dariolopez/bge-m3-es-legal-tmp-3

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-3")
# 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	0.5335
cosine_accuracy@3	0.7927
cosine_accuracy@5	0.8476
cosine_accuracy@10	0.8811
cosine_precision@1	0.5335
cosine_precision@3	0.2642
cosine_precision@5	0.1695
cosine_precision@10	0.0881
cosine_recall@1	0.5335
cosine_recall@3	0.7927
cosine_recall@5	0.8476
cosine_recall@10	0.8811
cosine_ndcg@10	0.7187
cosine_mrr@10	0.6652
cosine_map@100	0.6706

Information Retrieval

• Dataset: dim_768
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5366
cosine_accuracy@3	0.7988
cosine_accuracy@5	0.8445
cosine_accuracy@10	0.8872
cosine_precision@1	0.5366
cosine_precision@3	0.2663
cosine_precision@5	0.1689
cosine_precision@10	0.0887
cosine_recall@1	0.5366
cosine_recall@3	0.7988
cosine_recall@5	0.8445
cosine_recall@10	0.8872
cosine_ndcg@10	0.722
cosine_mrr@10	0.6678
cosine_map@100	0.6725

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5396
cosine_accuracy@3	0.7988
cosine_accuracy@5	0.8415
cosine_accuracy@10	0.8841
cosine_precision@1	0.5396
cosine_precision@3	0.2663
cosine_precision@5	0.1683
cosine_precision@10	0.0884
cosine_recall@1	0.5396
cosine_recall@3	0.7988
cosine_recall@5	0.8415
cosine_recall@10	0.8841
cosine_ndcg@10	0.7235
cosine_mrr@10	0.6706
cosine_map@100	0.6753

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5488
cosine_accuracy@3	0.7866
cosine_accuracy@5	0.8201
cosine_accuracy@10	0.878
cosine_precision@1	0.5488
cosine_precision@3	0.2622
cosine_precision@5	0.164
cosine_precision@10	0.0878
cosine_recall@1	0.5488
cosine_recall@3	0.7866
cosine_recall@5	0.8201
cosine_recall@10	0.878
cosine_ndcg@10	0.7222
cosine_mrr@10	0.6713
cosine_map@100	0.6765

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5274
cosine_accuracy@3	0.7713
cosine_accuracy@5	0.8201
cosine_accuracy@10	0.8628
cosine_precision@1	0.5274
cosine_precision@3	0.2571
cosine_precision@5	0.164
cosine_precision@10	0.0863
cosine_recall@1	0.5274
cosine_recall@3	0.7713
cosine_recall@5	0.8201
cosine_recall@10	0.8628
cosine_ndcg@10	0.7052
cosine_mrr@10	0.6535
cosine_map@100	0.6594

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5061
cosine_accuracy@3	0.7378
cosine_accuracy@5	0.8018
cosine_accuracy@10	0.8598
cosine_precision@1	0.5061
cosine_precision@3	0.2459
cosine_precision@5	0.1604
cosine_precision@10	0.086
cosine_recall@1	0.5061
cosine_recall@3	0.7378
cosine_recall@5	0.8018
cosine_recall@10	0.8598
cosine_ndcg@10	0.6884
cosine_mrr@10	0.6329
cosine_map@100	0.6381

