codersan/FaLaBSE-v6

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 semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: sentence-transformers/LaBSE
• Maximum Sequence Length: 256 tokens
• Output Dimensionality: 768 dimensions
• 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': 256, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Dense({'in_features': 768, 'out_features': 768, 'bias': True, 'activation_function': 'torch.nn.modules.activation.Tanh'})
  (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("codersan/FaLaBSE-v6")
# Run inference
sentences = [
    'آیا با دختری که باکره نیست ازدواج خواهید کرد؟',
    'آیا با کسی که باکره نیست ازدواج می کنید؟',
    'زنی با شلوار جین کنار اسبی با زین ایستاده است',
]
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]

Training Details

Training Dataset

Unnamed Dataset

• Size: 149,098 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 5 tokens mean: 15.1 tokens max: 76 tokens	min: 5 tokens mean: 14.54 tokens max: 57 tokens

• Samples:

  anchor	positive
  اگر هند تقسیم نشده بود ، هند امروز چگونه به نظر می رسد؟	اگر پارتیشن اتفاق نیفتاد ، هند امروز چگونه خواهد بود؟
  چگونه می توانم وارد امنیت اینترنت شوم؟	چگونه می توانم شروع به یادگیری امنیت اطلاعات کنم؟
  برخی از بهترین مؤسسات مربیگری GMAT در دهلی/NCR چیست؟	بهترین مؤسسات مربیگری برای GMAT در NCR چیست؟

