samehkhattab/biomed-retrieval-phi-4-mini-base-checkpoint

SentenceTransformer based on microsoft/Phi-4-mini-instruct

This is a sentence-transformers model finetuned from microsoft/Phi-4-mini-instruct on the biomed_retrieval_dataset dataset. It maps sentences & paragraphs to a 5120-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: microsoft/Phi-4-mini-instruct
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 5120 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • biomed_retrieval_dataset

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: Phi3Model 
  (1): Pooling({'word_embedding_dimension': 5120, 'pooling_mode_cls_token': False, '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': True, '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("sentence_transformers_model_id")
# Run inference
sentences = [
    'Given a question, retrieve relevant Pubmed passages that answer the question: Do experimental pain models reveal no sex differences in pentazocine analgesia in humans?',
    'Accumulating evidence suggests that there are sex differences in analgesic responses to opioid agonists. Several studies using an oral surgery pain model have reported more robust analgesia to kappa-agonist-antagonists (e.g., pentazocine, nalbuphine, butorphanol) among women than among men. However, evidence of sex differences in kappa-agonist-antagonist effects from studies of experimentally induced pain in humans is lacking. Therefore, the analgesic effects of intravenous pentazocine (0.5 mg/kg) were determined in healthy women (n = 41) and men (n = 38) using three experimental pain models: heat pain, pressure pain, and ischemic pain. Each pain procedure was conducted before and after double-blind administration of both pentazocine and saline, which occurred on separate days in counterbalanced order. Compared with saline, pentazocine produced significant analgesic responses for all pain stimuli. However, no sex differences in pentazocine analgesia emerged. Effect sizes for the sex differences were computed; the magnitude of effects was small, and an equal number of measures showed greater analgesia in men than in women. Also, analgesic responses were not highly correlated across pain modalities, suggesting that different mechanisms may underlie analgesia for disparate types of pain',
    "The purpose of this study was to elucidate the relationship between thermal and mechanical sensation, as well as pain thresholds degrees and the dynamics of the TRPV1 level in almost healthy young males and females in the follicular and luteal phases of the OMC. We found gender differences for some pain sensation indices, taking into account OMC phases of females. Mechanical pain tolerance and heat pain thresholds were significantly higher in males compared with females in both phases of the OMC, also, mechanical pain, mechanical pressure, cold pain and heat sensation thresholds were insignificantly higher in males compared with females in follicular phase of the OMC and significantly higher - in luteal phase of the OMC. We haven't found any differences in cold sensation threshold between males and females in both phases of OMC. Moreover, we found significant gender and interphase differences in receptor protein TRPV1 level - the maximal level in females in luteal phase of the OMC, lower in males and minimal in females in follicular phase of the OMC. Worldwide, women account for approximately 51% of human immunodeficiency virus-1 (HIV) seropositive individuals. The prevalence of neuropathic pain among individuals with HIV and a lack of preclinical data characterizing sex differences prompted us to address this knowledge gap. C57BL/6 male and female mice received multiple intrathecal injections of HIV-glycoprotein 120 (gp120), followed by determination of mechanical allodynia and thermal hypersensitivity for four weeks. The influence of ovarian hormones in the gp120 pain model was evaluated by comparison of ovariectomized (OVX) mice versus sham control. We found that gp120-induced neuropathic pain-like behaviors are sex-dependent. Female mice showed both increased mechanical allodynia and increased cold sensitivity relative to their male counterparts. The OVX mice showed reduced pain sensitivity compared to sham, suggesting a role of the ovarian hormones in sex differences in pain sensitivity to gp120. Gp120-induced neuropathic pain caused a shift in estrous cycle toward the estrus phase. However, there is a lack of clear correlation between the estrous cycle and the development of neuropathic pain-like behaviors during the four week recording period. This data provided the first evidence for sex differences in a rodent model of HIV-related neuropathic pain, along with a potential role of ovarian hormones.",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 5120]

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

Evaluation

Metrics

Triplet

• Dataset: bmretriever
• Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.9805

Training Details

Training Dataset

biomed_retrieval_dataset

• Dataset: biomed_retrieval_dataset at dff25ba
• Size: 1,260,000 training samples
• Columns: anchor, positive, negative, and source
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative	source
  type	string	string	string	string
  details	min: 16 tokens mean: 29.68 tokens max: 512 tokens	min: 3 tokens mean: 199.28 tokens max: 512 tokens	min: 2 tokens mean: 319.38 tokens max: 512 tokens	min: 1 tokens mean: 2.3 tokens max: 4 tokens

