This model is a fine-tuned version of intfloat/e5-base on german subset of the stsb_multi_mt dataset. It achieves the following results on the evaluation set:

Validation:

Loss: 0.9118
Pearson: 0.7952 Test:
Loss: 1.2162
Pearson: 0.7252

Usage

python

#install sentence-transformers
!pip install -U sentence-transformers

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("kaixkhazaki/e5-base-german-sentence-similarity")

sentences = [
    'Ein älterer Herr genießt die Natur auf einer Parkbank.',
    'Ein alter Mann vertieft sich in eine Zeitung im Stadtpark.',
    'Ein Teenager hört Musik auf einer Bank.'
]
embeddings = model.encode(sentences)

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)

>>
tensor([[1.0000, 0.9044, 0.7919],
        [0.9044, 1.0000, 0.7757],
        [0.7919, 0.7757, 1.0000]])

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

Dataset: STS Benchmark

Training procedure

additional dropout implemented on model to avoid overfitting(hidden_dropout_prob=0.25 ,attention_probs_dropout_prob = 0.25)

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 1e-05
train_batch_size: 16
eval_batch_size: 32
seed: 42
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 500
num_epochs: 12

Training results

Training Loss	Epoch	Step	Validation Loss	Pearson
1.6556	1.0	360	1.3086	0.7176
1.0113	2.0	720	1.2079	0.7563
1.0205	3.0	1080	1.2091	0.7639
0.8876	4.0	1440	0.9325	0.7910
0.6762	5.0	1800	0.9118	0.7952
0.5615	6.0	2160	0.9772	0.7910
0.6387	7.0	2520	0.9085	0.7933

Framework versions

Transformers 4.48.0
Pytorch 2.2.2
Datasets 3.2.0
Tokenizers 0.21.0