gubartz/facet_retriever

SentenceTransformer based on sentence-transformers/paraphrase-MiniLM-L12-v2

This is a sentence-transformers model finetuned from sentence-transformers/paraphrase-MiniLM-L12-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: sentence-transformers/paraphrase-MiniLM-L12-v2
• Maximum Sequence Length: 256 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': 256, '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("gubartz/facet_retriever")
# Run inference
sentences = [
    'purpose: 2.2 Decentralization and participation',
    'purpose: Social norm approach and feedback',
    'findings: The upper path of the model represents how counter-knowledge directly affects ACAP, reducing HC.',
]
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

Triplet

• Dataset: triplet
• Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.6998
dot_accuracy	0.3967
manhattan_accuracy	0.6999
euclidean_accuracy	0.7153
max_accuracy	0.7153

Training Details

Training Dataset

Unnamed Dataset

• Size: 1,541,116 training samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 9 tokens mean: 42.16 tokens max: 187 tokens	min: 10 tokens mean: 42.77 tokens max: 183 tokens	min: 8 tokens mean: 38.65 tokens max: 227 tokens

• Samples:

  anchor	positive	negative
  purpose: study attempts to fill this gap by examining firm-specific capabilities of Turkish outward FDI firms.	purpose: In short, the above-mentioned percentages show the lack of usage of knowledge sharing and collaborative technologies in some research institutions in Malaysia due to perceived causes such as non-availability of technology, lack of support, absent of teamwork culture, and lack of knowledge and training.	purpose: While SMA alone must not be used to gather and analyze these voices, these tools can guide organizations in relating to their publics, increasing the way groups identify with them and motivating these groups to enter into relationships with them.
  purpose: In this section of the paper, we try to explain citizen attitudes towards sustainable procurement.	purpose: Different from previous studies to concern key factors for motivating consumers' online buying behavior and behavioral intention (Liang and Lim, 2011; Zhang et al., 2013), such finding add knowledge in the filed by finding the meaningful affective mechanism of consumers in OFGB.	purpose: Task significance is not significantly different among generational cohorts of knowledge workers.
  purpose: However, the extensive use of information technology (IT) also comes with related security problems caused by the abstract nature of interacting systems - technical and organizational - and the seemingly lack of or inferior control of data or information.	purpose: No previous research using cluster analysis in nursing homes was found, but clusters identified in this study are lower than in previous hospital-based research into patients experiences and satisfaction used as cluster variables (Grondahl et al., 2011).	purpose: Yet, this engagement has tended to only involve a small section of the overall medical workforce in practice, raising questions about the nature of medical engagement more broadly and the mechanisms needed to enhance these processes.

• Loss: TripletLoss with these parameters:

  {
      "distance_metric": "TripletDistanceMetric.EUCLIDEAN",
      "triplet_margin": 5
  }
  

Evaluation Dataset

Unnamed Dataset

• Size: 199,564 evaluation samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 9 tokens mean: 42.64 tokens max: 165 tokens	min: 9 tokens mean: 42.42 tokens max: 197 tokens	min: 6 tokens mean: 38.23 tokens max: 193 tokens

• Samples:

  anchor	positive	negative
  purpose: However, it seems obvious that, in the long run, Green OA can be seen as leading progressively to the disappearance of the "traditional" publication model and, possibly, of scientific publishers altogether unless they reconsider their business model and adapt to the new situation.	purpose: Considering the transcendence of the sustainable development agenda in the UDRD, it was decided to search for explicit references to the issue of risk in the proposed indicators, finding a correspondence between four indicators of the development agenda and indicators proposed for the implementation of the Sendai Framework (Maskrey, 2016).	purpose: Finally, the terms of the permanent multinomial corresponding to the particular manufacturing system may be listed and the resulting graphs may be obtained and used for structurally analyzing the capabilities of the manufacturing system in different areas.
  purpose: To what extent do information science and the other disciplines demonstrate interest in social network theory and social network analysis?RQ2.	purpose: This study explores relationships between relationship commitment, cooperative behavior and alliance performance from the perspectives of both companies and contract farmers.	purpose: 4.1 The respondents' health literacy skills
  purpose: The evidence discussed above shows the nature of forecasting connections in the income growth across the globe.	purpose: Namely, the paper confirms that there is vast deviation between the European countries when it comes to consumer trust in banking in general but also related to each studied banking service.	purpose: Healthcare is one of the major sectors in which Lean production is being considered and adopted as an improvement program (Poksinska, 2010).

• Loss: TripletLoss with these parameters:

  {
      "distance_metric": "TripletDistanceMetric.EUCLIDEAN",
      "triplet_margin": 5
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 128
• gradient_accumulation_steps: 16
• num_train_epochs: 5
• warmup_ratio: 0.1
• fp16: True
• load_best_model_at_end: True
• auto_find_batch_size: True

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 128
• 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: 5
• max_steps: -1
• lr_scheduler_type: linear
• 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: False
• fp16: True
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• 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
• 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: True
• 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
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	loss	triplet_cosine_accuracy
0.3322	500	4.2859	-	-
0.6645	1000	3.693	-	-
0.9967	1500	3.5602	-	-
1.0	1505	-	3.4908	0.6914
1.3289	2000	3.427	-	-
1.6611	2500	3.3854	-	-
1.9934	3000	3.3551	-	-
2.0	3010	-	3.3604	0.7000
2.3256	3500	3.2353	-	-
2.6578	4000	3.221	-	-
2.9900	4500	3.2038	-	-
3.0	4515	-	3.3203	0.7026
3.3223	5000	3.1019	-	-
3.6545	5500	3.0942	-	-
3.9867	6000	3.085	-	-
4.0	6020	-	3.3177	0.7014
4.3189	6500	3.0129	-	-
4.6512	7000	3.0083	-	-
4.9834	7500	2.9971	-	-
5.0	7525	-	3.3264	0.6998

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.0.0
• Transformers: 4.41.0
• PyTorch: 2.3.0+cu121
• Accelerate: 0.30.1
• Datasets: 2.19.1
• 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",
}

TripletLoss

@misc{hermans2017defense,
    title={In Defense of the Triplet Loss for Person Re-Identification}, 
    author={Alexander Hermans and Lucas Beyer and Bastian Leibe},
    year={2017},
    eprint={1703.07737},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}