosine_spearman	euclidean_pearson	euclidean_spearman	manhattan_pearson	manhattan_spearman	dot_pearson	dot_spearman
test	-	0.834	0.830	0.820	0.819	0.819	0.818	0.799	0.789
validation	1	0.850	0.848	0.831	0.835	0.83	0.83	0.80	0.806
validation	2	0.857	0.857	0.844	0.848	0.844	0.848	0.813	0.810
validation	3	0.860	0.859	0.846	0.851	0.846	0.850	0.825	0.822
validation	4	0.859	0.860	0.846	0.851	0.846	0.851	0.825	0.823

Training

Training scripts training_nli_v2.py and training_stsbenchmark_continue_training.py were used to train the model.

The model was trained with the parameters:

DataLoader:

torch.utils.data.dataloader.DataLoader of length 360 with parameters:

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

Loss:

sentence_transformers.losses.CosineSimilarityLoss.CosineSimilarityLoss

Parameters of the fit()-Method:

{
    "epochs": 4,
    "evaluation_steps": 200,
    "evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'transformers.optimization.AdamW'>",
    "optimizer_params": {
        "lr": 2e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 144,
    "weight_decay": 0.01
}

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 75, '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})
)

Citing & Authors

@misc{celik2020turkishsbert,
  author       = {Emrecan Çelik},
  title        = {bert-base-turkish-cased-mean-nli-stsb-tr},
  howpublished = {Hugging Face model repository},
  url          = {https://huggingface.co/emrecan/bert-base-turkish-cased-mean-nli-stsb-tr},
  note         = {Sentence-transformer model trained on Turkish NLI and STS-B data}
}