luanafelbarros/bert-es-pt-cased-matryoshka

SentenceTransformer based on google-bert/bert-base-multilingual-cased

This is a sentence-transformers model finetuned from google-bert/bert-base-multilingual-cased on the en-es, en-pt and en-pt-br datasets. 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: google-bert/bert-base-multilingual-cased
• Maximum Sequence Length: 128 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity
• Training Datasets:
  • en-es
  • en-pt
  • en-pt-br
• Languages: en, multilingual, ar, bg, ca, cs, da, de, el, es, et, fa, fi, fr, gl, gu, he, hi, hr, hu, hy, id, it, ja, ka, ko, ku, lt, lv, mk, mn, mr, ms, my, nb, nl, pl, pt, ro, ru, sk, sl, sq, sr, sv, th, tr, uk, ur, vi, zh

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': 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("luanafelbarros/bert-es-pt-cased-matryoshka")
# Run inference
sentences = [
    "It's hands-on, it's in-your-face, it requires an active engagement, and it allows kids to apply all the core subject learning in real ways.",
    'É prático, é presencial, isso requer uma participação ativa, e permite que as crianças apliquem todos os tópicos importantes de aprendizagem de forma real.',
    'Vamos encarar a realidade, o contrato de uma grande marca multinacional para um fornecedor na Índia ou China tem um poder persuasivo muito maior do que as leis locais de trabalho, do que as regras ambientais locais, do que os padrões locais de Direitos Humanos.',
]
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

Knowledge Distillation

• Datasets: MSE-val-en-es, MSE-val-en-pt and MSE-val-en-pt-br
• Evaluated with MSEEvaluator

Metric	MSE-val-en-es	MSE-val-en-pt	MSE-val-en-pt-br
negative_mse	-31.555	-31.7247	-30.2442

Training Details

Training Datasets

en-es

• Dataset: en-es
• Size: 1,612,538 training samples
• Columns: english, non_english, and label
• Approximate statistics based on the first 1000 samples:

  	english	non_english	label
  type	string	string	list
  details	min: 4 tokens mean: 25.46 tokens max: 128 tokens	min: 4 tokens mean: 26.67 tokens max: 128 tokens	size: 768 elements

• Samples:

  english	non_english	label
  And then there are certain conceptual things that can also benefit from hand calculating, but I think they're relatively small in number.	Y luego hay ciertas aspectos conceptuales que pueden beneficiarse del cálculo a mano pero creo que son relativamente pocos.	[-0.015244179405272007, 0.04601434990763664, -0.052873335778713226, 0.03535117208957672, -0.039562877267599106, ...]
  One thing I often ask about is ancient Greek and how this relates.	Algo que pregunto a menudo es sobre el griego antiguo y cómo se relaciona.	[0.0012022971641272306, -0.009590390138328075, -0.032977133989334106, 0.017047710716724396, -0.0028919472824782133, ...]
  See, the thing we're doing right now is we're forcing people to learn mathematics.	Vean, lo que estamos haciendo ahora es forzar a la gente a aprender matemáticas.	[-0.019420800730586052, 0.10435999929904938, 0.009455346502363682, -0.02814250998198986, -0.017036104574799538, ...]

• Loss: main.ModifiedMatryoshkaLoss with these parameters:

  {
      "loss": "MSELoss",
      "matryoshka_dims": [
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

en-pt

• Dataset: en-pt
• Size: 1,542,353 training samples
• Columns: english, non_english, and label
• Approximate statistics based on the first 1000 samples:

  	english	non_english	label
  type	string	string	list
  details	min: 5 tokens mean: 24.95 tokens max: 128 tokens	min: 5 tokens mean: 27.08 tokens max: 128 tokens	size: 768 elements

• Samples:

  english	non_english	label
  And the country that does this first will, in my view, leapfrog others in achieving a new economy even, an improved economy, an improved outlook.	E o país que fizer isto primeiro vai, na minha opinião, ultrapassar outros em alcançar uma nova economia até uma economia melhorada, uma visão melhorada.	[-0.016568265855312347, 0.10754051059484482, -0.025950804352760315, -0.045048732310533524, 0.01812679134309292, ...]
  In fact, I even talk about us moving from what we often call now the "knowledge economy" to what we might call a "computational knowledge economy," where high-level math is integral to what everyone does in the way that knowledge currently is.	De facto, eu até falo de mudarmos do que chamamos hoje a economia do conhecimento para o que poderemos chamar a economia do conhecimento computacional, onde a matemática de alto nível está integrada no que toda a gente faz da forma que o conhecimento actualmente está.	[-0.014394757337868214, 0.11997982114553452, -0.041491635143756866, -0.024539340287446976, 0.01425645500421524, ...]
  We can engage so many more students with this, and they can have a better time doing it.	Podemos cativar tantos mais estudantes com isto, e eles podem divertir-se mais a fazê-lo.	[-0.034232210367918015, 0.04277702793478966, -0.05683526396751404, -0.006559622474014759, -0.00639274762943387, ...]

