Nimsara2001/labse-sinhala-finetuned

SentenceTransformer based on sentence-transformers/LaBSE

This is a sentence-transformers model finetuned from sentence-transformers/LaBSE. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: sentence-transformers/LaBSE
• Maximum Sequence Length: 64 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity
• Supported Modality: Text

Model Sources

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

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'BertModel'})
  (1): Pooling({'embedding_dimension': 768, 'pooling_mode': 'cls', 'include_prompt': True})
  (2): Dense({'in_features': 768, 'out_features': 768, 'bias': True, 'activation_function': 'torch.nn.modules.activation.Tanh', 'module_input_name': 'sentence_embedding', 'module_output_name': 'sentence_embedding'})
  (3): Normalize({})
)

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("sentence_transformers_model_id")
# Run inference
sentences = [
    'ආසන්න වශයෙන් සන්නද්ධ සේවා සහ පොලිස් නිලධාරීන්  75ක් වඩා හොඳ පරිපාලනය සඳහා අනුයුක්ත කරන ලදී.',
    'සිවිල් ආරක්ෂක දෙපාර්තමේන්තුවේ වත්මන් තුන්වන අධ්\u200dයක්ෂ ජනරාල්\xa0චන්ද්\u200dරරත්න පල්ලේගම ( MA, BSc (Hons), PgD, JP (All-Island), FCPM, MAAT (SL) ),\xa0මහතා ශ්\u200dරි ලංකා පරිපාලන සේවයේ (SLAS) විශේෂ ශ්\u200dරේණියේ නිලධාරියෙකි',
    '39,800 කට අධික පිරිසක් සියලු දිස්ත්\u200dරික්ක සහ පළාත්වල සේවය කරන නමුත් , ඉන් බොහෝ පිරිසක් උතුරු හා නැගෙනහිර පළාත්වල, එල්ටීටීඊ ප්\u200dරහාර එල්ල වීමෙන්\xa0පීඩාවට පත් ගම්මානවල සේවයේ යොදවා ඇත .',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.6430, 0.5702],
#         [0.6430, 1.0000, 0.2757],
#         [0.5702, 0.2757, 1.0000]])

Training Details

Training Dataset

Unnamed Dataset

• Size: 262,664 training samples
• Columns: sentence_0, sentence_1, and sentence_2
• Approximate statistics based on the first 100 samples:

  	sentence_0	sentence_1	sentence_2
  type	string	string	string
  modality	text	text	text
  details	min: 6 tokens mean: 30.52 tokens max: 64 tokens	min: 9 tokens mean: 30.84 tokens max: 64 tokens	min: 5 tokens mean: 34.91 tokens max: 64 tokens

• Samples:

  sentence_0	sentence_1	sentence_2
  කෙසේනමුත්, මෙම පංච රථ ඉන්දියානු දේවාල ගෘහනිර්මාණ ශිල්පයේ ප්‍රගමනයට පූර්වාදර්ශයක් වී ඇත.	සෙස පංච රථ සතර මෙන් මෙම පාෂාණමය රථය ද මින් පෙර පැවති දැවමය නිර්මාණයක අනුරුවක් විය හැක.	සියලුම පංච රථ උතුරු-දකුණු දිශානතිය ඔස්සේ පිහිටා ඇති අතර, පොදු පාදමක පිහිටා ඇත.මේවාට පෙර කිසිදු මේ ආකාරයේ ගෘහනිර්මාණ ක්‍රමවේදයක් දක්නට නොලැබෙන අතර, ඒවා පසුකාලීන විශාල දකුණු ඉන්දියානු ද්‍රවිඩියානු දේවාල ගෘහනිර්මාණ සඳහා "මූලාදර්ශ" වන්නට ඇතැයි විශ්වාස කෙරේ.
  සතර දේවාලයේ කප් සිටුවයි.	දිය කපන දිනයේ උගුල්ලා ගඟට විසි කරන්නේ මෙම කපයි.පාන්දර හතරට පමණ කප් සිටුවන අතර අලුත් නුවර සිට කප ගෙන ඒමද සිරිතකි.	හය වන දවසේ ඇරඹෙන්නේ කුඹල් පෙරහැරයි.
  විශාල වළාකුලක් එයට සමාන බූ සීමා විශාලත්වයකින් යුතු ඉතා නොගැඹුරු වතුර වලක ඇති තරම් ජලය ඇත.	මෙම විද්‍යාවේ වෛද්‍ය විද්‍යාත්මක අතින් වැදගත් වන්නේ වාතය හරහා බෝවන රෝග පිළිබඳ අධ්‍යයනයයි.	ඉන් පසුව ජූලි 27 දින ප්‍රාණ රහිත ඔහුගේ දේහය ඇඳ අසල තිබෙනු උපස්ථායකයා විසින් දක්නා ලදී.

