--- language: - en license: apache-2.0 tags: - sentence-transformers - sentence-similarity - feature-extraction - generated_from_trainer - dataset_size:23290 - loss:MatryoshkaLoss - loss:MultipleNegativesRankingLoss base_model: nomic-ai/modernbert-embed-base widget: - source_sentence: "Intervention \nSession \ntime \n(mins)\nMaterial resources \n\ Occupations \n(rehabilitation specialists)\nAssistive products \nEquipment \n\ Consumables \nActivities of daily living Target: Activities of daily living (ADL)\n\ Assessment of ADL\n30\n–\n•\tUtensils for activities of daily \nliving\n–\n•\t\ Occupational therapist \n•\tPhysiotherapist ADL training\n30\n–\n•\tUtensils for\ \ activities of daily \nliving\n•\tAssistive products for toileting \n•\tAssistive\ \ products for dressing\n–\n•\tOccupational therapist •\tPhysiotherapist\nProvision\ \ and training \nin the use of assistive \nproducts for self-care\n30\n•\tAssistive\ \ products for toileting \n•\tAssistive products for dressing\n– –\n•\tOccupational\ \ therapist \n•\tPhysiotherapist\nModification of the home \nenvironment\n60\n\ •\tHandrail/grab bar\n•\tRamps\n•\tMeasuring tape \n– •\tOccupational therapist\ \ \n•\tPhysiotherapist\nEducation and vocation\nTarget: Work and employment\n\ Vocational assessment\n90\n–\n•\tWork-related tools and \nequipment\n– •\tOccupational\ \ therapist\n•\tSocial work and counselling \nprofessional\nVocational counselling,\ \ \ntraining and support\n60\n–\n•\tWork-related tools and \nequipment •\tInformation\ \ materials (e.g. \nflyers, brochures)\n•\tOccupational therapist\n•\tSocial work\ \ and counselling \nprofessional\nSelf-management\nTarget: Self-management\nEducation,\ \ advice and" sentences: - self-management education in rehabilitation - rehabilitation interventions for dementia due to Alzheimer's disease - comparison of treadmill and swimming exercise for DCM - source_sentence: "correlations. However, no association was found between kidney\ \ \nstones and homeostatic dysregulation (p overall >0.05).\n3.4 Association between\ \ healthy dietary \nscores and aging indicators We conducted an analysis to examine\ \ the association between \nhealthy dietary scores and aging indicators listed\ \ in \nSupplementary Table S2. After adjusting for all covariates, \nwe  observed\ \ significant negative associations between healthy dietary scores (the AHEI,\ \ DASHI, HEI-2020, and MEDI) and \naging indicators (KDMAge, PhenoAge, homeostatic\ \ dysregulation, KDMAge accelerated aging, and PhenoAge accelerated aging). \n\ Additionally, the results of the multiple logistic regression analysis \nindicated\ \ that higher healthy dietary scores were associated with a decreased risk of\ \ accelerated aging, as shown by KDMAge and \nPhenoAge. Restricted cubic splines\ \ in Figure 3 visualize both linear \nand non-linear relationships between healthy\ \ dietary scores and aging indicators. KDMAge, PhenoAge, KDMAge accelerated \n\ aging, and PhenoAge accelerated aging all exhibited linear \nassociations with\ \ the four dietary scores.\n3.5 Aging indicators partially mediated the association\ \ between a healthy diet and \nkidney stones\nWe assessed the mediating effect\ \ in the association between \nkidney stones and a healthy diet using mediation\ \ analysis, with aging indicators as mediators (Supplementary Table S3). Our results\ \ suggest \nthat the aging indicators KDMAge, PhenoAge, KDMAge accelerated \n\ aging, and PhenoAge accelerated aging significantly mediated the" sentences: - Mental health assessment for Parkinson disease by psychologists - linear relationships between dietary scores and KDMAge PhenoAge - wrist joint fractures immobilization - source_sentence: "•\tMoisturizing product\n•\tMouthpieces\n•\tNeedles and syringes\ \ \n•\tNose clips\n•\tNutritional diary\n•\tOral anesthetic spray\n•\tOral swabs\ \ •\tReplaceable sticky electrode pads \n•\tSpeaking valve and tracheostomy \n\ cap\n•\tSpecimen cup\n•\tStraws \n•\tSyringes\n•\tTissues •\tTongue depressor\n\ •\tTopical antiseptics\n•\tTubing\n•\tWound dressings\n101\nAssistive products\ \ (for prescription)\nEquipment (for service facilities) Consumables (for service\ \ facilities)\n•\tStanding frames, adjustable\n•\tWalking frames/walkers\n•\t\ Wheelchair (manual or electrical)\n•\tCasting kit\n•\tOrthoses kit •\tSplinting\ \ kit (static/dynamic)\n•\tAssistant support belt\n•\tTransfer boards/slide sheet\n\ •\tFoam rollers/wedges\n•\tPillows\n•\tTreatment table •\tStools/small benches\ \ of varying \nheight\n•\t(Functional) electrical stimulation kit\n•\tTENS Supply\ \ kit\n•\tResistance bands \n•\tResistive exercise putty \n•\tWeights •\tExercise\ \ mat\n•\tExercise ball \n•\tBalance board/cushion\n•\tTraining stairs\n•\tSteps\ \ (stackable)\n•\tRamps (temporary/mobile)\n•\tMobile mirror\n•\tParallel bar\ \ •\tTimer\n•\tCycle ergometer (arm or leg)\n•\tUpper limb workstation\n•\tEquipment\ \ for sport and recreational \nand leisure activities\n•\tAssistive products for\ \ recreational \nand leisure activities" sentences: - assistive products for work and employment workstation adaptation - Wound dressings and topical antiseptics for injury care - Short-chain carbohydrates in grains and cereals - source_sentence: "made little to no difference at other \ntime-points;\n\t›\n\t\ improved health-related quality of life \nin the immediate term (small effect)\ \ \nand made little to no difference at other \ntime-points; ›\n\treduced catastrophic\ \ thinking in the \nimmediate term (moderate effect); or\n\t›\n\tmade little to\ \ no difference to pain or \nsocial participation at any time-point. In a single\ \ trial among older people, no \nbenefits for pain or function were observed \n\ (very low certainty evidence).\nFive trials reported on adverse events. A \nfew\ \ serious adverse events were reported, although none was related to the SMT \n\ intervention. Three trials which examined \nadverse effects reported that they\ \ were \ncommonly transient, ranged from mild \nto moderate severity and were\ \ related to musculoskeletal soreness and tiredness. \nNo data on harms among\ \ older people were \nreported. \n•\t In the comparison of SMT with no \nintervention\ \ (four trials), benefits were observed for pain, function, health-related \n\ quality of life and psychological outcomes. \nHowever, since the certainty of\ \ the evidence \nwas very low, it was uncertain whether SMT:\n\t› reduced pain\ \ and improved back-\nspecific function in the immediate term \n(moderate effects)\ \ and made little to no \ndifference at other time-points;\n\t›\n\timproved general\ \ function in the immediate and short term (small to \nmoderate effects) and made\ \ little to no \ndifference at other time-points;\n\t›\n\timproved health-related\ \ quality in the \nimmediate term (small effect);" sentences: - impact of low FODMAP diet on bloating and pain in IBS patients - swallowing therapy tools and products - Adverse events of SMT in dietary science - source_sentence: "CRONIN (Occupational therapist, USA); Diane DAMIANO (Physiotherapist,\ \ USA); Wouter de GROOTE (PRM physician, Belgium); Pamela ENDERBY (Speech and\ \ language therapist, United Kingdom); Darcy \nFEHLINGS (Developmental pediatrician/Clinician\ \ scientist, Canada); Charne FERIS (Occupational therapist, Namibia); Ferdiliza\ \ Dandah GARCIA (Speech pathologist and medical doctor, Philippines); Mohammad\ \ Mohinul ISLAM (Physiotherapist, Bangladesh); Heakyung KIM (PRM physician, USA);\ \ Pavlina PSYCHOULI (Occupational therapist, Greece); Mehdi RASSAFIANI (Occupational\ \ therapist, Islamic Republic of Iran); Gillian SALOOJEE (Physiotherapist, South\ \ Africa); Abena TANNOR (PRM physician/\nFamily medicine, Ghana).