dariolopez/bge-m3-es-legal-tmp-5

BGE large Legal Spanish

This is a sentence-transformers model finetuned from BAAI/bge-m3. It maps sentences & paragraphs to a 1024-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: BAAI/bge-m3
• Maximum Sequence Length: 8192 tokens
• Output Dimensionality: 1024 tokens
• Similarity Function: Cosine Similarity
• Language: es
• License: apache-2.0

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': 8192, 'do_lower_case': False}) with Transformer model: XLMRobertaModel 
  (1): Pooling({'word_embedding_dimension': 1024, '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): 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("dariolopez/bge-m3-es-legal-tmp-5")
# Run inference
sentences = [
    'Artículo 6. Definiciones. 1. Discriminación directa e indirecta. b) La discriminación indirecta se produce cuando una disposición, criterio o práctica aparentemente neutros ocasiona o puede ocasionar a una o varias personas una desventaja particular con respecto a otras por razón de las causas previstas en el apartado 1 del artículo 2.',
    '¿Qué se considera discriminación indirecta?',
    '¿Qué tipo de información se considera veraz?',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

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

Evaluation

Metrics

Information Retrieval

• Dataset: dim_1024
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5457
cosine_accuracy@3	0.7957
cosine_accuracy@5	0.8384
cosine_accuracy@10	0.8933
cosine_precision@1	0.5457
cosine_precision@3	0.2652
cosine_precision@5	0.1677
cosine_precision@10	0.0893
cosine_recall@1	0.5457
cosine_recall@3	0.7957
cosine_recall@5	0.8384
cosine_recall@10	0.8933
cosine_ndcg@10	0.7303
cosine_mrr@10	0.6768
cosine_map@100	0.6813

Information Retrieval

• Dataset: dim_768
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5366
cosine_accuracy@3	0.8079
cosine_accuracy@5	0.8415
cosine_accuracy@10	0.8933
cosine_precision@1	0.5366
cosine_precision@3	0.2693
cosine_precision@5	0.1683
cosine_precision@10	0.0893
cosine_recall@1	0.5366
cosine_recall@3	0.8079
cosine_recall@5	0.8415
cosine_recall@10	0.8933
cosine_ndcg@10	0.7282
cosine_mrr@10	0.6737
cosine_map@100	0.6781

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5518
cosine_accuracy@3	0.8079
cosine_accuracy@5	0.8476
cosine_accuracy@10	0.8902
cosine_precision@1	0.5518
cosine_precision@3	0.2693
cosine_precision@5	0.1695
cosine_precision@10	0.089
cosine_recall@1	0.5518
cosine_recall@3	0.8079
cosine_recall@5	0.8476
cosine_recall@10	0.8902
cosine_ndcg@10	0.7326
cosine_mrr@10	0.6805
cosine_map@100	0.6848

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5518
cosine_accuracy@3	0.7835
cosine_accuracy@5	0.8384
cosine_accuracy@10	0.8841
cosine_precision@1	0.5518
cosine_precision@3	0.2612
cosine_precision@5	0.1677
cosine_precision@10	0.0884
cosine_recall@1	0.5518
cosine_recall@3	0.7835
cosine_recall@5	0.8384
cosine_recall@10	0.8841
cosine_ndcg@10	0.7255
cosine_mrr@10	0.6738
cosine_map@100	0.6784

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5244
cosine_accuracy@3	0.7683
cosine_accuracy@5	0.8201
cosine_accuracy@10	0.878
cosine_precision@1	0.5244
cosine_precision@3	0.2561
cosine_precision@5	0.164
cosine_precision@10	0.0878
cosine_recall@1	0.5244
cosine_recall@3	0.7683
cosine_recall@5	0.8201
cosine_recall@10	0.878
cosine_ndcg@10	0.709
cosine_mrr@10	0.6541
cosine_map@100	0.6583

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.503
cosine_accuracy@3	0.7256
cosine_accuracy@5	0.7896
cosine_accuracy@10	0.8567
cosine_precision@1	0.503
cosine_precision@3	0.2419
cosine_precision@5	0.1579
cosine_precision@10	0.0857
cosine_recall@1	0.503
cosine_recall@3	0.7256
cosine_recall@5	0.7896
cosine_recall@10	0.8567
cosine_ndcg@10	0.6822
cosine_mrr@10	0.626
cosine_map@100	0.631

Training Details

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• gradient_accumulation_steps: 16
• learning_rate: 2e-05
• num_train_epochs: 8
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• bf16: True
• tf32: True
• load_best_model_at_end: True
• optim: adamw_torch_fused
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: epoch
• 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: 16
• eval_accumulation_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: 8
• max_steps: -1
• lr_scheduler_type: cosine
• 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: True
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: True
• 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: 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_fused
• 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
• 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
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	loss	dim_1024_cosine_map@100	dim_128_cosine_map@100	dim_256_cosine_map@100	dim_512_cosine_map@100	dim_64_cosine_map@100	dim_768_cosine_map@100
0.4324	5	1.6729	-	-	-	-	-	-	-
0.8649	10	1.0155	-	-	-	-	-	-	-
0.9514	11	-	0.5773	0.6769	0.6526	0.6771	0.6782	0.5960	0.6752
1.2973	15	0.8661	-	-	-	-	-	-	-
1.7297	20	0.4311	-	-	-	-	-	-	-
1.9892	23	-	0.4496	0.6637	0.6494	0.6749	0.6729	0.6203	0.6656
2.1622	25	0.3745	-	-	-	-	-	-	-
2.5946	30	0.19	-	-	-	-	-	-	-
2.9405	34	-	0.4119	0.6714	0.6530	0.6777	0.6753	0.6162	0.6746
3.0270	35	0.1448	-	-	-	-	-	-	-
3.4595	40	0.0926	-	-	-	-	-	-	-
3.8919	45	0.0536	-	-	-	-	-	-	-
3.9784	46	-	0.3744	0.6852	0.6585	0.6778	0.6827	0.6273	0.6811
4.3243	50	0.0583	-	-	-	-	-	-	-
4.7568	55	0.0377	-	-	-	-	-	-	-
4.9297	57	-	0.3594	0.6829	0.6523	0.6786	0.6837	0.6302	0.6772
5.1892	60	0.0401	-	-	-	-	-	-	-
5.6216	65	0.0294	-	-	-	-	-	-	-
5.9676	69	-	0.3519	0.6831	0.6567	0.6774	0.6859	0.6329	0.6800
6.0541	70	0.0288	-	-	-	-	-	-	-
6.4865	75	0.0273	-	-	-	-	-	-	-
6.9189	80	0.0227	0.3513	0.6807	0.6551	0.6757	0.6832	0.6298	0.6781
7.3514	85	0.0223	-	-	-	-	-	-	-
7.6108	88	-	0.3523	0.6813	0.6583	0.6784	0.6848	0.631	0.6781

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.0.1
• Transformers: 4.42.3
• PyTorch: 2.2.0+cu121
• Accelerate: 0.32.1
• Datasets: 2.20.0
• Tokenizers: 0.19.1

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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning}, 
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

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}
}