yahyaabd/paraphrase-multilingual-miniLM-L12-V2-mnrl

SentenceTransformer based on sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2

This is a sentence-transformers model finetuned from sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2 on the statictable-triplets-all dataset. It maps sentences & paragraphs to a 384-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-MiniLM-L12-v2
• Maximum Sequence Length: 128 tokens
• Output Dimensionality: 384 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • statictable-triplets-all

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: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, '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/paraphrase-multilingual-miniLM-L12-V2-mnrl")
# Run inference
sentences = [
    'Informasi lengkap dan terbaru mengenai statistik edukasi',
    'Statistik Pendidikan Tahunan',
    'Struktur Ongkos Riil Usaha Ternak dan Unggas di Rumah Tangga dengan Pola Pemeliharaan Dikandangkan, 2017',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

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

Evaluation

Metrics

Information Retrieval

• Dataset: bps-statictable-ir
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.8925
cosine_accuracy@3	0.9772
cosine_accuracy@5	0.987
cosine_accuracy@10	0.987
cosine_precision@1	0.8925
cosine_precision@3	0.3616
cosine_precision@5	0.2397
cosine_precision@10	0.1404
cosine_recall@1	0.6962
cosine_recall@3	0.7776
cosine_recall@5	0.7978
cosine_recall@10	0.8202
cosine_ndcg@1	0.8925
cosine_ndcg@3	0.839
cosine_ndcg@5	0.8325
cosine_ndcg@10	0.8287
cosine_mrr@1	0.8925
cosine_mrr@3	0.9321
cosine_mrr@5	0.9344
cosine_mrr@10	0.9344
cosine_map@1	0.8925
cosine_map@3	0.7985
cosine_map@5	0.7862
cosine_map@10	0.7774

Information Retrieval

• Dataset: bps-statictable-ir
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.8893
cosine_accuracy@3	0.9707
cosine_accuracy@5	0.987
cosine_accuracy@10	0.9967
cosine_precision@1	0.8893
cosine_precision@3	0.3822
cosine_precision@5	0.2697
cosine_precision@10	0.1684
cosine_recall@1	0.6681
cosine_recall@3	0.7612
cosine_recall@5	0.7949
cosine_recall@10	0.8141
cosine_ndcg@1	0.8893
cosine_ndcg@3	0.8449
cosine_ndcg@5	0.8453
cosine_ndcg@10	0.8374
cosine_mrr@1	0.8893
cosine_mrr@3	0.9273
cosine_mrr@5	0.931
cosine_mrr@10	0.9322
cosine_map@1	0.8893
cosine_map@3	0.8074
cosine_map@5	0.7989
cosine_map@10	0.7836

Training Details

Training Dataset

statictable-triplets-all

• Dataset: statictable-triplets-all at 0ef226c
• Size: 10,998 training samples
• Columns: query, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative
  type	string	string	string
  details	min: 6 tokens mean: 17.69 tokens max: 33 tokens	min: 5 tokens mean: 25.79 tokens max: 58 tokens	min: 4 tokens mean: 25.72 tokens max: 58 tokens

• Samples:

  query	positive	negative
  Neraca arus kas triwulan II 2005 (ringkasan, )	Ringkasan Neraca Arus Dana, Triwulan Kedua, 2005, (Miliar Rupiah)	Rata-rata Upah/Gaji Bersih Sebulan Buruh/Karyawan/Pegawai Menurut Provinsi dan Jenis Pekerjaan Utama (Rupiah), 2017
  Hasil tangkapan ikan per provinsi, bedakan jenis penangkapan, 2013	Produksi Perikanan Tangkap Menurut Provinsi dan Jenis Penangkapan, 2000-2020	Ringkasan Neraca Arus Dana, Triwulan II, 2006, (Miliar Rupiah)
  Bagaimana perubahan distribusi pengeluaran?	Persentase Perkembangan Distribusi Pengeluaran	Angka Kematian Bayi/AKB (Infant Mortality Rate/IMR) Menurut Provinsi, 1971-2020

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Evaluation Dataset

statictable-triplets-all

• Dataset: statictable-triplets-all at 0ef226c
• Size: 10,998 evaluation samples
• Columns: query, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative
  type	string	string	string
  details	min: 5 tokens mean: 17.61 tokens max: 34 tokens	min: 4 tokens mean: 25.58 tokens max: 58 tokens	min: 5 tokens mean: 25.33 tokens max: 58 tokens