• Samples:

  anchor	positive	negative	source
  Given a question, retrieve relevant passages that answer the question: when does the new season of real housewives of beverly hills air?	The Real Housewives of Beverly Hills (season 10) This is the latest accepted revision, reviewed on 27 July 2020. The tenth season of The Real Housewives of Beverly Hills, an American reality television series, is broadcast on Bravo. It premiered on April 15 2020, and is primarily filmed in Beverly Hills, California.	Seasons 1–4. The Real Housewives of Beverly Hills was announced in March 2010 as the sixth installment of The Real Housewives franchise. The first season premiered on October 14, 2010, and starred Kyle Richards, Adrienne Maloof, Kim Richards, Lisa Vanderpump, Camille Grammer and Taylor Armstrong. This is the latest accepted revision, reviewed on 27 July 2020. The tenth season of The Real Housewives of Beverly Hills, an American reality television series, is broadcast on Bravo. It premiered on April 15 2020, and is primarily filmed in Beverly Hills, California.	gooaq
  Given a question, retrieve Pubmed passages that answer the question: determinants of the pathway to emergency obstetric care in south africa	BACKGROUND: Maternity referral systems have been under-documented, under-researched, and under-theorised. Responsive emergency referral systems and appropriate transportation are cornerstones in the continuum of care and central to the complex health system. The pathways that women follow to reach Emergency Obstetric and Neonatal Care (EmONC) once a decision has been made to seek care have received relatively little attention. The aim of this research was to identify patterns and determinants of the pathways pregnant women follow from the onset of labour or complications until they reach an appropriate health facility.METHODS: This study was conducted in Renk County in South Sudan between 2010 and 2012. Data was collected using Critical Incident Technique (CIT) and stakeholder interviews. CIT systematically identified pathways to healthcare during labour, and factors associated with an event of maternal mortality or near miss through a series of in-depth interviews with witnesses or th...	OBJECTIVES: In a rural district hospital in Burundi offering Emergency Obstetric care-(EmOC), we assessed the a) characteristics of women at risk of, or with an obstetric complication and their types b) the number and type of obstetric surgical procedures and anaesthesia performed c) human resource cadres who performed surgery and anaesthesia and d) hospital exit outcomes.METHODS: A retrospective analysis of EmOC data (2011 and 2012).RESULTS: A total of 6084 women were referred for EmOC of whom 2534(42%) underwent a major surgical procedure while 1345(22%) required a minor procedure (36% women did not require any surgical procedure). All cases with uterine rupture(73) and extra-uterine pregnancy(10) and the majority with pre-uterine rupture and foetal distress required major surgery. The two most prevalent conditions requiring a minor surgical procedure were abortions (61%) and normal delivery (34%). A total of 2544 major procedures were performed on 2534 admitted individuals. Of these...	synthetic
  Given a question, retrieve Wikipedia passages that answer the question: hindi festival where sisters give bracelets to brothers	Raksha Bandhan Raksha Bandhan, also Rakshabandhan, or Rakhi, is a popular, traditionally Hindu, annual rite, or ceremony, which is central to a festival of the same name, originating from the Indian subcontinent, celebrated in parts of Indian subcontinent, and among people influenced by Hindu and Indian culture around the world. On this day, sisters of all ages tie a talisman, or amulet, called the ""rakhi"", around the wrists of their brothers, symbolically protecting them, receiving a gift in return, and traditionally investing the brothers with a share of the responsibility of their potential care. Raksha Bandhan is observed on the	Bonalu Bonalu or Goddess Mahankali bonalu (Telugu: బోనాలు ) is a Hindu Festival, Goddess Mahakali is worshiped. Bonalu is an annual festival of Telangana celebrated in Twin Cities Hyderabad, Secunderabad and other parts of Telangana. It is celebrated in the month of Ashada Masam, in July/August. Special poojas are performed for Yellamma on the first and last day of the festival. The festival is also considered a thanksgiving to the Goddess for fulfillment of vows. The word ""Bonam"" is a contraction of the word ""Bhojanam"", a Sanskrit loanword which means a meal or a feast in Telugu, is an ""Offering related to the sixth Guru, Guru Hargobind. According to Sikh history, on this day, Guru Hargobind was released from prison by the Mughal Emperor Jahangir who freed 52 other Hindu kings with him. The Bandi Chhor Divas is celebrated in a manner similar to Diwali, with the lighting of homes and Gurdwaras, feasts, gift giving and family time. It is an important Sikh celebration along with Vai...	nq