\nMembers of\ \ the peer review group An MIHEE (Physiotherapist, Republic of Korea); Uthman\ \ Olayiwola ANJORIN (Physiotherapist, Nigeria); Merce AVELLANET (PRM physician,\ \ Andorra); Marie BRIEN (Physiotherapist, Canada); Annemieke BUIZER" sentences: - Developmental pediatrician research and studies - materials needed for plaster of Paris casts - impact of client medications on exercise responses pipeline_tag: sentence-similarity library_name: sentence-transformers metrics: - cosine_accuracy@1 - cosine_accuracy@3 - cosine_accuracy@5 - cosine_accuracy@10 - cosine_precision@1 - cosine_precision@3 - cosine_precision@5 - cosine_precision@10 - cosine_recall@1 - cosine_recall@3 - cosine_recall@5 - cosine_recall@10 - cosine_ndcg@10 - cosine_mrr@10 - cosine_map@100 model-index: - name: ModernBERT Embed base fitness health Matryoshka results: - task: type: information-retrieval name: Information Retrieval dataset: name: dim 768 type: dim_768 metrics: - type: cosine_accuracy@1 value: 0.5370942812982998 name: Cosine Accuracy@1 - type: cosine_accuracy@3 value: 0.5374806800618238 name: Cosine Accuracy@3 - type: cosine_accuracy@5 value: 0.5374806800618238 name: Cosine Accuracy@5 - type: cosine_accuracy@10 value: 0.5664605873261206 name: Cosine Accuracy@10 - type: cosine_precision@1 value: 0.5370942812982998 name: Cosine Precision@1 - type: cosine_precision@3 value: 0.5372230808861412 name: Cosine Precision@3 - type: cosine_precision@5 value: 0.5372488408037095 name: Cosine Precision@5 - type: cosine_precision@10 value: 0.48898763523956723 name: Cosine Precision@10 - type: cosine_recall@1 value: 0.032969689163661345 name: Cosine Recall@1 - type: cosine_recall@3 value: 0.09892838742916021 name: Cosine Recall@3 - type: cosine_recall@5 value: 0.1648870856946591 name: Cosine Recall@5 - type: cosine_recall@10 value: 0.29061909668555724 name: Cosine Recall@10 - type: cosine_ndcg@10 value: 0.5034792038193298 name: Cosine Ndcg@10 - type: cosine_mrr@10 value: 0.5420530654301913 name: Cosine Mrr@10 - type: cosine_map@100 value: 0.3249220491378387 name: Cosine Map@100 - task: type: information-retrieval name: Information Retrieval dataset: name: dim 512 type: dim_512 metrics: - type: cosine_accuracy@1 value: 0.5312982998454405 name: Cosine Accuracy@1 - type: cosine_accuracy@3 value: 0.5312982998454405 name: Cosine Accuracy@3 - type: cosine_accuracy@5 value: 0.5312982998454405 name: Cosine Accuracy@5 - type: cosine_accuracy@10 value: 0.5618238021638331 name: Cosine Accuracy@10 - type: cosine_precision@1 value: 0.5312982998454405 name: Cosine Precision@1 - type: cosine_precision@3 value: 0.5312982998454405 name: Cosine Precision@3 - type: cosine_precision@5 value: 0.5312982998454405 name: Cosine Precision@5 - type: cosine_precision@10 value: 0.4845440494590417 name: Cosine Precision@10 - type: cosine_recall@1 value: 0.032712089987978706 name: Cosine Recall@1 - type: cosine_recall@3 value: 0.09813626996393612 name: Cosine Recall@3 - type: cosine_recall@5 value: 0.1635604499398935 name: Cosine Recall@5 - type: cosine_recall@10 value: 0.2885840631976644 name: Cosine Recall@10 - type: cosine_ndcg@10 value: 0.4986060701912577 name: Cosine Ndcg@10 - type: cosine_mrr@10 value: 0.536385883565173 name: Cosine Mrr@10 - type: cosine_map@100 value: 0.32255571092212365 name: Cosine Map@100 - task: type: information-retrieval name: Information Retrieval dataset: name: dim 256 type: dim_256 metrics: - type: cosine_accuracy@1 value: 0.5162287480680062 name: Cosine Accuracy@1 - type: cosine_accuracy@3 value: 0.5162287480680062 name: Cosine Accuracy@3 - type: cosine_accuracy@5 value: 0.5162287480680062 name: Cosine Accuracy@5 - type: cosine_accuracy@10 value: 0.5498454404945904 name: Cosine Accuracy@10 - type: cosine_precision@1 value: 0.5162287480680062 name: Cosine Precision@1 - type: cosine_precision@3 value: 0.5162287480680062 