SentenceTransformer based on sentence-transformers/all-mpnet-base-v2

This is a sentence-transformers model finetuned from sentence-transformers/all-mpnet-base-v2. It maps sentences & paragraphs to a 768-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/all-mpnet-base-v2
• Maximum Sequence Length: 384 tokens
• Output Dimensionality: 768 dimensions
• 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': 384, 'do_lower_case': False, 'architecture': 'MPNetModel'})
  (1): Pooling({'word_embedding_dimension': 768, '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})
  (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("ozgur-celik/all-mpnet-base-v2-qaitrain500-500")
# Run inference
sentences = [
    'How long should a patient wait for soft tissue to heal after a tooth extraction before undergoing regenerative surgery for ridge augmentation?',
    '# Vertical periodontal regeneration\nin combination with ridge\naugmentation\n\ninterproximal bone loss, the smile line, and esthetic\nexpectations should be considered. In cases where\nthe decision is difficult to make, even after the clin-\nician has evaluated these criteria, it can also be de-\ncided during the regenerative surgery. In some cas-\nes, the clinician should consider whether there will\nbe a soft tissue defect when the extraction is per-\nformed during the regenerative surgery, and then\nthe flap design has to be at least one tooth larger. In\ncases where the extraction is performed before the\nidge augmentation, a complete soft tissue healing\ntime of about 2 months should be allowed before\nthe regenerative surgery.\nIn this chapter, technical details of interproximal\nbone and soft tissue regeneration are reviewed\nthrough a representative case of vertical periodon-\ntal regeneration in combination with ridge augmen-\ntation. The 55-year-old, healthy, male patient was\ntreated with an immediate implant at site 12 a dec-\nade previously at another clinic. The patient experi-\nenced bleeding and purulent exudate around the\nimplant and sensitivity of the neighboring lateral\nincisor shortly after implant placement. He sought\ntreatment due to an abscess around the implant.',
    'Fig 9   Ridge preservation via placement of a biomaterial to reduce shrink-\nage following tooth extraction.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.6634, 0.3224],
#         [0.6634, 1.0000, 0.3342],
#         [0.3224, 0.3342, 1.0000]])

Training Details

Training Dataset

Unnamed Dataset

• Size: 793 training samples
• Columns: anchor, positive, negative_1, negative_2, negative_3, and negative_4
• Approximate statistics based on the first 793 samples:

  	anchor	positive	negative_1	negative_2	negative_3	negative_4
  type	string	string	string	string	string	string
  details	min: 9 tokensmean: 26.75 tokensmax: 70 tokens	min: 13 tokensmean: 220.18 tokensmax: 384 tokens	min: 7 tokensmean: 130.22 tokensmax: 384 tokens	min: 9 tokensmean: 124.06 tokensmax: 384 tokens	min: 9 tokensmean: 129.57 tokensmax: 384 tokens	min: 8 tokensmean: 132.48 tokensmax: 384 tokens

• Samples:

  anchor	positive	negative_1	negative_2	negative_3	negative_4
  What specific term does Larsson use to describe the personality-dependent, additional sucking urge observed in children who engage in non-nutritive sucking behaviors?	# B) Abbau schädlicher Gewohnheiten (Habits)llerdings lutschen nicht alle diese Kinder am Daumen, während viele Kin-der auch in offensichtlich nicht belastender Umwelt dieses Verhalten durch-us zeigen. Dies führte zur Annahme, dass persönlichkeitsabhängige Ursa-chen für das Zustandekommen des Habits zumindest mitveranwortlich ind. Larsson spricht hier von einem »surplus sucking urge« (zusätzlichen Saugbedürfnis), das manche Kinder haben und das evtl. mit dem zu tunhat, was viele Eltern als »besondere Anhänglichkeit« oder » Verschmustheit«bezeichnen.	# ugar and childhood obesitysame but the amount of sugar was reduced, positive effects such as lowered blood sugar,lowered LDL cholesterol, lowered triglycerides, improved liver function, and loweredinsulin levels became apparent quickly. In addition, the subjects under a low-sugar dietreacted much more strongly to satiety stimuli. This shows that “one calorie is not equalto one calorie."27Obesity and its accompanying symptoms are now among the most pressing health issues worldwide. In most cases, it is children who suffer from obesity for the rest of theirlives, because, contrary to popular belief, people do not grow out of being overweight;rather it is a serious disease. While diet is not the only contributing factor to obesity, it isour responsibility as pediatric dentists to address it for the sake of our young patients’health. Specifically, we can raise parents' awareness of this subject and expose andthereby prevent possible dietary traps. These include too frequent ...	# 8.5.1 Gingival recessionsPatient-related factors include oral hygiene habitsand general factors that affect wound healing after oralsurgery, including the patient's age and general medic-al factors such as preexisting medical conditions likediabetes, current medications like immunosuppres-sants, and external factors like smoking and stress.Tooth- and defect-related factors of prognostic rele-vance include tooth malpositions, the presence of cer-vical defects, and the initial recession depth as well asthe vertical level of the papillae adjacent to the reces-sion defect. Technique-related factors include, for ex-ample, the type of flap used, the amount of soft tissuetrauma during the procedure, and the manual skill ofthe clinician. While patient- as well as tooth- and de-fect-related factors can be controlled to some extentby careful case and defect selection, it is above all thetechnique-related factors that give the clinician the op-portunity to directly influence ...	# WHEN BRUSHING IS A STRUGGLEAt routine checkups, many parents report that brushing their child’s teethis a daily struggle and they are helplessly seeking tips and tricks to avoid a“wrestling match.” What worked entirely fine when the child was an infant hasuddenly become an ordeal for all concerned. Parents do not always manageto motivate their little ones to brush or have their teeth brushed with patienceand fun by including play. Especially with the youngest children, discussionsor positive reinforcement with different reward systems are only possible toa limited extent.Nonetheless, a solution must be found, because not brushing for daysor weeks is not an option in terms of the child’s well-being and oral health.The solution is simple: Parents just have to persevere. The more calmly andconfidently parents act, the quicker this period of refusal from children is over.Refusal to brush may be a child’s way of expressing desire for independence,which can turn other everyd...	# REFERENCES23. Roberts GJ, Cleaton-Jones PE, Fatti LP, et al. Patterns of breast and bottle feeding and their associ-ation with dental caries in 1- to 4-year-old South African children. 1. Dental caries prevalence andexperience. Community Dent Health 1993;10:405-413.24. Koletzko B, Brönstrup A, Cremer M, et al. Säuglingsernährung und Ernährung der stillenden Mutter:
  What type of implant is recommended for a narrow-diameter implant case according to the Esthetic Risk Assessment?	# Table 1 Esthetic Risk Assessment (ERA) * If three-dimensional imaging is available with the tooth in place1 Standard-diameter implant, regular connection2 Narrow-diameter implant, narrow connection1 Standard-diameter implant, regular connection2 Narrow-diameter implant, narrow connection	# Single-tooth gapThe implant rehabilitation of a single-tooth gap inthe anterior mandible, following the loss of onemandibular incisor, requires not only a special setof instruments, but also a particularly high level ofprecision from the implantologist. In most cases, theresultant gap is very narrow and the available spacebetween the roots of the neighboring teeth extreme-ly limited. Iatrogenic injury to the neighboring teethcan only be avoided with detailed preoperative radi-ologic investigations and by performing the surgicalprocedure with great care. If the residual ridge hashealed, the freshly inserted implant can be loadeddirectly with a prosthesis, which is highly advanta-geous for both the patient and the dentist.Because of the limited space, implants of reduceddiameter, ie, narrow platform (NP) implants such asNobelActive (diameter 3.0 and 3.3 mm) and Nobel-Direct (diameter 3.0 mm) from Nobel Biocare, orTouareg (diameter 3.0 mm) from Adin or the K.S.I.Bau...	