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Evaluation Dataset

biomed_retrieval_dataset

• Dataset: biomed_retrieval_dataset at dff25ba
• Size: 70,000 evaluation samples
• Columns: anchor, positive, negative, and source
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative	source
  type	string	string	string	string
  details	min: 16 tokens mean: 29.46 tokens max: 292 tokens	min: 3 tokens mean: 188.07 tokens max: 512 tokens	min: 3 tokens mean: 313.94 tokens max: 512 tokens	min: 1 tokens mean: 2.31 tokens max: 4 tokens

• Samples:

  anchor	positive	negative	source
  Given a question, retrieve Pubmed passages that answer the question: what is considered the reference gene for bone marrow osteoblasts?	Quantitative real-time polymerase chain reaction (qRT-PCR) is a powerful tool to evaluate gene expression, but its accuracy depends on the choice and stability of the reference genes used for normalization. In this study, we aimed to identify reference genes for studies on osteoblasts derived from rat bone marrow mesenchymal stem cells (bone marrow osteoblasts), osteoblasts derived from newborn rat calvarial (calvarial osteoblasts), and rat osteosarcoma cell line UMR-106. The osteoblast phenotype was characterized by ALP activity and extracellular matrix mineralization. Thirty-one candidates for reference genes from a Taqman array were assessed by qRT-PCR, and their expressions were analyzed by five different approaches. The data showed that several of the most traditional reference genes, such as Actb and Gapdh, were inadequate for normalization and that the experimental conditions may affect gene stability. Eif2b1 was frequently identified among the best reference genes in bone marro...	Fast progress of the next generation sequencing (NGS) technology has allowed global transcriptional profiling and genome-wide mapping of transcription factor binding sites in various cellular contexts. However, limited number of replicates and high amount of data processing may weaken the significance of the findings. Comparative analyses of independent data sets acquired in the different laboratories would greatly increase the validity of the data. Runx2 is the key transcription factor regulating osteoblast differentiation and bone formation. We performed a comparative analysis of three published Runx2 data sets of chromatin immunoprecipitation followed by deep sequencing (ChIP-seq) analysis in osteoblasts from mouse and human origin. Moreover, we assessed the similarity of the corresponding transcription data of these studies available online. The ChIP-seq data analysis confirmed general features of Runx2 binding, including location at genic vs intergenic regions and abundant Runx2 b...	synthetic
  Given a question, retrieve Pubmed passages that answer the question: what does gper1 do in the oviduct	Oviducts play roles in reproductive processes, including gametes transport, fertilization and early embryo development. Oviductal transport is controlled by various factors such as endothelins (EDNs) and nitric oxide (NO), smooth muscle contracting and relaxing factor, respectively. EDNs and NO production depend on an ovarian steroid hormone, oestradiol-17 (E2) and E2 quickly exerts their biological functions through G protein-coupled oestrogen receptor 1 (GPER1), which mediates rapid intracellular signalling. Because follicular fluid which contains a high concentration of E2 enters the oviduct, we hypothesized that E2 in the follicular fluid participates via GPER1 in producing EDNs and NO. To test this hypothesis, we investigated 1) the expression and localization of GPER1 in bovine oviductal tissues and 2) rapid effects of E2 via GPER1 on EDN1, EDN2 and inducible NO synthase (iNOS) expression in cultured bovine oviductal isthmic epithelial cells. GPER1 was observed in the oviductal e...	BACKGROUND: The G protein estrogen receptor GPER/GPR30 mediates estrogen action in breast cancer cells as well as in breast cancer-associated fibroblasts (CAFs), which are key components of microenvironment driving tumor progression. GPER is a transcriptional target of hypoxia inducible factor 1 alpha (HIF-1) and activates VEGF expression and angiogenesis in hypoxic breast tumor microenvironment. Furthermore, IGF1/IGF1R signaling, which has angiogenic effects, has been shown to activate GPER in breast cancer cells.METHODS: We analyzed gene expression data from published studies representing almost 5000 breast cancer patients to investigate whether GPER and IGF1 signaling establish an angiocrine gene signature in breast cancer patients. Next, we used GPER-positive but estrogen receptor (ER)-negative primary CAF cells derived from patient breast tumours and SKBR3 breast cancer cells to investigate the role of GPER in the regulation of VEGF expression and angiogenesis triggered by IGF1. W...	synthetic
  Given a question, retrieve Pubmed passages that answer the question: what are ovine hair follicle stem cells	Hair follicle stem cells (HFSCs) possess fascinating self-renewal capacity and multipotency, which play important roles in mammalian hair growth and skin wound repair. Although HFSCs from other mammalian species have been obtained, the characteristics of ovine HFSCs, as well as the methods to isolate them have not been well addressed. Here, we report an efficient strategy to obtain multipotent ovine HFSCs. Through microdissection and organ culture, we obtained keratinocytes that grew from the bulge area of vibrissa hair follicles, and even abundant keratinocytes were harvested from a single hair follicle. These bulge-derived keratinocytes are highly positive for Krt15, Krt14, Tp63, Krt19 and Itga6; in addition to their strong proliferation abilities in vitro, these keratinocytes formed new epidermis, hair follicles and sebaceous glands in skin reconstitution experiments, showing that these are HFSCs from the bulge outer root sheath. Taken together, we developed an efficient in vitro sy...	Hair differentiates from follicle stem cells through progenitor cells in the matrix. In contrast to stem cells in the bulge, the identities of the progenitors and the mechanisms by which they regulate hair shaft components are poorly understood. Hair is also pigmented by melanocytes in the follicle. However, the niche that regulates follicular melanocytes is not well characterized. Here, we report the identification of hair shaft progenitors in the matrix that are differentiated from follicular epithelial cells expressing transcription factor KROX20. Depletion of Krox20 lineage cells results in arrest of hair growth, confirming the critical role of KROX20+ cells as antecedents of structural cells found in hair. Expression of stem cell factor (SCF) by these cells is necessary for the maintenance of differentiated melanocytes and for hair pigmentation. Our findings reveal the identities of hair matrix progenitors that regulate hair growth and pigmentation, partly by creating an SCF-depen...	synthetic