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false,
      "directions": [
          "query_to_doc"
      ],
      "partition_mode": "joint",
      "hardness_mode": null,
      "hardness_strength": 0.0
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 16
• num_train_epochs: 1
• fp16: True
• per_device_eval_batch_size: 16
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• per_device_train_batch_size: 16
• num_train_epochs: 1
• max_steps: -1
• learning_rate: 5e-05
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_steps: 0
• optim: adamw_torch
• optim_args: None
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• optim_target_modules: None
• gradient_accumulation_steps: 1
• average_tokens_across_devices: True
• max_grad_norm: 1
• label_smoothing_factor: 0.0
• bf16: False
• fp16: True
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• use_liger_kernel: False
• liger_kernel_config: None
• use_cache: False
• neftune_noise_alpha: None
• torch_empty_cache_steps: None
• auto_find_batch_size: False
• log_on_each_node: True
• logging_nan_inf_filter: True
• include_num_input_tokens_seen: no
• log_level: passive
• log_level_replica: warning
• disable_tqdm: False
• project: huggingface
• trackio_space_id: None
• trackio_bucket_id: None
• trackio_static_space_id: None
• per_device_eval_batch_size: 16
• prediction_loss_only: True
• eval_on_start: False
• eval_do_concat_batches: True
• eval_use_gather_object: False
• eval_accumulation_steps: None
• include_for_metrics: []
• batch_eval_metrics: False
• save_only_model: False
• save_on_each_node: False
• enable_jit_checkpoint: False
• push_to_hub: False
• hub_private_repo: None
• hub_model_id: None
• hub_strategy: every_save
• hub_always_push: False
• hub_revision: None
• load_best_model_at_end: False
• ignore_data_skip: False
• restore_callback_states_from_checkpoint: False
• full_determinism: False
• seed: 42
• data_seed: None
• use_cpu: False
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• parallelism_config: None
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• dataloader_prefetch_factor: None
• remove_unused_columns: True
• label_names: None
• train_sampling_strategy: random
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• ddp_static_graph: None
• ddp_backend: None
• ddp_timeout: 1800
• fsdp: None
• fsdp_config: None
• deepspeed: None
• debug: []
• skip_memory_metrics: True
• do_predict: False
• resume_from_checkpoint: None
• warmup_ratio: None
• local_rank: -1
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Click to expand

