adriansanz/sitges10242608-4ep-rerankv3

SentenceTransformer based on cross-encoder/ms-marco-MiniLM-L-4-v2

This is a sentence-transformers model finetuned from cross-encoder/ms-marco-MiniLM-L-4-v2. It maps sentences & paragraphs to a 384-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: cross-encoder/ms-marco-MiniLM-L-4-v2
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 384 tokens
• Similarity Function: Cosine Similarity

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': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, '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("adriansanz/sitges10242608-4ep-rerankv3")
# Run inference
sentences = [
    "Aquest tràmit permet sol·licitar la llicència per a realitzar obres d'excavació a la via pública per a la instal·lació o reparació d'infraestructures de serveis i subministraments.",
    'Quin és el paper de la via pública en aquest tràmit?',
    "Quin és l'objectiu de presentar una denúncia per presumpta infracció urbanística?",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

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

Evaluation

Metrics

Information Retrieval

• Dataset: dim_768
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0517
cosine_accuracy@3	0.1272
cosine_accuracy@5	0.1789
cosine_accuracy@10	0.3254
cosine_precision@1	0.0517
cosine_precision@3	0.0424
cosine_precision@5	0.0358
cosine_precision@10	0.0325
cosine_recall@1	0.0517
cosine_recall@3	0.1272
cosine_recall@5	0.1789
cosine_recall@10	0.3254
cosine_ndcg@10	0.1628
cosine_mrr@10	0.1143
cosine_map@100	0.1362

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0517
cosine_accuracy@3	0.1272
cosine_accuracy@5	0.1789
cosine_accuracy@10	0.3254
cosine_precision@1	0.0517
cosine_precision@3	0.0424
cosine_precision@5	0.0358
cosine_precision@10	0.0325
cosine_recall@1	0.0517
cosine_recall@3	0.1272
cosine_recall@5	0.1789
cosine_recall@10	0.3254
cosine_ndcg@10	0.1628
cosine_mrr@10	0.1143
cosine_map@100	0.1362

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0453
cosine_accuracy@3	0.1207
cosine_accuracy@5	0.1703
cosine_accuracy@10	0.3233
cosine_precision@1	0.0453
cosine_precision@3	0.0402
cosine_precision@5	0.0341
cosine_precision@10	0.0323
cosine_recall@1	0.0453
cosine_recall@3	0.1207
cosine_recall@5	0.1703
cosine_recall@10	0.3233
cosine_ndcg@10	0.1576
cosine_mrr@10	0.1083
cosine_map@100	0.1311

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0474
cosine_accuracy@3	0.1207
cosine_accuracy@5	0.1767
cosine_accuracy@10	0.3147
cosine_precision@1	0.0474
cosine_precision@3	0.0402
cosine_precision@5	0.0353
cosine_precision@10	0.0315
cosine_recall@1	0.0474
cosine_recall@3	0.1207
cosine_recall@5	0.1767
cosine_recall@10	0.3147
cosine_ndcg@10	0.1556
cosine_mrr@10	0.1083
cosine_map@100	0.1316

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0366
cosine_accuracy@3	0.1013
cosine_accuracy@5	0.153
cosine_accuracy@10	0.2845
cosine_precision@1	0.0366
cosine_precision@3	0.0338
cosine_precision@5	0.0306
cosine_precision@10	0.0284
cosine_recall@1	0.0366
cosine_recall@3	0.1013
cosine_recall@5	0.153
cosine_recall@10	0.2845
cosine_ndcg@10	0.1358
cosine_mrr@10	0.0917
cosine_map@100	0.1149

Training Details

Training Dataset

Unnamed Dataset

• Size: 4,173 training samples
• Columns: positive and anchor
• Approximate statistics based on the first 1000 samples:

  	positive	anchor
  type	string	string
  details	min: 10 tokens mean: 67.49 tokens max: 214 tokens	min: 11 tokens mean: 28.0 tokens max: 61 tokens

• Samples:

  positive	anchor
  Havent-se d'acreditar la matriculació i inscripció en el respectiu centre públic o concertat, així com el cost de les llars d'infants, de l'educació especialitzada per les discapacitats físiques, psíquiques i sensorials en centres públics, concertats o privats.	Quin és el requisit per acreditar la llar d'infants?
  El volant històric de convivència és el document que informa de la residencia en el municipi de Sitges, així com altres fets relatius a l'empadronament d'una persona, i detalla tots els domicilis, la data inicial i final en els que ha estat empadronada en cadascun d'ells, i les persones amb les què constava inscrites, segons les dades que consten al Padró Municipal d'Habitants fins a la data d'expedició.	Quin és el propòsit del volant històric de convivència?
  Instal·lació de tanques sense obra.	Quins són els exemples d'instal·lacions que es poden comunicar amb aquest tràmit?

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          384,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 16
• gradient_accumulation_steps: 16
• num_train_epochs: 10
• lr_scheduler_type: cosine
• warmup_ratio: 0.2
• bf16: True
• tf32: False
• 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: 32
• 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: 5e-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: 10
• 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: False
• 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	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
0.6130	10	11.3695	-	-	-	-	-
0.9808	16	-	0.0214	0.0243	0.0234	0.0199	0.0234
1.2261	20	10.653	-	-	-	-	-
1.8391	30	9.0745	-	-	-	-	-
1.9617	32	-	0.0495	0.0517	0.0589	0.0481	0.0589
2.4521	40	7.3468	-	-	-	-	-
2.9425	48	-	0.0764	0.0734	0.0811	0.0709	0.0811
3.0651	50	5.887	-	-	-	-	-
3.6782	60	5.3568	-	-	-	-	-
3.9847	65	-	0.0922	0.0857	0.0896	0.0808	0.0896
4.2912	70	4.8338	-	-	-	-	-
4.9042	80	4.9251	0.0899	0.0899	0.0906	0.0837	0.0906
0.9771	8	-	0.0953	0.0965	0.0957	0.0841	0.0957
1.2214	10	6.7779	-	-	-	-	-
1.9542	16	-	0.1056	0.1036	0.1078	0.0948	0.1078
2.4427	20	5.8485	-	-	-	-	-
2.9313	24	-	0.1112	0.1107	0.1170	0.1009	0.1170
3.6641	30	4.6394	-	-	-	-	-
3.9084	32	-	0.1243	0.1189	0.1247	0.1152	0.1247
4.8855	40	3.8786	0.1248	0.1248	0.1335	0.1148	0.1335
5.9847	49	-	0.1298	0.1298	0.1371	0.1204	0.1371
6.1069	50	3.3198	-	-	-	-	-
6.9618	57	-	0.1284	0.1347	0.1370	0.1208	0.1370
7.3282	60	3.081	-	-	-	-	-
7.9389	65	-	0.1273	0.1344	0.1360	0.1215	0.1360
8.5496	70	2.8556	-	-	-	-	-
8.9160	73	-	0.1313	0.1315	0.1350	0.1147	0.1350
9.771	80	2.7635	0.1316	0.1311	0.1362	0.1149	0.1362

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.0.1
• Transformers: 4.42.4
• PyTorch: 2.4.0+cu121
• Accelerate: 0.34.0.dev0
• Datasets: 2.21.0
• 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}
}