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• gradient_accumulation_steps: 2
• learning_rate: 2e-05
• num_train_epochs: 1
• warmup_steps: 100
• bf16: True
• dataloader_drop_last: True
• optim: adamw_bnb_8bit
• ddp_find_unused_parameters: False
• gradient_checkpointing: True
• gradient_checkpointing_kwargs: {'use_reentrant': False}
• use_liger_kernel: 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: 64
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 2
• 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: 1
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.0
• warmup_steps: 100
• 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
• bf16: True
• 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: True
• 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}
• 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}
• parallelism_config: None
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_bnb_8bit
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• project: huggingface
• trackio_space_id: trackio
• ddp_find_unused_parameters: False
• 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
• hub_revision: None
• gradient_checkpointing: True
• gradient_checkpointing_kwargs: {'use_reentrant': False}
• 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: no
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: True
• liger_kernel_config: None
• eval_use_gather_object: False
• average_tokens_across_devices: True
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss	Validation Loss	bmretriever_cosine_accuracy
0.0051	50	6.2071	-	-
0.0102	100	3.1273	-	-
0.0152	150	0.4119	-	-
0.0203	200	0.1551	-	-
0.0254	250	0.1219	-	-
0.0305	300	0.1021	-	-
0.0356	350	0.0895	-	-
0.0406	400	0.0854	-	-
0.0457	450	0.0729	-	-
0.0508	500	0.068	-	-
0.0559	550	0.07	-	-
0.0610	600	0.0669	-	-
0.0660	650	0.0632	-	-
0.0711	700	0.0651	-	-
0.0762	750	0.0638	-	-
0.0813	800	0.0569	-	-
0.0864	850	0.059	-	-
0.0914	900	0.061	-	-
0.0965	950	0.0546	-	-
0.1016	1000	0.0556	-	-
0.1067	1050	0.0499	-	-
0.1117	1100	0.054	-	-
0.1168	1150	0.0496	-	-
0.1219	1200	0.0539	-	-
0.1270	1250	0.0506	-	-
0.1321	1300	0.0488	-	-
0.1371	1350	0.0509	-	-
0.1422	1400	0.0465	-	-
0.1473	1450	0.047	-	-
0.1524	1500	0.0456	-	-
0.1575	1550	0.043	-	-
0.1625	1600	0.0479	-	-
0.1676	1650	0.05	-	-
0.1727	1700	0.0465	-	-
0.1778	1750	0.0448	-	-
0.1829	1800	0.041	-	-
0.1879	1850	0.047	-	-
0.1930	1900	0.0446	-	-
0.1981	1950	0.0433	-	-
0.2032	2000	0.0429	-	-
0.2083	2050	0.0434	-	-
0.2133	2100	0.0453	-	-