name: Cosine Precision@3 - type: cosine_precision@5 value: 0.5162287480680062 name: Cosine Precision@5 - type: cosine_precision@10 value: 0.4728748068006182 name: Cosine Precision@10 - type: cosine_recall@1 value: 0.03184054611025245 name: Cosine Recall@1 - type: cosine_recall@3 value: 0.09552163833075734 name: Cosine Recall@3 - type: cosine_recall@5 value: 0.15920273055126222 name: Cosine Recall@5 - type: cosine_recall@10 value: 0.2826893353941267 name: Cosine Recall@10 - type: cosine_ndcg@10 value: 0.4859196209845146 name: Cosine Ndcg@10 - type: cosine_mrr@10 value: 0.5218154301906238 name: Cosine Mrr@10 - type: cosine_map@100 value: 0.31600673687689496 name: Cosine Map@100 - task: type: information-retrieval name: Information Retrieval dataset: name: dim 128 type: dim_128 metrics: - type: cosine_accuracy@1 value: 0.49265842349304484 name: Cosine Accuracy@1 - type: cosine_accuracy@3 value: 0.49265842349304484 name: Cosine Accuracy@3 - type: cosine_accuracy@5 value: 0.49304482225656876 name: Cosine Accuracy@5 - type: cosine_accuracy@10 value: 0.5193199381761978 name: Cosine Accuracy@10 - type: cosine_precision@1 value: 0.49265842349304484 name: Cosine Precision@1 - type: cosine_precision@3 value: 0.49265842349304484 name: Cosine Precision@3 - type: cosine_precision@5 value: 0.4927357032457496 name: Cosine Precision@5 - type: cosine_precision@10 value: 0.44837712519319944 name: Cosine Precision@10 - type: cosine_recall@1 value: 0.030391550747037612 name: Cosine Recall@1 - type: cosine_recall@3 value: 0.09117465224111283 name: Cosine Recall@3 - type: cosine_recall@5 value: 0.15200927357032457 name: Cosine Recall@5 - type: cosine_recall@10 value: 0.2676391035548686 name: Cosine Recall@10 - type: cosine_ndcg@10 value: 0.4617161176693884 name: Cosine Ndcg@10 - type: cosine_mrr@10 value: 0.4971342091705308 name: Cosine Mrr@10 - type: cosine_map@100 value: 0.3035061522343962 name: Cosine Map@100 - task: type: information-retrieval name: Information Retrieval dataset: name: dim 64 type: dim_64 metrics: - type: cosine_accuracy@1 value: 0.42619783616692425 name: Cosine Accuracy@1 - type: cosine_accuracy@3 value: 0.42619783616692425 name: Cosine Accuracy@3 - type: cosine_accuracy@5 value: 0.42619783616692425 name: Cosine Accuracy@5 - type: cosine_accuracy@10 value: 0.45904173106646057 name: Cosine Accuracy@10 - type: cosine_precision@1 value: 0.42619783616692425 name: Cosine Precision@1 - type: cosine_precision@3 value: 0.4259402369912416 name: Cosine Precision@3 - type: cosine_precision@5 value: 0.42604327666151465 name: Cosine Precision@5 - type: cosine_precision@10 value: 0.39316074188562594 name: Cosine Precision@10 - type: cosine_recall@1 value: 0.026257942641250212 name: Cosine Recall@1 - type: cosine_recall@3 value: 0.0787223080886141 name: Cosine Recall@3 - type: cosine_recall@5 value: 0.13123819337111453 name: Cosine Recall@5 - type: cosine_recall@10 value: 0.23491971492357888 name: Cosine Recall@10 - type: cosine_ndcg@10 value: 0.4030993901354209 name: Cosine Ndcg@10 - type: cosine_mrr@10 value: 0.4316718186501803 name: Cosine Mrr@10 - type: cosine_map@100 value: 0.2688982568901479 name: Cosine Map@100 --- # ModernBERT Embed base fitness health Matryoshka This is a [sentence-transformers](https://www.SBERT.net) model finetuned from [nomic-ai/modernbert-embed-base](https://huggingface.co/nomic-ai/modernbert-embed-base) on the json dataset. 