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: 32
• 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: 32
• 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
0.8649	10	1.5054	-	-	-	-	-	-	-
0.9514	11	-	0.8399	0.6684	0.6148	0.6574	0.6770	0.5281	0.6691
1.7297	20	1.0127	-	-	-	-	-	-	-
1.9892	23	-	0.5057	0.6757	0.6596	0.6715	0.6738	0.6017	0.6719
2.5946	30	0.5708	-	-	-	-	-	-	-
2.9405	34	-	0.4593	0.6781	0.6551	0.6795	0.6806	0.6165	0.6697
3.4595	40	0.2618	-	-	-	-	-	-	-
3.9784	46	-	0.4122	0.6787	0.6607	0.6842	0.6795	0.6227	0.6793
4.3243	50	0.1079	-	-	-	-	-	-	-
4.9297	57	-	0.3717	0.6827	0.6609	0.6810	0.6868	0.6277	0.6769
5.1892	60	0.0574	-	-	-	-	-	-	-
5.9676	69	-	0.3394	0.6824	0.6493	0.6777	0.6784	0.6344	0.6685
6.0541	70	0.0342	-	-	-	-	-	-	-
6.9189	80	0.0211	0.3379	0.6771	0.6627	0.6764	0.6766	0.6395	0.6723
7.7838	90	0.0136	-	-	-	-	-	-	-
7.9568	92	-	0.3128	0.6790	0.6536	0.6789	0.6782	0.6279	0.6730
8.6486	100	0.0087	-	-	-	-	-	-	-
8.9946	104	-	0.3163	0.6811	0.6542	0.6716	0.6744	0.6413	0.6758
9.5135	110	0.0073	-	-	-	-	-	-	-
9.9459	115	-	0.2937	0.6730	0.6569	0.6735	0.6747	0.6380	0.6710
10.3784	120	0.0049	-	-	-	-	-	-	-
10.9838	127	-	0.2927	0.6701	0.6578	0.6772	0.6724	0.6355	0.6738
11.2432	130	0.0044	-	-	-	-	-	-	-
11.9351	138	-	0.2837	0.6720	0.6558	0.6791	0.6752	0.6376	0.6783
12.1081	140	0.0035	-	-	-	-	-	-	-
12.9730	150	0.0031	0.2897	0.6746	0.6610	0.6708	0.6739	0.6375	0.6769
13.8378	160	0.0027	-	-	-	-	-	-	-
13.9243	161	-	0.2961	0.6733	0.6562	0.6692	0.6704	0.6402	0.6740
14.7027	170	0.0026	-	-	-	-	-	-	-
14.9622	173	-	0.2934	0.6734	0.6557	0.6720	0.6720	0.6368	0.6726
15.5676	180	0.0025	-	-	-	-	-	-	-
16.0	185	-	0.2932	0.6735	0.6561	0.6718	0.6744	0.6414	0.6773
16.4324	190	0.0023	-	-	-	-	-	-	-
16.9514	196	-	0.2912	0.6708	0.6582	0.6761	0.6794	0.6367	0.6753
17.2973	200	0.0021	-	-	-	-	-	-	-
17.9892	208	-	0.2925	0.6726	0.6582	0.6747	0.6773	0.6357	0.6737
18.1622	210	0.0022	-	-	-	-	-	-	-
18.9405	219	-	0.2965	0.6688	0.6563	0.6758	0.6769	0.6372	0.6765
19.0270	220	0.002	-	-	-	-	-	-	-
19.8919	230	0.0019	-	-	-	-	-	-	-
19.9784	231	-	0.3010	0.6697	0.6563	0.6768	0.6775	0.6380	0.6730
20.7568	240	0.0018	-	-	-	-	-	-	-
20.9297	242	-	0.3025	0.6728	0.6564	0.6764	0.6757	0.6367	0.6728
21.6216	250	0.0019	-	-	-	-	-	-	-
21.9676	254	-	0.3043	0.6707	0.6533	0.6733	0.6750	0.6352	0.6729
22.4865	260	0.0018	-	-	-	-	-	-	-
22.9189	265	-	0.3029	0.6706	0.6554	0.6734	0.6757	0.6355	0.6715
23.3514	270	0.0018	-	-	-	-	-	-	-
23.9568	277	-	0.3046	0.6706	0.6586	0.6733	0.6740	0.6383	0.6731
24.2162	280	0.0018	-	-	-	-	-	-	-
24.9946	289	-	0.3045	0.6722	0.6553	0.6740	0.6752	0.6364	0.6735
25.0811	290	0.0016	-	-	-	-	-	-	-
25.9459	300	0.0017	0.3061	0.6703	0.6564	0.6770	0.6736	0.6371	0.6724
26.8108	310	0.0016	-	-	-	-	-	-	-
26.9838	312	-	0.3023	0.6694	0.6581	0.6790	0.6771	0.6375	0.6731
27.6757	320	0.0015	-	-	-	-	-	-	-
27.9351	323	-	0.3035	0.6701	0.6585	0.6748	0.6787	0.6366	0.6729
28.5405	330	0.0016	-	-	-	-	-	-	-
28.9730	335	-	0.3017	0.6686	0.6568	0.6748	0.6710	0.6357	0.6713
29.4054	340	0.0016	-	-	-	-	-	-	-
29.9243	346	-	0.3043	0.6683	0.6549	0.6722	0.6762	0.6367	0.6712
30.2703	350	0.0017	-	-	-	-	-	-	-
30.4432	352	-	0.3056	0.6706	0.6594	0.6765	0.6753	0.6381	0.6725

• 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}
}