adriansanz/sqv-v3

SentenceTransformer based on BAAI/bge-m3

This is a sentence-transformers model finetuned from BAAI/bge-m3 on the json dataset. 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
• Training Dataset:
  • json

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("adriansanz/sqv-v3")
# Run inference
sentences = [
    "Permet sol·licitar l’autorització per a l’ús comú especial de la via pública per a reserves temporals d’estacionament i espai públic per: càrrega/descàrrega de materials diversos davant d'una obra;",
    "Quins són els materials que es poden càrregar/descarregar en l'ocupació i reserves temporals amb càrrega/descàrrega de materials?",
    'Quin és el tràmit per canviar el domicili del permís de conducció i del permís de circulació?',
]
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	0.1957
cosine_accuracy@3	0.5435
cosine_accuracy@5	0.6739
cosine_accuracy@10	0.7717
cosine_precision@1	0.1957
cosine_precision@3	0.1812
cosine_precision@5	0.1348
cosine_precision@10	0.0772
cosine_recall@1	0.1957
cosine_recall@3	0.5435
cosine_recall@5	0.6739
cosine_recall@10	0.7717
cosine_ndcg@10	0.485
cosine_mrr@10	0.3923
cosine_map@100	0.4003

Information Retrieval

• Dataset: dim_768
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.1957
cosine_accuracy@3	0.5543
cosine_accuracy@5	0.6739
cosine_accuracy@10	0.7717
cosine_precision@1	0.1957
cosine_precision@3	0.1848
cosine_precision@5	0.1348
cosine_precision@10	0.0772
cosine_recall@1	0.1957
cosine_recall@3	0.5543
cosine_recall@5	0.6739
cosine_recall@10	0.7717
cosine_ndcg@10	0.488
cosine_mrr@10	0.3962
cosine_map@100	0.4042

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.2065
cosine_accuracy@3	0.5435
cosine_accuracy@5	0.6522
cosine_accuracy@10	0.7609
cosine_precision@1	0.2065
cosine_precision@3	0.1812
cosine_precision@5	0.1304
cosine_precision@10	0.0761
cosine_recall@1	0.2065
cosine_recall@3	0.5435
cosine_recall@5	0.6522
cosine_recall@10	0.7609
cosine_ndcg@10	0.4841
cosine_mrr@10	0.395
cosine_map@100	0.4031

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.1848
cosine_accuracy@3	0.5435
cosine_accuracy@5	0.6522
cosine_accuracy@10	0.75
cosine_precision@1	0.1848
cosine_precision@3	0.1812
cosine_precision@5	0.1304
cosine_precision@10	0.075
cosine_recall@1	0.1848
cosine_recall@3	0.5435
cosine_recall@5	0.6522
cosine_recall@10	0.75
cosine_ndcg@10	0.4702
cosine_mrr@10	0.3799
cosine_map@100	0.3886

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.2283
cosine_accuracy@3	0.5435
cosine_accuracy@5	0.6957
cosine_accuracy@10	0.7826
cosine_precision@1	0.2283
cosine_precision@3	0.1812
cosine_precision@5	0.1391
cosine_precision@10	0.0783
cosine_recall@1	0.2283
cosine_recall@3	0.5435
cosine_recall@5	0.6957
cosine_recall@10	0.7826
cosine_ndcg@10	0.5046
cosine_mrr@10	0.4149
cosine_map@100	0.4207

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.1739
cosine_accuracy@3	0.4891
cosine_accuracy@5	0.663
cosine_accuracy@10	0.7609
cosine_precision@1	0.1739
cosine_precision@3	0.163
cosine_precision@5	0.1326
cosine_precision@10	0.0761
cosine_recall@1	0.1739
cosine_recall@3	0.4891
cosine_recall@5	0.663
cosine_recall@10	0.7609
cosine_ndcg@10	0.4628
cosine_mrr@10	0.3667
cosine_map@100	0.3729

Training Details

Training Dataset

json

• Dataset: json
• Size: 828 training samples
• Columns: positive and anchor
• Approximate statistics based on the first 828 samples:

  	positive	anchor
  type	string	string
  details	min: 5 tokens mean: 41.95 tokens max: 117 tokens	min: 9 tokens mean: 20.81 tokens max: 50 tokens

• Samples:

  positive	anchor
  Consultar l'estat tributari d'un contribuent. Us permet consultar l'estat dels rebuts i liquidacions que estan a nom del contribuent titular d'un certificat electrònic, així com els elements que configuren el càlcul per determinar el deute tributari de cadascun d'ells.	Com puc consultar l'estat tributari d'un contribuent?
  L'informe facultatiu servirà per tramitar una autorització de residència temporal per arrelament social.	Quin és el tràmit relacionat amb la residència a l'Ajuntament?
  Aquesta targeta, és el document que dona dret a persones físiques o jurídiques titulars de vehicles adaptats destinats al transport col·lectiu de persones amb discapacitat...	Quin és el benefici de tenir la targeta d'aparcament de transport col·lectiu per a les persones amb discapacitat?

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          1024,
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

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: 5
• lr_scheduler_type: cosine
• warmup_ratio: 0.2
• 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
• 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: 5
• max_steps: -1
• lr_scheduler_type: cosine
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.2
• 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
• eval_use_gather_object: False
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training 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
0.9231	3	-	0.3914	0.3466	0.3625	0.3778	0.3067	0.3810
1.8462	6	-	0.3835	0.3940	0.3789	0.3857	0.3407	0.3808
2.7692	9	-	0.4028	0.4159	0.3961	0.4098	0.3803	0.4029
3.0769	10	3.1546	-	-	-	-	-	-
4.0	13	-	0.3992	0.4209	0.3905	0.4121	0.3806	0.4009
4.6154	15	-	0.4003	0.4207	0.3886	0.4031	0.3729	0.4042

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.1.1
• Transformers: 4.44.2
• PyTorch: 2.4.1+cu121
• Accelerate: 0.35.0.dev0
• Datasets: 3.0.1
• 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}
}