• Samples:

  query	positive	negative
  Data total penghasilan berbagai golongan rumah tangga setelah dipotong pajak, tahun 2000 (dalam )	Jumlah Pendapatan Setelah Pajak Menurut Golongan Rumah Tangga (miliar rupiah), 2000, 2005, dan 2008	Indeks Harga Konsumen per Kelompok di 82 Kota 1 (2012=100)
  Bagaimana perkembangan impor barang modal pada tahun 2020	Impor Barang Modal, 1996-2023	Indeks Harga yang Diterima Petani (It), Indes Harga yang Dibayar Petani (Ib), dan Nilai Tukar Petani Subsektor Hortikultura (NTPH) di Indonesia (2007=100), 2008-2016
  Konsumsi makanan per orang di Kalut: data mingguan, beda kelompok pengeluaran (2018)	Rata-rata Konsumsi dan Pengeluaran Perkapita Seminggu Menurut Komoditi Makanan dan Golongan Pengeluaran per Kapita Seminggu di Provinsi Kalimantan Utara, 2018-2023	Ekspor Kimia Dasar Organik yang Bersumber dari Hasil Pertanian menurut Negara Tujuan Utama, 2012-2023

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• weight_decay: 0.01
• warmup_ratio: 0.1
• save_on_each_node: True
• fp16: True
• dataloader_num_workers: 2
• load_best_model_at_end: True
• eval_on_start: True
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• 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: 0.01
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 3
• 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: True
• 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: 2
• dataloader_prefetch_factor: None
• past_index: -1
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: True
• 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: True
• use_liger_kernel: False
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	Validation Loss	bps-statictable-ir_cosine_ndcg@10
0	0	-	1.6770	0.4644
0.0645	20	1.2483	0.7232	0.6311
0.1290	40	0.5223	0.3231	0.7470
0.1935	60	0.2825	0.2193	0.7888
0.2581	80	0.1617	0.1823	0.7956
0.3226	100	0.2261	0.1141	0.8087
0.3871	120	0.1326	0.0731	0.8114
0.4516	140	0.0782	0.0611	0.8259
0.5161	160	0.0948	0.0605	0.8227
0.5806	180	0.1009	0.0534	0.8315
0.6452	200	0.0696	0.0511	0.8280
0.7097	220	0.0698	0.0431	0.8312
0.7742	240	0.0424	0.0405	0.8369
0.8387	260	0.0521	0.0378	0.8243
0.9032	280	0.0393	0.0339	0.8261
0.9677	300	0.0309	0.0315	0.8339
1.0323	320	0.0226	0.0325	0.8322
1.0968	340	0.0249	0.0323	0.8299
1.1613	360	0.0318	0.0320	0.8297
1.2258	380	0.0192	0.0278	0.8270
1.2903	400	0.025	0.0247	0.8279
1.3548	420	0.0253	0.0219	0.8280
1.4194	440	0.0359	0.0216	0.8294
1.4839	460	0.0252	0.0225	0.8227
1.5484	480	0.0217	0.0209	0.8347
1.6129	500	0.0233	0.0210	0.8346
1.6774	520	0.0131	0.0225	0.8343
1.7419	540	0.0135	0.0211	0.8330
1.8065	560	0.0133	0.0207	0.8308
1.8710	580	0.0235	0.0208	0.8352
1.9355	600	0.02	0.0190	0.8317
2.0	620	0.0142	0.0182	0.8305
2.0645	640	0.0088	0.0174	0.8330
2.1290	660	0.0103	0.0171	0.8324
2.1935	680	0.0141	0.0174	0.8281
2.2581	700	0.0209	0.0169	0.8329
2.3226	720	0.0137	0.0159	0.8316
2.3871	740	0.0173	0.0159	0.8271
2.4516	760	0.0115	0.0167	0.8230
2.5161	780	0.0155	0.0167	0.8278
2.5806	800	0.0136	0.0164	0.8280
2.6452	820	0.0085	0.0156	0.8297
2.7097	840	0.0113	0.0154	0.8293
2.7742	860	0.0094	0.0156	0.8280
2.8387	880	0.0108	0.0150	0.8280
2.9032	900	0.0081	0.0149	0.8275
2.9677	920	0.0179	0.0147	0.8287
-1	-1	-	-	0.8374

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.4.0
• Transformers: 4.48.1
• PyTorch: 2.5.1+cu124
• Accelerate: 1.3.0
• 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",
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}