MesTruck/STS-multilingual-mpnet-base-v2-GGUF

State-of-the-Art Results Comparison (MTEB STS Multilingual Leaderboard)

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.

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:

  {
      "scale": 20.0,
      "similarity_fct": "pairwise_angle_sim"
  }
  

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

@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

@misc{li2023angleoptimized,
    title={AnglE-optimized Text Embeddings}, 
    author={Xianming Li and Jing Li},
    year={2023},
    eprint={2309.12871},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}