# 2.3.1Application of ComputerTechnology in Surgical ImplantDentistryComputer-guided (static) surgery. The use of a static sur-gical template that reproduces the virtual implant posi-tion directly from computerized tomographic data with-out allowing the intraoperative modification of implantposition.Computer-navigated (dynamic) surgery. The use of a sur-gical navigation system that reproduces the virtual im-plant position directly from computerized tomographyand allows intraoperative changes in implant position.The group recommended that, with appropriate training,experience, and pre-surgical planning, these systemsmight be clinically beneficial in the following clinical sit-uations:Complex anatomy••Minimally invasive surgery.Optimization of implant placementin critical esthetic cases.Immediate loading	# The Relationship Between Peri-implantEsthetics and the Biologic WidthThe presence of a papilla between teeth and implants and between implants is fundamental to an accept-able esthetic result. When single implants are placedbetween healthy teeth, the interproximal soft tissuesare maintained by adjacent bone crests. The position ofthe bone crest must be analyzed through an ultrasound	# Treatment guidelinesEsthetic outcomes can be achieved at post-extractionsites irrespective of the timing of implant placement.Different placement times, however, present with spe-cific treatment challenges and variable predictability ofesthetic outcomes.With immediate placement, a high level of clinical com-petence and experience in performing the treatment isneeded. Careful case selection is required to achieve sat-isfactory esthetic outcomes. The following clinical condi-tions should be satisfied:
  Which specific shade of composite material was utilized to simulate the incisal halo effect on the left central incisor during the restorative procedure described in the Manauta case?	Q: You invented the centripetal technique in the year 1994. Is it only for posteriors?n search of a solution to overcome the challenges of restoring theproximal cavities of posterior teeth and in order to facilitate an easierand quicker technique with a more effective result, the idea of thecentripetal buildup was born.The idea was to define and delimit the proximal surface with athin layer of composite that is actually contoured with the aid of aprecontoured matrix band and then to fill up the inner part of thecavity with a composite resin restorative material. This processallows for the utilization of different translucencies, transparencies,and colors of the composite resins according to the area that isbeing restored, meaning that the enamel that is more translucent and has a more pronounced opalescence will be restored with acertain type of composite, whereas the dentin part will be restoredwith a more chromatic dentin shade.The initial idea was to implement this te...	# The incisal hooktechniqueThe workflow selected in this case was first a morphologicalrestoration of the palatal areas together with an increasein the vertical dimension and design of the incisal edges(type 1 palatal veneer).563	Fig 4-33h Left lateral incisor. Folded protectivematrices in place for enamel etching, followed bydrying.	# LITERATUR2 Chiche GJ, Pinault A. Artistic and scientific principlesapplied to esthetic dentistry. In: Chiche GJ, Pinault A(eds). Esthetics of Anterior Fixed Prosthodontics. Chi-cago: Quintessence, 1994:13-32	# ReferencesPfeiffer J. Dental CAD/CAM technologies: the optical impression (I). Int299.J Comput Dent 1998;1:29-33.

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• learning_rate: 2e-05
• num_train_epochs: 2
• warmup_ratio: 0.1
• data_seed: 42
• dataloader_pin_memory: False
• hub_model_id: ozgur-celik/all-mpnet-base-v2-qaitrain500-500
• hub_private_repo: True
• eval_on_start: True
• batch_sampler: no_duplicates

All Hyperparameters

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 8
• per_device_eval_batch_size: 8
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• 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: 2
• 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: 42
• jit_mode_eval: False
• use_ipex: False
• bf16: False
• fp16: False
• 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: False
• 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}
• tp_size: 0
• 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: False
• dataloader_persistent_workers: False
• skip_memory_metrics: True
• use_legacy_prediction_loop: False
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: ozgur-celik/all-mpnet-base-v2-qaitrain500-500
• hub_strategy: every_save
• hub_private_repo: True
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• include_for_metrics: []
• 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
• 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: True
• use_liger_kernel: False
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Framework Versions

• Python: 3.12.3
• Sentence Transformers: 5.2.2
• Transformers: 4.51.3
• PyTorch: 2.8.0+cu128
• Accelerate: 1.14.0
• Datasets: 5.0.0
• Tokenizers: 0.21.4

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",
}

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}
}