panagoa/LaBSE-kbd-v0.2

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("panagoa/LaBSE-kbd-v0.2")
# Run inference
sentences = [
    "We've never seen Tom this angry before.",
    "Tom'u daha önce asla bu kadar öfkeli görmedik.",
    'Soyez attentive aux voleurs à la tire.',
]
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: validation
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	-0.28
spearman_cosine	-0.3212

Training Details

Training Dataset

Unnamed Dataset

• Size: 3,395,988 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: 5 tokens mean: 10.33 tokens max: 50 tokens	min: 5 tokens mean: 13.81 tokens max: 46 tokens	min: 0.0 mean: 0.36 max: 0.98

• Samples:

  sentence_0	sentence_1	label
  Почему вас это удивило?	Сыт ар щIывгъэщIэгъуар?	0.9298050403594972
  Ребёнка кто-нибудь видел?	Quelqu'un a-t-il vu l'enfant ?	0.0
  Marie se couchait.	Мэри гъуэлъырт.	0.9330472946166992

• 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: 16
• per_device_eval_batch_size: 16
• num_train_epochs: 2
• multi_dataset_batch_sampler: round_robin

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: 2
• 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
• 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: batch_sampler
• multi_dataset_batch_sampler: round_robin

Training Logs

Click to expand

Epoch	Step	Training Loss	validation_spearman_cosine
0.0005	100	-	-0.7761
0.0009	200	-	-0.7598
0.0014	300	-	-0.7485
0.0019	400	-	-0.7412
0.0024	500	0.2864	-0.7354
0.0028	600	-	-0.7307
0.0033	700	-	-0.7191
0.0038	800	-	-0.7206
0.0042	900	-	-0.7197
0.0047	1000	0.0463	-0.7037
0.0052	1100	-	-0.6866
0.0057	1200	-	-0.6798
0.0061	1300	-	-0.6844
0.0066	1400	-	-0.6716
0.0071	1500	0.0184	-0.6658
0.0075	1600	-	-0.6620
0.0080	1700	-	-0.6532
0.0085	1800	-	-0.6455
0.0090	1900	-	-0.6452
0.0094	2000	0.011	-0.6360
0.0099	2100	-	-0.6240
0.0104	2200	-	-0.6220
0.0108	2300	-	-0.6294
0.0113	2400	-	-0.6038
0.0118	2500	0.0092	-0.6116
0.0122	2600	-	-0.5996
0.0127	2700	-	-0.6120
0.0132	2800	-	-0.5940
0.0137	2900	-	-0.5848
0.0141	3000	0.0071	-0.5958
0.0146	3100	-	-0.5840
0.0151	3200	-	-0.5944
0.0155	3300	-	-0.5895
0.0160	3400	-	-0.5849
0.0165	3500	0.0056	-0.5708
0.0005	100	-	-0.5686
0.0009	200	-	-0.5608
0.0014	300	-	-0.5587
0.0024	500	0.0053	-
0.0047	1000	0.0081	-0.5882
0.0071	1500	0.0058	-
0.0094	2000	0.0064	-0.5127
0.0118	2500	0.004	-
0.0141	3000	0.0042	-0.4934
0.0165	3500	0.0048	-
0.0188	4000	0.0036	-0.4762
0.0212	4500	0.0051	-
0.0236	5000	0.0054	-0.4754
0.0259	5500	0.0054	-
0.0283	6000	0.0054	-0.4609
0.0306	6500	0.0044	-
0.0330	7000	0.0048	-0.4716
0.0353	7500	0.0061	-
0.0377	8000	0.0018	-0.4293
0.0400	8500	0.0047	-
0.0424	9000	0.0043	-0.4311
0.0448	9500	0.0034	-
0.0471	10000	0.0041	-0.4429
0.0495	10500	0.0028	-
0.0518	11000	0.0032	-0.4324
0.0542	11500	0.0025	-
0.0565	12000	0.0037	-0.4374
0.0589	12500	0.003	-
0.0612	13000	0.005	-0.4522
0.0636	13500	0.0051	-
0.0660	14000	0.0048	-0.3994
0.0683	14500	0.0034	-
0.0707	15000	0.0032	-0.4148
0.0730	15500	0.0046	-
0.0754	16000	0.0026	-0.3848
0.0777	16500	0.0036	-
0.0801	17000	0.0051	-0.3845
0.0824	17500	0.0031	-
0.0848	18000	0.0035	-0.3500
0.0872	18500	0.0028	-
0.0895	19000	0.0021	-0.3634
0.0919	19500	0.0025	-
0.0942	20000	0.0023	-0.3428
0.0966	20500	0.0042	-
0.0989	21000	0.0038	-0.3432
0.1013	21500	0.005	-
0.1037	22000	0.0024	-0.3515
0.1060	22500	0.0029	-
0.1084	23000	0.0033	-0.3929
0.1107	23500	0.003	-
0.1131	24000	0.0029	-0.3309
0.1154	24500	0.0038	-
0.1178	25000	0.0028	-0.3369
0.1201	25500	0.0025	-
0.1225	26000	0.002	-0.3257
0.1249	26500	0.0025	-
0.1272	27000	0.0033	-0.3659
0.1296	27500	0.0023	-
0.1319	28000	0.0031	-0.3208
0.1343	28500	0.0027	-
0.1366	29000	0.0031	-0.3298
0.1390	29500	0.0047	-
0.1413	30000	0.003	-0.3460
0.1437	30500	0.004	-
0.1461	31000	0.0027	-0.3567
0.1484	31500	0.0063	-
0.1508	32000	0.003	-0.3382
0.1531	32500	0.0022	-
0.1555	33000	0.0048	-0.3475
0.1578	33500	0.0021	-
0.1602	34000	0.0043	-0.3323
0.1625	34500	0.0031	-
0.1649	35000	0.0024	-0.3207
0.1673	35500	0.0029	-
0.1696	36000	0.0032	-0.3004
0.1720	36500	0.0046	-
0.1743	37000	0.0033	-0.3085
0.1767	37500	0.002	-
0.1790	38000	0.0022	-0.3270
0.1814	38500	0.0036	-
0.1837	39000	0.0034	-0.3042
0.1861	39500	0.0034	-
0.1885	40000	0.0016	-0.3193
0.1908	40500	0.0026	-
0.1932	41000	0.0028	-0.2945
0.1955	41500	0.0031	-
0.1979	42000	0.0016	-0.2942
0.2002	42500	0.0021	-
0.2026	43000	0.003	-0.2998
0.2049	43500	0.0042	-
0.2073	44000	0.0023	-0.3245
0.2097	44500	0.0018	-
0.2120	45000	0.0021	-0.3212

Framework Versions

• Python: 3.11.11
• Sentence Transformers: 3.4.1
• Transformers: 4.48.3
• PyTorch: 2.5.1+cu124
• Accelerate: 1.3.0
• Datasets: 3.3.2
• 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}
}