Dataset	State-of-the-art (Multi)	STSb-XLM-RoBERTa-base	STS Multilingual MPNet base v2
Average	73.17	71.68	73.89
STS17 (ar-ar)	81.87	80.43	81.24
STS17 (en-ar)	81.22	76.3	77.03
STS17 (en-de)	87.3	91.06	91.09
STS17 (en-tr)	77.18	80.74	79.87
STS17 (es-en)	88.24	83.09	85.53
STS17 (es-es)	88.25	84.16	87.27
STS17 (fr-en)	88.06	91.33	90.68
STS17 (it-en)	89.68	92.87	92.47
STS17 (ko-ko)	83.69	97.67	97.66
STS17 (nl-en)	88.25	92.13	91.15
STS22 (ar)	58.67	58.67	62.66
STS22 (de)	60.12	52.17	57.74
STS22 (de-en)	60.92	58.5	57.5
STS22 (de-fr)	67.79	51.28	57.99
STS22 (de-pl)	58.69	44.56	44.22
STS22 (es)	68.57	63.68	66.21
STS22 (es-en)	78.8	70.65	75.18
STS22 (es-it)	75.04	60.88	66.25
STS22 (fr)	83.75	76.46	78.76
STS22 (fr-pl)	84.52	84.52	84.52
STS22 (it)	79.28	66.73	68.47
STS22 (pl)	42.08	41.18	43.36
STS22 (pl-en)	77.5	64.35	75.11
STS22 (ru)	61.71	58.59	58.67
STS22 (tr)	68.72	57.52	63.84
STS22 (zh-en)	71.88	60.69	65.37
STSb	89.86	95.05	95.15

Bold indicates the best result in each row.

SentenceTransformer based on sentence-transformers/paraphrase-multilingual-mpnet-base-v2

This is a sentence-transformers model finetuned from sentence-transformers/paraphrase-multilingual-mpnet-base-v2. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: sentence-transformers/paraphrase-multilingual-mpnet-base-v2
Maximum Sequence Length: 128 tokens
Output Dimensionality: 768 tokens
Similarity Function: Cosine Similarity

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 128, '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, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("Gameselo/STS-multilingual-mpnet-base-v2")
# Run inference
sentences = [
    '一个女人正在洗澡。',
    'A woman is taking a bath.',
    'En jente børster håret sitt',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Semantic Similarity

Dataset: sts-dev
Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.9551
spearman_cosine	0.9593
pearson_manhattan	0.927
spearman_manhattan	0.9383
pearson_euclidean	0.9278
spearman_euclidean	0.9394
pearson_dot	0.876
spearman_dot	0.8865
pearson_max	0.9551
spearman_max	0.9593

Evalutation results vs SOTA results

Dataset: sts-test
Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.948
spearman_cosine	0.9515
pearson_manhattan	0.9252
spearman_manhattan	0.9352
pearson_euclidean	0.9258
spearman_euclidean	0.9364
pearson_dot	0.8443
spearman_dot	0.8435
pearson_max	0.948
spearman_max	0.9515

Training Details

Training Dataset

Unnamed Dataset

Size: 226,547 training samples
Columns: sentence_0, sentence_1, and label
Approximate statistics based on the first 1000 samples:
sentence_0 sentence_1 label
type string string float
details
min: 3 tokens
mean: 20.05 tokens
max: 128 tokens
min: 4 tokens
mean: 19.94 tokens
max: 128 tokens
min: 0.0
mean: 1.92
max: 398.6

Samples:

sentence_0	sentence_1	label
`Bir kadın makineye dikiş dikiyor.`	`Bir kadın biraz et ekiyor.`	`0.12`
`Snowden 'gegeven vluchtelingendocument door Ecuador'.`	`Snowden staat op het punt om uit Moskou te vliegen`	`0.24000000953674316`
`Czarny pies idzie mostem przez wodę`	`Czarny pies nie idzie mostem przez wodę`	`0.74000000954`

Loss: AnglELoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "pairwise_angle_sim" 4}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 256
per_device_eval_batch_size: 256
num_train_epochs: 10
multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
prediction_loss_only: True
per_device_train_batch_size: 256
per_device_eval_batch_size: 256
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 1
eval_accumulation_steps: None
learning_rate: 5e-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1
num_train_epochs: 10
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.0
warmup_steps: 0
log_level: passive
log_level_replica: warning
log_on_each_node: True
logging_nan_inf_filter: True
save_safetensors: True
save_on_each_node: False
save_only_model: False
no_cuda: False
use_cpu: False
use_mps_device: False
seed: 42
data_seed: None
jit_mode_eval: False
use_ipex: False
bf16: False
fp16: False
fp16_opt_level: O1
half_precision_backend: auto
bf16_full_eval: False
fp16_full_eval: False
tf32: None
local_rank: 0
ddp_backend: None
tpu_num_cores: None
tpu_metrics_debug: False
debug: []
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_prefetch_factor: None
past_index: -1
disable_tqdm: False
remove_unused_columns: True
label_names: None
load_best_model_at_end: False
ignore_data_skip: False
fsdp: []
fsdp_min_num_params: 0
fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
fsdp_transformer_layer_cls_to_wrap: None
accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'gradient_accumulation_kwargs': None}
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch
optim_args: None
adafactor: False
group_by_length: False
length_column_name: length
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
dataloader_pin_memory: True
dataloader_persistent_workers: False
skip_memory_metrics: True
use_legacy_prediction_loop: False
push_to_hub: False
resume_from_checkpoint: None
hub_model_id: None
hub_strategy: every_save
hub_private_repo: False
hub_always_push: False
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_inputs_for_metrics: False
eval_do_concat_batches: True
fp16_backend: auto
push_to_hub_model_id: None
push_to_hub_organization: None
mp_parameters:
auto_find_batch_size: False
full_determinism: False
torchdynamo: None
ray_scope: last
ddp_timeout: 1800
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
dispatch_batches: None
split_batches: None
include_tokens_per_second: False
include_num_input_tokens_seen: False
neftune_noise_alpha: None
optim_target_modules: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: round_robin

Training Logs

Epoch	Step	Training Loss	sts-dev_spearman_cosine	sts-test_spearman_cosine
0.5650	500	10.9426	-	-
1.0	885	-	0.9202	-
1.1299	1000	9.7184	-	-
1.6949	1500	9.5348	-	-
2.0	1770	-	0.9400	-
2.2599	2000	9.4412	-	-
2.8249	2500	9.3097	-	-
3.0	2655	-	0.9489	-
3.3898	3000	9.2357	-	-
3.9548	3500	9.1594	-	-
4.0	3540	-	0.9528	-
4.5198	4000	9.0963	-	-
5.0	4425	-	0.9553	-
5.0847	4500	9.0382	-	-
5.6497	5000	8.9837	-	-
6.0	5310	-	0.9567	-
6.2147	5500	8.9403	-	-
6.7797	6000	8.8841	-	-
7.0	6195	-	0.9581	-
7.3446	6500	8.8513	-	-
7.9096	7000	8.81	-	-
8.0	7080	-	0.9582	-
8.4746	7500	8.8069	-	-
9.0	7965	-	0.9589	-
9.0395	8000	8.7616	-	-
9.6045	8500	8.7521	-	-
10.0	8850	-	0.9593	0.6266

Framework Versions

Python: 3.9.7
Sentence Transformers: 3.0.0
Transformers: 4.40.1
PyTorch: 2.3.0+cu121
Accelerate: 0.29.3
Datasets: 2.19.0
Tokenizers: 0.19.1

Citation

BibTeX

Sentence Transformers

bibtex

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

AnglELoss