yahyaabd/allstat-semantic-search-mpnet-base-v3-sts

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 on the bps-query-publication-similarity-pairs dataset. 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 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • bps-query-publication-similarity-pairs

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("yahyaabd/allstat-semantic-search-mpnet-base-v3-sts")
# Run inference
sentences = [
    'Sistem neraca lingkungan dan ekonomi Indonesia, -',
    'Sistem Terintegrasi Neraca Lingkungan dan Ekonomi Indonesia -',
    'Distribusi Perdagangan Komoditas Minyak Goreng Indonesia ',
]
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

• Datasets: allstat-semantic-dev and allstat-semantic-test
• Evaluated with EmbeddingSimilarityEvaluator

Metric	allstat-semantic-dev	allstat-semantic-test
pearson_cosine	0.9672	0.9644
spearman_cosine	0.8714	0.8572

Training Details

Training Dataset

bps-query-publication-similarity-pairs

• Dataset: bps-query-publication-similarity-pairs at cf2836e
• Size: 44,668 training samples
• Columns: query, doc_title, and score
• Approximate statistics based on the first 1000 samples:

  	query	doc_title	score
  type	string	string	float
  details	min: 4 tokens mean: 9.66 tokens max: 36 tokens	min: 4 tokens mean: 11.88 tokens max: 49 tokens	min: 0.0 mean: 0.49 max: 1.0

• Samples:

  query	doc_title	score
  Tren bisnis perikanan di Indonesia	Statistik Perusahaan Perikanan	0.88
  Statistik APBDes	Statistik Perusahaan Peternakan Ternak Besar dan Kecil	0.29
  Laporan Indikator Konstruksi semester 1	Statistik Air Bersih -	0.25

• Loss: CosineSimilarityLoss with these parameters:

  {
      "loss_fct": "torch.nn.modules.loss.MSELoss"
  }
  

Evaluation Dataset

bps-query-publication-similarity-pairs

• Dataset: bps-query-publication-similarity-pairs at cf2836e
• Size: 2,482 evaluation samples
• Columns: query, doc_title, and score
• Approximate statistics based on the first 1000 samples:

  	query	doc_title	score
  type	string	string	float
  details	min: 4 tokens mean: 9.55 tokens max: 31 tokens	min: 4 tokens mean: 11.62 tokens max: 36 tokens	min: 0.0 mean: 0.54 max: 1.0

• Samples:

  query	doc_title	score
  Dampak COVID-19 pada usaha mikro kecil	Statistik Penyedia Makan Minum	0.2
  Sektor konstruksi Aceh, data UMKM	Profil Usaha Konstruksi Perorangan Provinsi Aceh,	0.88
  SP2010: Statistik lansia Sumatera Selatan	Statistik Penduduk Lanjut Usia Provinsi Sumatera Selatan -Hasil Sensus Penduduk	0.81

• Loss: CosineSimilarityLoss with these parameters:

  {
      "loss_fct": "torch.nn.modules.loss.MSELoss"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• num_train_epochs: 4
• warmup_ratio: 0.1
• fp16: True