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:** [nomic-ai/modernbert-embed-base](https://huggingface.co/nomic-ai/modernbert-embed-base) - **Maximum Sequence Length:** 8192 tokens - **Output Dimensionality:** 768 dimensions - **Similarity Function:** Cosine Similarity - **Training Dataset:** - json - **Language:** en - **License:** apache-2.0 ### Model Sources - **Documentation:** [Sentence Transformers Documentation](https://sbert.net) - **Repository:** [Sentence Transformers on GitHub](https://github.com/UKPLab/sentence-transformers) - **Hugging Face:** [Sentence Transformers on Hugging Face](https://huggingface.co/models?library=sentence-transformers) ### Full Model Architecture ``` SentenceTransformer( (0): Transformer({'max_seq_length': 8192, 'do_lower_case': False}) with Transformer model: ModernBertModel (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: ```bash pip install -U sentence-transformers ``` Then you can load this model and run inference. ```python from sentence_transformers import SentenceTransformer # Download from the 🤗 Hub model = SentenceTransformer("kokojake/modernbert-embed-base-fitness-health-matryoshka-5-epochs-25k") # Run inference sentences = [ 'CRONIN (Occupational therapist, USA); Diane DAMIANO (Physiotherapist, USA); Wouter de GROOTE (PRM physician, Belgium); Pamela ENDERBY (Speech and language therapist, United Kingdom); Darcy \nFEHLINGS (Developmental pediatrician/Clinician scientist, Canada); Charne FERIS (Occupational therapist, Namibia); Ferdiliza Dandah GARCIA (Speech pathologist and medical doctor, Philippines); Mohammad Mohinul ISLAM (Physiotherapist, Bangladesh); Heakyung KIM (PRM physician, USA); Pavlina PSYCHOULI (Occupational therapist, Greece); Mehdi RASSAFIANI (Occupational therapist, Islamic Republic of Iran); Gillian SALOOJEE (Physiotherapist, South Africa); Abena TANNOR (PRM physician/\nFamily medicine, Ghana).\nMembers of the peer review group An MIHEE (Physiotherapist, Republic of Korea); Uthman Olayiwola ANJORIN (Physiotherapist, Nigeria); Merce AVELLANET (PRM physician, Andorra); Marie BRIEN (Physiotherapist, Canada); Annemieke BUIZER', 'Developmental pediatrician research and studies', 'materials needed for plaster of Paris casts', ] embeddings = model.encode(sentences) print(embeddings.shape) # [3, 768] # 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](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.InformationRetrievalEvaluator) with these parameters: ```json { "truncate_dim": 768 } ``` | Metric | Value | |:--------------------|:-----------| | cosine_accuracy@1 | 0.5371 | | cosine_accuracy@3 | 0.5375 | | cosine_accuracy@5 | 0.5375 | | cosine_accuracy@10 | 0.5665 | | cosine_precision@1 | 0.5371 | | cosine_precision@3 | 0.5372 | | cosine_precision@5 | 0.5372 | | cosine_precision@10 | 0.489 | | cosine_recall@1 | 0.033 | | cosine_recall@3 | 0.0989 | | cosine_recall@5 | 0.1649 | | cosine_recall@10 | 0.2906 | | **cosine_ndcg@10** | **0.5035** | | cosine_mrr@10 | 0.5421 | | cosine_map@100 | 0.3249 | #### Information Retrieval * Dataset: `dim_512` * Evaluated with [InformationRetrievalEvaluator](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.InformationRetrievalEvaluator) with these parameters: ```json { "truncate_dim": 512 } ``` | Metric | Value | |:--------------------|:-----------| | cosine_accuracy@1 | 0.5313 | | cosine_accuracy@3 | 0.5313 | | cosine_accuracy@5 | 0.5313 | | cosine_accuracy@10 | 0.5618 | | cosine_precision@1 | 0.5313 | | cosine_precision@3 | 0.5313 | | cosine_precision@5 | 0.5313 | | cosine_precision@10 | 0.4845 | | cosine_recall@1 | 0.0327 | | cosine_recall@3 | 0.0981 | | cosine_recall@5 | 0.1636 | | cosine_recall@10 | 0.2886 | | **cosine_ndcg@10** | **0.4986** | | cosine_mrr@10 | 0.5364 | | cosine_map@100 | 0.3226 | #### Information Retrieval * Dataset: `dim_256` * Evaluated with [InformationRetrievalEvaluator](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.InformationRetrievalEvaluator) with these parameters: ```json { "truncate_dim": 256 } ``` | Metric | Value | |:--------------------|:-----------| | cosine_accuracy@1 | 0.5162 | | cosine_accuracy@3 | 0.5162 | | cosine_accuracy@5 | 0.5162 | | cosine_accuracy@10 | 0.5498 | | cosine_precision@1 | 0.5162 | | cosine_precision@3 | 0.5162 | | cosine_precision@5 | 0.5162 | | cosine_precision@10 | 0.4729 | | cosine_recall@1 | 0.0318 | | cosine_recall@3 | 0.0955 | | cosine_recall@5 | 0.1592 | | cosine_recall@10 | 0.2827 | | **cosine_ndcg@10** | **0.4859** | | cosine_mrr@10 | 0.5218 | | cosine_map@100 | 0.316 | #### Information Retrieval * Dataset: `dim_128` * Evaluated with [InformationRetrievalEvaluator](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.InformationRetrievalEvaluator) with these parameters: ```json { "truncate_dim": 128 } ``` | Metric | Value | |:--------------------|:-----------| | cosine_accuracy@1 | 0.4927 | | cosine_accuracy@3 | 0.4927 | | cosine_accuracy@5 | 0.493 | | cosine_accuracy@10 | 0.5193 | | cosine_precision@1 | 0.4927 | | cosine_precision@3 | 0.4927 | | cosine_precision@5 | 0.4927 | | cosine_precision@10 | 0.4484 | | cosine_recall@1 | 0.0304 | | cosine_recall@3 | 0.0912 | | cosine_recall@5 | 0.152 | | cosine_recall@10 | 0.2676 | | **cosine_ndcg@10** | **0.4617** | | cosine_mrr@10 | 0.4971 | | cosine_map@100 | 0.3035 | #### Information Retrieval * Dataset: `dim_64` * Evaluated with [InformationRetrievalEvaluator](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.InformationRetrievalEvaluator) with these parameters: ```json { "truncate_dim": 64 } ``` | Metric | Value | |:--------------------|:-----------| | cosine_accuracy@1 | 0.4262 | | cosine_accuracy@3 | 0.4262 | | cosine_accuracy@5 | 0.4262 | | cosine_accuracy@10 | 0.459 | | cosine_precision@1 | 0.4262 | | cosine_precision@3 | 0.4259 | | cosine_precision@5 | 0.426 | | cosine_precision@10 | 0.3932 | | cosine_recall@1 | 0.0263 | | cosine_recall@3 | 0.0787 | | cosine_recall@5 | 0.1312 | | cosine_recall@10 | 0.2349 | | **cosine_ndcg@10** | **0.4031** | | cosine_mrr@10 | 0.4317 | | cosine_map@100 | 0.2689 | ## Training Details ### Training Dataset #### json * Dataset: json * Size: 23,290 training samples * Columns: positive and anchor * Approximate statistics based on the first 1000 samples: | | positive | anchor | |:--------|:------------------------------------------------------------------------------------|:----------------------------------------------------------------------------------| | type | string | string | | details | | | * Samples: | positive | anchor | |:--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|:--------------------------------------------------------------| | 5. Zeng CY, Zhang ZR, Tang ZM, Hua FZ. Benefits and mechanisms of exercise training for knee osteoarthritis.
Frontiers in Physiology. 2021;12. 6. Büssing A, Ostermann T, Lüdtke R, Michalsen A. Effects of yoga interventions on pain and pain-associated disability: a meta-analysis. J Pain. 2012;13(1):1-9. doi:10.1016/j.jpain.2011.10.001 7. Wren AA, Wright MA, Carson JW, Keefe FJ. Yoga for persistent pain: new findings and directions for an ancient practice. Pain. 2011;152(3):477-480. doi:10.1016/j.pain.2010.11.017 8. Lauche R, Hunter DJ, Adams J, Cramer H. Yoga for osteoarthritis: a systematic review and meta-analysis. Curr Rheumatol Rep. 2019;21(9):47. doi:10.1007/s11926-019-0846-5 9. Zhang Q, Young L, Li F. Network meta-analysis of various nonpharmacological interventions on pain relief in | yoga for persistent pain management | | CRONIN (Occupational therapist, USA); Diane DAMIANO (Physiotherapist, USA); Wouter de GROOTE (PRM physician, Belgium); Pamela ENDERBY (Speech and language therapist, United Kingdom); Darcy
FEHLINGS (Developmental pediatrician/Clinician scientist, Canada); Charne FERIS (Occupational therapist, Namibia); Ferdiliza Dandah GARCIA (Speech pathologist and medical doctor, Philippines); Mohammad Mohinul ISLAM (Physiotherapist, Bangladesh); Heakyung KIM (PRM physician, USA); Pavlina PSYCHOULI (Occupational therapist, Greece); Mehdi RASSAFIANI (Occupational therapist, Islamic Republic of Iran); Gillian SALOOJEE (Physiotherapist, South Africa); Abena TANNOR (PRM physician/
Family medicine, Ghana).