• Loss: main.ModifiedMatryoshkaLoss with these parameters:

  {
      "loss": "MSELoss",
      "matryoshka_dims": [
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

en-pt-br

• Dataset: en-pt-br at 0c70bc6
• Size: 405,807 training samples
• Columns: english, non_english, and label
• Approximate statistics based on the first 1000 samples:

  	english	non_english	label
  type	string	string	list
  details	min: 4 tokens mean: 25.39 tokens max: 128 tokens	min: 5 tokens mean: 27.52 tokens max: 128 tokens	size: 768 elements

• Samples:

  english	non_english	label
  And then there are certain conceptual things that can also benefit from hand calculating, but I think they're relatively small in number.	E também existem alguns aspectos conceituais que também podem se beneficiar do cálculo manual, mas eu acho que eles são relativamente poucos.	[-0.015244179405272007, 0.04601434990763664, -0.052873335778713226, 0.03535117208957672, -0.039562877267599106, ...]
  One thing I often ask about is ancient Greek and how this relates.	Uma coisa sobre a qual eu pergunto com frequencia é grego antigo e como ele se relaciona a isto.	[0.0012022971641272306, -0.009590390138328075, -0.032977133989334106, 0.017047710716724396, -0.0028919472824782133, ...]
  See, the thing we're doing right now is we're forcing people to learn mathematics.	Vejam, o que estamos fazendo agora, é que estamos forçando as pessoas a aprender matemática.	[-0.019420800730586052, 0.10435999929904938, 0.009455346502363682, -0.02814250998198986, -0.017036104574799538, ...]

• Loss: main.ModifiedMatryoshkaLoss with these parameters:

  {
      "loss": "MSELoss",
      "matryoshka_dims": [
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Evaluation Datasets

en-es

• Dataset: en-es
• Size: 2,990 evaluation samples
• Columns: english, non_english, and label
• Approximate statistics based on the first 1000 samples:

  	english	non_english	label
  type	string	string	list
  details	min: 4 tokens mean: 25.68 tokens max: 128 tokens	min: 4 tokens mean: 27.31 tokens max: 128 tokens	size: 768 elements

• Samples:

  english	non_english	label
  Thank you so much, Chris.	Muchas gracias Chris.	[-0.061677999794483185, -0.04450423642992973, -0.0325058177113533, -0.06641444563865662, 0.003981702029705048, ...]
  And it's truly a great honor to have the opportunity to come to this stage twice; I'm extremely grateful.	Y es en verdad un gran honor tener la oportunidad de venir a este escenario por segunda vez. Estoy extremadamente agradecido.	[0.011398610658943653, -0.02500406838953495, -0.009884772822260857, 0.009336909279227257, 0.0030828709714114666, ...]
  I have been blown away by this conference, and I want to thank all of you for the many nice comments about what I had to say the other night.	He quedado conmovido por esta conferencia, y deseo agradecer a todos ustedes sus amables comentarios acerca de lo que tenía que decir la otra noche.	[-0.03842132166028023, 0.03635749593377113, -0.02491452544927597, -0.0032229204662144184, 0.0003549510147422552, ...]

• Loss: main.ModifiedMatryoshkaLoss with these parameters:

  {
      "loss": "MSELoss",
      "matryoshka_dims": [
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

en-pt

• Dataset: en-pt
• Size: 2,992 evaluation samples
• Columns: english, non_english, and label
• Approximate statistics based on the first 1000 samples:

  	english	non_english	label
  type	string	string	list
  details	min: 4 tokens mean: 25.05 tokens max: 128 tokens	min: 4 tokens mean: 27.58 tokens max: 128 tokens	size: 768 elements

• Samples:

  english	non_english	label
  Thank you so much, Chris.	Muito obrigado, Chris.	[-0.06167794018983841, -0.04450422152876854, -0.032505810260772705, -0.06641443818807602, 0.0039817155338823795, ...]
  And it's truly a great honor to have the opportunity to come to this stage twice; I'm extremely grateful.	É realmente uma grande honra ter a oportunidade de pisar este palco pela segunda vez. Estou muito agradecido.	[0.011398610658943653, -0.02500406838953495, -0.009884772822260857, 0.009336909279227257, 0.0030828709714114666, ...]
  I have been blown away by this conference, and I want to thank all of you for the many nice comments about what I had to say the other night.	Fiquei muito impressionado com esta conferência e quero agradecer a todos os imensos comentários simpáticos sobre o que eu tinha a dizer naquela noite.	[-0.03842132166028023, 0.03635749593377113, -0.02491452544927597, -0.0032229204662144184, 0.0003549510147422552, ...]