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• torch_empty_cache_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: 1.0
• num_train_epochs: 4
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.1
• 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
• restore_callback_states_from_checkpoint: 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: True
• 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, 'non_blocking': False, '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: None
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• include_for_metrics: []
• 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_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss	Validation Loss	allstat-semantic-dev_spearman_cosine	allstat-semantic-test_spearman_cosine
0.0358	100	0.0498	0.0311	0.7840	-
0.0716	200	0.0294	0.0245	0.7970	-
0.1074	300	0.0241	0.0210	0.8040	-
0.1433	400	0.0215	0.0192	0.8078	-
0.1791	500	0.0208	0.0200	0.8091	-
0.2149	600	0.0208	0.0183	0.8183	-
0.2507	700	0.0216	0.0176	0.8177	-
0.2865	800	0.02	0.0177	0.8192	-
0.3223	900	0.0183	0.0180	0.8107	-
0.3582	1000	0.0197	0.0190	0.8058	-
0.3940	1100	0.0199	0.0176	0.8182	-
0.4298	1200	0.0207	0.0193	0.8097	-
0.4656	1300	0.0186	0.0190	0.8088	-
0.5014	1400	0.0197	0.0178	0.8122	-
0.5372	1500	0.0179	0.0177	0.8161	-
0.5731	1600	0.0171	0.0169	0.8197	-
0.6089	1700	0.0178	0.0162	0.8152	-
0.6447	1800	0.0152	0.0162	0.8234	-
0.6805	1900	0.0171	0.0162	0.8187	-
0.7163	2000	0.0179	0.0154	0.8194	-
0.7521	2100	0.0164	0.0158	0.8126	-
0.7880	2200	0.016	0.0149	0.8254	-
0.8238	2300	0.0164	0.0149	0.8193	-
0.8596	2400	0.0151	0.0139	0.8297	-
0.8954	2500	0.0151	0.0142	0.8306	-
0.9312	2600	0.0136	0.0143	0.8315	-
0.9670	2700	0.0157	0.0135	0.8342	-
1.0029	2800	0.0133	0.0135	0.8330	-
1.0387	2900	0.0116	0.0133	0.8369	-
1.0745	3000	0.0106	0.0132	0.8357	-
1.1103	3100	0.0113	0.0126	0.8395	-
1.1461	3200	0.0123	0.0131	0.8362	-
1.1819	3300	0.0117	0.0142	0.8289	-
1.2178	3400	0.0133	0.0135	0.8322	-
1.2536	3500	0.0113	0.0129	0.8358	-
1.2894	3600	0.0109	0.0132	0.8352	-
1.3252	3700	0.0107	0.0122	0.8394	-
1.3610	3800	0.0125	0.0128	0.8364	-
1.3968	3900	0.012	0.0126	0.8342	-
1.4327	4000	0.0123	0.0128	0.8364	-
1.4685	4100	0.0109	0.0127	0.8369	-
1.5043	4200	0.0108	0.0125	0.8385	-
1.5401	4300	0.011	0.0124	0.8416	-
1.5759	4400	0.0104	0.0120	0.8455	-
1.6117	4500	0.0107	0.0114	0.8498	-
1.6476	4600	0.0095	0.0114	0.8485	-
1.6834	4700	0.0114	0.0118	0.8457	-
1.7192	4800	0.0101	0.0118	0.8417	-
1.7550	4900	0.0127	0.0113	0.8466	-
1.7908	5000	0.0112	0.0114	0.8466	-
1.8266	5100	0.0095	0.0109	0.8485	-
1.8625	5200	0.0107	0.0114	0.8465	-
1.8983	5300	0.0113	0.0115	0.8454	-
1.9341	5400	0.0107	0.0116	0.8473	-
1.9699	5500	0.0102	0.0111	0.8526	-
2.0057	5600	0.0097	0.0109	0.8542	-
2.0415	5700	0.0082	0.0106	0.8534	-
2.0774	5800	0.0069	0.0107	0.8551	-
2.1132	5900	0.0077	0.0107	0.8533	-
2.1490	6000	0.0076	0.0109	0.8532	-
2.1848	6100	0.0071	0.0107	0.8515	-
2.2206	6200	0.0075	0.0104	0.8563	-
2.2564	6300	0.0074	0.0102	0.8567	-
2.2923	6400	0.0083	0.0105	0.8567	-
2.3281	6500	0.0075	0.0107	0.8515	-
2.3639	6600	0.007	0.0103	0.8546	-
2.3997	6700	0.0079	0.0103	0.8559	-
2.4355	6800	0.0072	0.0102	0.8550	-
2.4713	6900	0.0069	0.0098	0.8618	-
2.5072	7000	0.0082	0.0099	0.8611	-
2.5430	7100	0.0067	0.0101	0.8596	-
2.5788	7200	0.0062	0.0097	0.8593	-
2.6146	7300	0.0074	0.0094	0.8622	-
2.6504	7400	0.008	0.0093	0.8624	-
2.6862	7500	0.0066	0.0097	0.8610	-
2.7221	7600	0.0066	0.0098	0.8616	-
2.7579	7700	0.0066	0.0097	0.8593	-
2.7937	7800	0.0076	0.0099	0.8582	-
2.8295	7900	0.0078	0.0094	0.8625	-
2.8653	8000	0.0075	0.0092	0.8639	-
2.9011	8100	0.0077	0.0092	0.8620	-
2.9370	8200	0.0067	0.0092	0.8643	-
2.9728	8300	0.0069	0.0095	0.8625	-
3.0086	8400	0.0067	0.0095	0.8632	-
3.0444	8500	0.0051	0.0093	0.8652	-
3.0802	8600	0.0046	0.0094	0.8662	-
3.1160	8700	0.0046	0.0094	0.8669	-
3.1519	8800	0.0047	0.0095	0.8671	-
3.1877	8900	0.0049	0.0091	0.8688	-
3.2235	9000	0.0048	0.0090	0.8688	-
3.2593	9100	0.0047	0.0092	0.8697	-
3.2951	9200	0.0058	0.0092	0.8686	-
3.3309	9300	0.005	0.0091	0.8681	-
3.3668	9400	0.0049	0.0090	0.8694	-
3.4026	9500	0.0051	0.0091	0.8670	-
3.4384	9600	0.0048	0.0090	0.8666	-
3.4742	9700	0.0047	0.0089	0.8672	-
3.5100	9800	0.0046	0.0091	0.8658	-
3.5458	9900	0.0051	0.0090	0.8658	-
3.5817	10000	0.0054	0.0089	0.8681	-
3.6175	10100	0.0049	0.0089	0.8679	-
3.6533	10200	0.0042	0.0089	0.8681	-
3.6891	10300	0.0049	0.0089	0.8684	-
3.7249	10400	0.0046	0.0088	0.8692	-
3.7607	10500	0.0048	0.0088	0.8691	-
3.7966	10600	0.0042	0.0088	0.8704	-
3.8324	10700	0.0049	0.0088	0.8702	-
3.8682	10800	0.0045	0.0088	0.8709	-
3.9040	10900	0.0047	0.0088	0.8712	-
3.9398	11000	0.0046	0.0088	0.8711	-
3.9756	11100	0.0045	0.0088	0.8714	-
4.0	11168	-	-	-	0.8572

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.3.1
• Transformers: 4.47.1
• PyTorch: 2.2.2+cu121
• Accelerate: 1.2.1
• Datasets: 3.2.0
• Tokenizers: 0.21.0

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",
}