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 32
• learning_rate: 3e-05
• weight_decay: 0.15
• num_train_epochs: 10
• warmup_ratio: 0.15
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: no
• prediction_loss_only: True
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 8
• 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: 3e-05
• weight_decay: 0.15
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 10
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.15
• 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: 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, '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: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss
0.0429	100	0.1219
0.0858	200	0.0626
0.1288	300	0.0489
0.1717	400	0.0414
0.2146	500	0.0432
0.2575	600	0.0419
0.3004	700	0.0313
0.3433	800	0.0339
0.3863	900	0.0317
0.4292	1000	0.035
0.4721	1100	0.0378
0.5150	1200	0.0308
0.5579	1300	0.0305
0.6009	1400	0.0312
0.6438	1500	0.0304
0.6867	1600	0.0295
0.7296	1700	0.0301
0.7725	1800	0.033
0.8155	1900	0.0263
0.8584	2000	0.0276
0.9013	2100	0.0236
0.9442	2200	0.0276
0.9871	2300	0.0278
1.0300	2400	0.0309
1.0730	2500	0.0269
1.1159	2600	0.0299
1.1588	2700	0.0272
1.2017	2800	0.029
1.2446	2900	0.0309
1.2876	3000	0.0247
1.3305	3100	0.0244
1.3734	3200	0.0261
1.4163	3300	0.0254
1.4592	3400	0.0273
1.5021	3500	0.0298
1.5451	3600	0.0225
1.5880	3700	0.0278
1.6309	3800	0.027
1.6738	3900	0.0218
1.7167	4000	0.0247
1.7597	4100	0.023
1.8026	4200	0.0225
1.8455	4300	0.0191
1.8884	4400	0.0174
1.9313	4500	0.0214
1.9742	4600	0.018
2.0172	4700	0.0227
2.0601	4800	0.0222
2.1030	4900	0.0211
2.1459	5000	0.0204
2.1888	5100	0.0215
2.2318	5200	0.0206
2.2747	5300	0.0213
2.3176	5400	0.0168
2.3605	5500	0.0189
2.4034	5600	0.0206
2.4464	5700	0.0194
2.4893	5800	0.0182
2.5322	5900	0.017
2.5751	6000	0.0186
2.6180	6100	0.017
2.6609	6200	0.0152
2.7039	6300	0.0164
2.7468	6400	0.0142
2.7897	6500	0.0162
2.8326	6600	0.0123
2.8755	6700	0.0162
2.9185	6800	0.0138
2.9614	6900	0.0163
3.0043	7000	0.0138
3.0472	7100	0.0164
3.0901	7200	0.016
3.1330	7300	0.0175
3.1760	7400	0.0143
3.2189	7500	0.0142
3.2618	7600	0.0176
3.3047	7700	0.0147
3.3476	7800	0.0164
3.3906	7900	0.0133
3.4335	8000	0.0168
3.4764	8100	0.0166
3.5193	8200	0.0138
3.5622	8300	0.0126
3.6052	8400	0.0145
3.6481	8500	0.0114
3.6910	8600	0.0137
3.7339	8700	0.014
3.7768	8800	0.0134
3.8197	8900	0.0108
3.8627	9000	0.012
3.9056	9100	0.0102
3.9485	9200	0.0119
3.9914	9300	0.0122
4.0343	9400	0.0116
4.0773	9500	0.0136
4.1202	9600	0.0135
4.1631	9700	0.0108
4.2060	9800	0.0119
4.2489	9900	0.0142
4.2918	10000	0.0111
4.3348	10100	0.0131
4.3777	10200	0.0103
4.4206	10300	0.0124
4.4635	10400	0.0163
4.5064	10500	0.0123
4.5494	10600	0.0112
4.5923	10700	0.01
4.6352	10800	0.0096
4.6781	10900	0.0103
4.7210	11000	0.0102
4.7639	11100	0.0092
4.8069	11200	0.0107
4.8498	11300	0.0114
4.8927	11400	0.0091
4.9356	11500	0.0108
4.9785	11600	0.0092
5.0215	11700	0.0086
5.0644	11800	0.0104
5.1073	11900	0.0123
5.1502	12000	0.009
5.1931	12100	0.0106
5.2361	12200	0.0114
5.2790	12300	0.0098
5.3219	12400	0.0093
5.3648	12500	0.0092
5.4077	12600	0.011
5.4506	12700	0.0113
5.4936	12800	0.0091
5.5365	12900	0.0079
5.5794	13000	0.01
5.6223	13100	0.0067
5.6652	13200	0.0081
5.7082	13300	0.0097
5.7511	13400	0.0081
5.7940	13500	0.0094
5.8369	13600	0.0074
5.8798	13700	0.0071
5.9227	13800	0.0074
5.9657	13900	0.0076
6.0086	14000	0.0063
6.0515	14100	0.0083
6.0944	14200	0.0101
6.1373	14300	0.0084
6.1803	14400	0.0074
6.2232	14500	0.007
6.2661	14600	0.0078
6.3090	14700	0.0074
6.3519	14800	0.0086
6.3948	14900	0.0069
6.4378	15000	0.0083
6.4807	15100	0.0082
6.5236	15200	0.0066
6.5665	15300	0.0086
6.6094	15400	0.0059
6.6524	15500	0.0052
6.6953	15600	0.0081
6.7382	15700	0.0054
6.7811	15800	0.0063
6.8240	15900	0.0065
6.8670	16000	0.0068
6.9099	16100	0.0047
6.9528	16200	0.0065
6.9957	16300	0.0064
7.0386	16400	0.0051
7.0815	16500	0.0066
7.1245	16600	0.0069
7.1674	16700	0.0074
7.2103	16800	0.0062
7.2532	16900	0.0071
7.2961	17000	0.005
7.3391	17100	0.008
7.3820	17200	0.0047
7.4249	17300	0.0073
7.4678	17400	0.0078
7.5107	17500	0.0058
7.5536	17600	0.0055
7.5966	17700	0.0049
7.6395	17800	0.0046
7.6824	17900	0.0051
7.7253	18000	0.005
7.7682	18100	0.0059
7.8112	18200	0.0056
7.8541	18300	0.0049
7.8970	18400	0.0038
7.9399	18500	0.005
7.9828	18600	0.005
8.0258	18700	0.0036
8.0687	18800	0.0049
8.1116	18900	0.0067
8.1545	19000	0.0056
8.1974	19100	0.0061
8.2403	19200	0.0054
8.2833	19300	0.0046
8.3262	19400	0.0048
8.3691	19500	0.0052
8.4120	19600	0.0059
8.4549	19700	0.0053
8.4979	19800	0.0049
8.5408	19900	0.0036
8.5837	20000	0.0049
8.6266	20100	0.0033
8.6695	20200	0.0049
8.7124	20300	0.0043
8.7554	20400	0.0039
8.7983	20500	0.0038
8.8412	20600	0.0035
8.8841	20700	0.0041
8.9270	20800	0.0042
8.9700	20900	0.0056
9.0129	21000	0.0031
9.0558	21100	0.004
9.0987	21200	0.0043
9.1416	21300	0.0047
9.1845	21400	0.0051
9.2275	21500	0.0032
9.2704	21600	0.0045
9.3133	21700	0.0038
9.3562	21800	0.0045
9.3991	21900	0.0047
9.4421	22000	0.0048
9.4850	22100	0.0042
9.5279	22200	0.0039
9.5708	22300	0.0042
9.6137	22400	0.003
9.6567	22500	0.0031
9.6996	22600	0.0042
9.7425	22700	0.0028
9.7854	22800	0.0037
9.8283	22900	0.0035
9.8712	23000	0.0033
9.9142	23100	0.0029
9.9571	23200	0.0048
10.0	23300	0.0039

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.3.1
• Transformers: 4.47.0
• PyTorch: 2.5.1+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",
}

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}
}