Epoch	Step	Training Loss
0.0305	500	1.5952
0.0609	1000	1.1837
0.0914	1500	1.0741
0.1218	2000	1.0447
0.1523	2500	1.0214
0.1827	3000	0.9567
0.2132	3500	0.9446
0.2436	4000	0.9445
0.2741	4500	0.9084
0.3046	5000	0.9049
0.3350	5500	0.8879
0.3655	6000	0.8746
0.3959	6500	0.8342
0.4264	7000	0.8318
0.4568	7500	0.8514
0.4873	8000	0.8166
0.5178	8500	0.8046
0.5482	9000	0.8257
0.5787	9500	0.8034
0.6091	10000	0.7874
0.6396	10500	0.7786
0.6700	11000	0.7563
0.7005	11500	0.7778
0.7309	12000	0.7434
0.7614	12500	0.7495
0.7919	13000	0.7639
0.8223	13500	0.7398
0.8528	14000	0.7493
0.8832	14500	0.7284
0.9137	15000	0.7317
0.9441	15500	0.7351
0.9746	16000	0.7062
0.0305	500	0.6713
0.0609	1000	0.6940
0.0914	1500	0.6673
0.1218	2000	0.6850
0.1523	2500	0.7092
0.1827	3000	0.6829
0.2132	3500	0.6903
0.2436	4000	0.6828
0.2741	4500	0.6641
0.3046	5000	0.6574
0.3350	5500	0.6672
0.3655	6000	0.6652
0.3959	6500	0.6573
0.4264	7000	0.6644
0.4568	7500	0.6387
0.4873	8000	0.6370
0.5178	8500	0.6411
0.5482	9000	0.6358
0.5787	9500	0.6207
0.6091	10000	0.6037
0.6396	10500	0.6328
0.6700	11000	0.6017
0.7005	11500	0.6248
0.7309	12000	0.5891
0.7614	12500	0.5956
0.7919	13000	0.5924
0.8223	13500	0.5777
0.8528	14000	0.5891
0.8832	14500	0.5714
0.9137	15000	0.5815
0.9441	15500	0.5812
0.9746	16000	0.5578
0.0305	500	0.1675
0.0609	1000	0.1821
0.0914	1500	0.1734
0.1218	2000	0.1907
0.1523	2500	0.2056
0.1827	3000	0.1925
0.2132	3500	0.2149
0.2436	4000	0.2120
0.2741	4500	0.2132
0.3046	5000	0.2159
0.3350	5500	0.2291
0.3655	6000	0.2418
0.3959	6500	0.2509
0.4264	7000	0.2597
0.4568	7500	0.2718
0.4873	8000	0.2700
0.5178	8500	0.2905
0.5482	9000	0.3004
0.5787	9500	0.3076
0.6091	10000	0.3025
0.6396	10500	0.3435
0.6700	11000	0.3351
0.7005	11500	0.3889
0.7309	12000	0.3750
0.7614	12500	0.3977
0.7919	13000	0.4149
0.8223	13500	0.4214
0.8528	14000	0.4504
0.8832	14500	0.4687
0.9137	15000	0.4966
0.9441	15500	0.5326
0.9746	16000	0.5350
0.0305	500	0.0357
0.0609	1000	0.0421
0.0914	1500	0.0443
0.1218	2000	0.0545
0.1523	2500	0.0527
0.1827	3000	0.0509
0.2132	3500	0.0579
0.2436	4000	0.0543
0.2741	4500	0.0620
0.3046	5000	0.0640
0.3350	5500	0.0656
0.3655	6000	0.0689
0.3959	6500	0.0721
0.4264	7000	0.0798
0.4568	7500	0.0835
0.4873	8000	0.0870
0.5178	8500	0.0944
0.5482	9000	0.1082
0.5787	9500	0.1137
0.6091	10000	0.1115
0.6396	10500	0.1397
0.6700	11000	0.1484
0.7005	11500	0.1895
0.7309	12000	0.1975
0.7614	12500	0.2279
0.7919	13000	0.2547
0.8223	13500	0.2777
0.8528	14000	0.3216
0.8832	14500	0.3612
0.9137	15000	0.4126
0.9441	15500	0.4811
0.9746	16000	0.5160
0.0305	500	0.0112
0.0609	1000	0.0173
0.0914	1500	0.0166
0.1218	2000	0.0204
0.1523	2500	0.0221
0.1827	3000	0.0197
0.2132	3500	0.0231
0.2436	4000	0.0218
0.2741	4500	0.0234
0.3046	5000	0.0245
0.3350	5500	0.0237
0.3655	6000	0.0255
0.3959	6500	0.0262
0.4264	7000	0.0298
0.4568	7500	0.0337
0.4873	8000	0.0332
0.5178	8500	0.0334
0.5482	9000	0.0406