• Loss: main.ModifiedMatryoshkaLoss with these parameters:

  {
      "loss": "MSELoss",
      "matryoshka_dims": [
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

en-pt-br

• Dataset: en-pt-br at 0c70bc6
• Size: 992 evaluation samples
• Columns: english, non_english, and label
• Approximate statistics based on the first 992 samples:

  	english	non_english	label
  type	string	string	list
  details	min: 4 tokens mean: 25.8 tokens max: 128 tokens	min: 4 tokens mean: 28.92 tokens max: 128 tokens	size: 768 elements

• Samples:

  english	non_english	label
  Thank you so much, Chris.	Muito obrigado, Chris.	[-0.0616779662668705, -0.044504180550575256, -0.032505787909030914, -0.06641441583633423, 0.003981734160333872, ...]
  And it's truly a great honor to have the opportunity to come to this stage twice; I'm extremely grateful.	É realmente uma grande honra ter a oportunidade de estar neste palco pela segunda vez. Estou muito agradecido.	[0.011398598551750183, -0.02500401996076107, -0.009884790517389774, 0.009336900897324085, 0.003082842566072941, ...]
  I have been blown away by this conference, and I want to thank all of you for the many nice comments about what I had to say the other night.	Eu fui muito aplaudido por esta conferência e quero agradecer a todos pelos muitos comentários delicados sobre o que eu tinha a dizer naquela noite.	[-0.03842132166028023, 0.03635749593377113, -0.02491452544927597, -0.0032229204662144184, 0.0003549510147422552, ...]

• Loss: main.ModifiedMatryoshkaLoss with these parameters:

  {
      "loss": "MSELoss",
      "matryoshka_dims": [
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 256
• per_device_eval_batch_size: 256
• learning_rate: 2e-05
• num_train_epochs: 1
• 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: 256
• per_device_eval_batch_size: 256
• 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: 2e-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: 1
• 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: False
• 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

Epoch	Step	Training Loss	en-es loss	en-pt loss	en-pt-br loss	MSE-val-en-es_negative_mse	MSE-val-en-pt_negative_mse	MSE-val-en-pt-br_negative_mse
0.0719	1000	0.028	0.0237	0.0237	0.0231	-24.8296	-24.6706	-25.9588
0.1438	2000	0.0227	0.0213	0.0215	0.0208	-26.2546	-26.2964	-25.9444
0.2157	3000	0.0213	0.0203	0.0205	0.0199	-27.7589	-27.8414	-27.1460
0.2876	4000	0.0206	0.0197	0.0199	0.0193	-29.1241	-29.2139	-28.3021
0.3595	5000	0.0201	0.0194	0.0195	0.0190	-30.1292	-30.2692	-29.0747
0.4313	6000	0.0198	0.0190	0.0192	0.0187	-30.3807	-30.4967	-29.3404
0.5032	7000	0.0195	0.0188	0.0190	0.0185	-31.0799	-31.2305	-29.9549
0.5751	8000	0.0193	0.0186	0.0188	0.0183	-31.1297	-31.2883	-30.0050
0.6470	9000	0.0192	0.0185	0.0186	0.0182	-31.2788	-31.4498	-30.0589
0.7189	10000	0.019	0.0184	0.0185	0.0181	-31.3215	-31.4903	-30.0056
0.7908	11000	0.019	0.0183	0.0184	0.0180	-31.4416	-31.6329	-30.1343
0.8627	12000	0.0189	0.0182	0.0184	0.0180	-31.5266	-31.6991	-30.1956
0.9346	13000	0.0188	0.0182	0.0183	0.0179	-31.5550	-31.7247	-30.2442

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.3.1
• Transformers: 4.46.3
• PyTorch: 2.5.1+cu121
• Accelerate: 1.1.1
• Datasets: 3.1.0
• Tokenizers: 0.20.3

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",
}