Members of the peer review group An MIHEE (Physiotherapist, Republic of Korea); Uthman Olayiwola ANJORIN (Physiotherapist, Nigeria); Merce AVELLANET (PRM physician, Andorra); Marie BRIEN (Physiotherapist, Canada); Annemieke BUIZER | Developmental pediatrician research and studies | | JAMA Network Open. 2025;8(4):e253698. doi:10.1001/jamanetworkopen.2025.3698
(Reprinted)
April 8, 2025 | JAMA Network Open 2025 study on medical research | * Loss: [MatryoshkaLoss](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#matryoshkaloss) with these parameters: ```json { "loss": "MultipleNegativesRankingLoss", "matryoshka_dims": [ 768, 512, 256, 128, 64 ], "matryoshka_weights": [ 1, 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 - `learning_rate`: 2e-05 - `num_train_epochs`: 5 - `lr_scheduler_type`: cosine - `warmup_ratio`: 0.1 - `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`: 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 - `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`: 0.1 - `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`: 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} - `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_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`: None - `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`: False - `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
### Training Logs | Epoch | Step | Training Loss | dim_768_cosine_ndcg@10 | dim_512_cosine_ndcg@10 | dim_256_cosine_ndcg@10 | dim_128_cosine_ndcg@10 | dim_64_cosine_ndcg@10 | |:-------:|:-------:|:-------------:|:----------------------:|:----------------------:|:----------------------:|:----------------------:|:---------------------:| | 0.2198 | 10 | 54.8397 | - | - | - | - | - | | 0.4396 | 20 | 26.5885 | - | - | - | - | - | | 0.6593 | 30 | 20.9275 | - | - | - | - | - | | 0.8791 | 40 | 17.6283 | - | - | - | - | - | | 1.0 | 46 | - | 0.4713 | 0.4725 | 0.4562 | 0.4333 | 0.3646 | | 1.0879 | 50 | 13.4942 | - | - | - | - | - | | 1.3077 | 60 | 12.4011 | - | - | - | - | - | | 1.5275 | 70 | 12.2302 | - | - | - | - | - | | 1.7473 | 80 | 11.7666 | - | - | - | - | - | | 1.9670 | 90 | 11.9032 | - | - | - | - | - | | 2.0 | 92 | - | 0.4909 | 0.4865 | 0.4760 | 0.4501 | 0.3923 | | 2.1758 | 100 | 9.4322 | - | - | - | - | - | | 2.3956 | 110 | 9.692 | - | - | - | - | - | | 2.6154 | 120 | 8.7793 | - | - | - | - | - | | 2.8352 | 130 | 8.3124 | - | - | - | - | - | | 3.0 | 138 | - | 0.5021 | 0.4964 | 0.4851 | 0.4572 | 0.3995 | | 3.0440 | 140 | 7.258 | - | - | - | - | - | | 3.2637 | 150 | 7.3585 | - | - | - | - | - | | 3.4835 | 160 | 7.5519 | - | - | - | - | - | | 3.7033 | 170 | 7.6819 | - | - | - | - | - | | 3.9231 | 180 | 7.3011 | - | - | - | - | - | | **4.0** | **184** | **-** | **0.5058** | **0.4973** | **0.4857** | **0.462** | **0.4015** | | 4.1319 | 190 | 7.4137 | - | - | - | - | - | | 4.3516 | 200 | 7.1914 | - | - | - | - | - | | 4.5714 | 210 | 7.38 | - | - | - | - | - | | 4.7912 | 220 | 7.3488 | - | - | - | - | - | | 4.9011 | 225 | - | 0.5035 | 0.4986 | 0.4859 | 0.4617 | 0.4031 | * The bold row denotes the saved checkpoint. ### Framework Versions - Python: 3.11.12 - Sentence Transformers: 4.0.2 - Transformers: 4.51.2 - PyTorch: 2.6.0+cu124 - Accelerate: 1.5.2 - Datasets: 3.5.0 - Tokenizers: 0.21.1 ## Citation ### BibTeX #### Sentence Transformers ```bibtex @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 ```bibtex @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 ```bibtex @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} } ```