adriansanz/sqv2

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/sqv2")
# Run inference
sentences = [
    'La presentació de la sol·licitud no dona dret al muntatge de la parada.',
    'Quin és el requisit per a la presentació de la sol·licitud d’autorització?',
    'Quin és el motiu per canviar la persona titular dels drets funeraris?',
]
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	0.044
cosine_accuracy@3	0.116
cosine_accuracy@5	0.18
cosine_accuracy@10	0.3507
cosine_precision@1	0.044
cosine_precision@3	0.0387
cosine_precision@5	0.036
cosine_precision@10	0.0351
cosine_recall@1	0.044
cosine_recall@3	0.116
cosine_recall@5	0.18
cosine_recall@10	0.3507
cosine_ndcg@10	0.1659
cosine_mrr@10	0.111
cosine_map@100	0.1341

Information Retrieval

• Dataset: dim_768
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0413
cosine_accuracy@3	0.116
cosine_accuracy@5	0.1787
cosine_accuracy@10	0.3627
cosine_precision@1	0.0413
cosine_precision@3	0.0387
cosine_precision@5	0.0357
cosine_precision@10	0.0363
cosine_recall@1	0.0413
cosine_recall@3	0.116
cosine_recall@5	0.1787
cosine_recall@10	0.3627
cosine_ndcg@10	0.169
cosine_mrr@10	0.1116
cosine_map@100	0.1341

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0467
cosine_accuracy@3	0.116
cosine_accuracy@5	0.1787
cosine_accuracy@10	0.356
cosine_precision@1	0.0467
cosine_precision@3	0.0387
cosine_precision@5	0.0357
cosine_precision@10	0.0356
cosine_recall@1	0.0467
cosine_recall@3	0.116
cosine_recall@5	0.1787
cosine_recall@10	0.356
cosine_ndcg@10	0.1677
cosine_mrr@10	0.1121
cosine_map@100	0.1346

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0387
cosine_accuracy@3	0.1067
cosine_accuracy@5	0.1707
cosine_accuracy@10	0.3413
cosine_precision@1	0.0387
cosine_precision@3	0.0356
cosine_precision@5	0.0341
cosine_precision@10	0.0341
cosine_recall@1	0.0387
cosine_recall@3	0.1067
cosine_recall@5	0.1707
cosine_recall@10	0.3413
cosine_ndcg@10	0.1587
cosine_mrr@10	0.1046
cosine_map@100	0.129

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0493
cosine_accuracy@3	0.1227
cosine_accuracy@5	0.1987
cosine_accuracy@10	0.3667
cosine_precision@1	0.0493
cosine_precision@3	0.0409
cosine_precision@5	0.0397
cosine_precision@10	0.0367
cosine_recall@1	0.0493
cosine_recall@3	0.1227
cosine_recall@5	0.1987
cosine_recall@10	0.3667
cosine_ndcg@10	0.1759
cosine_mrr@10	0.119
cosine_map@100	0.142

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0373
cosine_accuracy@3	0.0947
cosine_accuracy@5	0.1573
cosine_accuracy@10	0.34
cosine_precision@1	0.0373
cosine_precision@3	0.0316
cosine_precision@5	0.0315
cosine_precision@10	0.034
cosine_recall@1	0.0373
cosine_recall@3	0.0947
cosine_recall@5	0.1573
cosine_recall@10	0.34
cosine_ndcg@10	0.1535
cosine_mrr@10	0.0987
cosine_map@100	0.1226

Training Details

Training Dataset

json

• Dataset: json
• Size: 6,749 training samples
• Columns: positive and anchor
• Approximate statistics based on the first 1000 samples:

  	positive	anchor
  type	string	string
  details	min: 6 tokens mean: 42.03 tokens max: 106 tokens	min: 10 tokens mean: 20.32 tokens max: 54 tokens

• Samples:

  positive	anchor
  Aquest tràmit us permet compensar deutes de naturalesa pública a favor de l'Ajuntament, sigui quin sigui el seu estat (voluntari/executiu), amb crèdits reconeguts per aquest a favor del mateix deutor, i que el seu estat sigui pendent de pagament.	Quin és el benefici de la compensació de deutes amb crèdits?
  El seu objecte és que -prèviament a la seva execució material- l'Ajuntament comprovi l'adequació de l’actuació a la normativa i planejament, així com a les ordenances municipals sobre l’ús del sòl i edificació.	Quin és el paper de les ordenances municipals en aquest tràmit?
  Comunicació prèvia del manteniment en espais, zones o instal·lacions comunitàries interiors dels edificis (reparació i/o millora de materials).	Quin és el límit del manteniment en espais comunitaris interiors dels edificis?

• 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
0.3791	10	3.0867	-	-	-	-	-	-
0.7583	20	2.4414	-	-	-	-	-	-
0.9858	26	-	0.1266	0.1255	0.1232	0.1257	0.1091	0.1345
1.1351	30	1.7091	-	-	-	-	-	-
1.5142	40	1.2495	-	-	-	-	-	-
1.8934	50	0.9813	-	-	-	-	-	-
1.9692	52	-	0.1315	0.1325	0.1285	0.1328	0.1218	0.1309
2.2701	60	0.6918	-	-	-	-	-	-
2.6493	70	0.7146	-	-	-	-	-	-
2.9905	79	-	0.1370	0.1344	0.1355	0.1338	0.1269	0.1363
3.0261	80	0.6002	-	-	-	-	-	-
3.4052	90	0.4816	-	-	-	-	-	-
3.7844	100	0.4949	-	-	-	-	-	-
3.9739	105	-	0.1357	0.1393	0.1302	0.1347	0.1204	0.1354
4.1611	110	0.474	-	-	-	-	-	-
4.5403	120	0.4692	-	-	-	-	-	-
4.9194	130	0.4484	0.1341	0.142	0.129	0.1346	0.1226	0.1341

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.1.0
• Transformers: 4.44.2
• PyTorch: 2.4.0+cu121
• Accelerate: 0.35.0.dev0
• Datasets: 3.0.0
• 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}
}