0.5787	9500	0.0451
0.6091	10000	0.0445
0.6396	10500	0.0568
0.6700	11000	0.0589
0.7005	11500	0.0869
0.7309	12000	0.0951
0.7614	12500	0.1140
0.7919	13000	0.1461
0.8223	13500	0.1731
0.8528	14000	0.2140
0.8832	14500	0.2686
0.9137	15000	0.3328
0.9441	15500	0.4245
0.9746	16000	0.5006
0.0305	500	0.0048
0.0609	1000	0.0107
0.0914	1500	0.0111
0.1218	2000	0.0112
0.1523	2500	0.0104
0.1827	3000	0.0111
0.2132	3500	0.0114
0.2436	4000	0.0108
0.2741	4500	0.0107
0.3046	5000	0.0127
0.3350	5500	0.0132
0.3655	6000	0.0132
0.3959	6500	0.0140
0.4264	7000	0.0138
0.4568	7500	0.0164
0.4873	8000	0.0173
0.5178	8500	0.0159
0.5482	9000	0.0197
0.5787	9500	0.0205
0.6091	10000	0.0212
0.6396	10500	0.0261
0.6700	11000	0.0274
0.7005	11500	0.0413
0.7309	12000	0.0509
0.7614	12500	0.0586
0.7919	13000	0.0796
0.8223	13500	0.1056
0.8528	14000	0.1412
0.8832	14500	0.1975
0.9137	15000	0.2672
0.9441	15500	0.3724
0.9746	16000	0.4738
0.0305	500	0.0024
0.0609	1000	0.0040
0.0914	1500	0.0034
0.1218	2000	0.0060
0.1523	2500	0.0082
0.1827	3000	0.0075
0.2132	3500	0.0074
0.2436	4000	0.0069
0.2741	4500	0.0082
0.3046	5000	0.0076
0.3350	5500	0.0100
0.3655	6000	0.0084
0.3959	6500	0.0093
0.4264	7000	0.0097
0.4568	7500	0.0104
0.4873	8000	0.0097
0.5178	8500	0.0100
0.5482	9000	0.0119
0.5787	9500	0.0122
0.6091	10000	0.0138
0.6396	10500	0.0155
0.6700	11000	0.0186
0.7005	11500	0.0242
0.7309	12000	0.0276
0.7614	12500	0.0359
0.7919	13000	0.0489
0.8223	13500	0.0683
0.8528	14000	0.0981
0.8832	14500	0.1450
0.9137	15000	0.2173
0.9441	15500	0.3247
0.9746	16000	0.4529
0.0305	500	0.0025
0.0609	1000	0.0042
0.0914	1500	0.0038
0.1218	2000	0.0057
0.1523	2500	0.0068
0.1827	3000	0.0052
0.2132	3500	0.0058
0.2436	4000	0.0056
0.2741	4500	0.0058
0.3046	5000	0.0047
0.3350	5500	0.0056
0.3655	6000	0.0065
0.3959	6500	0.0069
0.4264	7000	0.0054
0.4568	7500	0.0060
0.4873	8000	0.0065
0.5178	8500	0.0063
0.5482	9000	0.0077
0.5787	9500	0.0081
0.6091	10000	0.0077
0.6396	10500	0.0098
0.6700	11000	0.0116
0.7005	11500	0.0149
0.7309	12000	0.0177
0.7614	12500	0.0188
0.7919	13000	0.0301
0.8223	13500	0.0415
0.8528	14000	0.0631
0.8832	14500	0.1010
0.9137	15000	0.1621
0.9441	15500	0.2720
0.9746	16000	0.4265

Training Time

• Training: 56.5 minutes

Evaluation

Evaluated on a held-out split of topically-coherent sentence pairs (positives) against paragraph-boundary hard negatives, used as the coherence signal in akshara-kit's neuro-symbolic chunker:

	AUC	mean sim, coherent	mean sim, hard-negative
Base LaBSE	0.7269	0.3871	0.2746
Fine-tuned (this model)	0.8832	0.5313	0.1620

Bootstrap 95% CI on the AUC improvement over base LaBSE: [+0.1419, +0.1694].

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 5.6.1
• Transformers: 5.14.1
• PyTorch: 2.5.1+cu121
• Accelerate: 1.14.0
• Datasets: 5.0.1
• Tokenizers: 0.22.2

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{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}