0.2184	2150	0.038	-	-
0.2235	2200	0.04	-	-
0.2286	2250	0.0417	-	-
0.2337	2300	0.0453	-	-
0.2387	2350	0.0349	-	-
0.2438	2400	0.0411	-	-
0.2489	2450	0.0426	-	-
0.2540	2500	0.0417	-	-
0.2591	2550	0.0377	-	-
0.2641	2600	0.0407	-	-
0.2692	2650	0.0353	-	-
0.2743	2700	0.0388	-	-
0.2794	2750	0.0395	-	-
0.2845	2800	0.0362	-	-
0.2895	2850	0.0387	-	-
0.2946	2900	0.0392	-	-
0.2997	2950	0.0354	-	-
0.3048	3000	0.0337	-	-
0.3098	3050	0.0348	-	-
0.3149	3100	0.0335	-	-
0.3200	3150	0.0353	-	-
0.3251	3200	0.0362	-	-
0.3302	3250	0.0362	-	-
0.3352	3300	0.0357	-	-
0.3403	3350	0.045	-	-
0.3454	3400	0.0421	-	-
0.3505	3450	0.033	-	-
0.3556	3500	0.0367	-	-
0.3606	3550	0.0363	-	-
0.3657	3600	0.0353	-	-
0.3708	3650	0.0295	-	-
0.3759	3700	0.0375	-	-
0.3810	3750	0.0347	-	-
0.3860	3800	0.0338	-	-
0.3911	3850	0.0313	-	-
0.3962	3900	0.0341	-	-
0.4013	3950	0.042	-	-
0.4064	4000	0.0372	-	-
0.4114	4050	0.0333	-	-
0.4165	4100	0.0325	-	-
0.4216	4150	0.0296	-	-
0.4267	4200	0.0307	-	-
0.4318	4250	0.0315	-	-
0.4368	4300	0.0351	-	-
0.4419	4350	0.0343	-	-
0.4470	4400	0.0336	-	-
0.4521	4450	0.0285	-	-
0.4572	4500	0.0334	-	-
0.4622	4550	0.0356	-	-
0.4673	4600	0.0295	-	-
0.4724	4650	0.0343	-	-
0.4775	4700	0.0292	-	-
0.4826	4750	0.0341	-	-
0.4876	4800	0.0332	-	-
0.4927	4850	0.0312	-	-
0.4978	4900	0.0298	-	-
0.5029	4950	0.0306	-	-
0.5079	5000	0.0276	-	-
0.5130	5050	0.0287	-	-
0.5181	5100	0.0345	-	-
0.5232	5150	0.0298	-	-
0.5283	5200	0.0298	-	-
0.5333	5250	0.0331	-	-
0.5384	5300	0.0288	-	-
0.5435	5350	0.0299	-	-
0.5486	5400	0.0363	-	-
0.5537	5450	0.0314	-	-
0.5587	5500	0.0296	-	-
0.5638	5550	0.0271	-	-
0.5689	5600	0.0304	-	-
0.5740	5650	0.0252	-	-
0.5791	5700	0.0266	-	-
0.5841	5750	0.0247	-	-
0.5892	5800	0.0278	-	-
0.5943	5850	0.0315	-	-
0.5994	5900	0.0285	-	-
0.6045	5950	0.0243	-	-
0.6095	6000	0.0287	-	-
0.6146	6050	0.0287	-	-
0.6197	6100	0.0311	-	-
0.6248	6150	0.0293	-	-
0.6299	6200	0.0323	-	-
0.6349	6250	0.0266	-	-
0.6400	6300	0.0289	-	-
0.6451	6350	0.0276	-	-
0.6502	6400	0.0326	-	-
0.6553	6450	0.0257	-	-
0.6603	6500	0.0264	-	-
0.6654	6550	0.0298	-	-
0.6705	6600	0.0264	-	-
0.6756	6650	0.0306	-	-
0.6807	6700	0.0256	-	-
0.6857	6750	0.0294	-	-
0.6908	6800	0.0263	-	-
0.6959	6850	0.0253	-	-
0.7010	6900	0.027	-	-
0.7060	6950	0.031	-	-
0.7111	7000	0.0257	-	-
0.7162	7050	0.0266	-	-
0.7213	7100	0.0282	-	-
0.7264	7150	0.0262	-	-
0.7314	7200	0.0244	-	-
0.7365	7250	0.0263	-	-
0.7416	7300	0.0269	-	-
0.7467	7350	0.0284	-	-
0.7518	7400	0.029	-	-
0.7568	7450	0.0311	-	-
0.7619	7500	0.0271	-	-
0.7670	7550	0.0267	-	-
0.7721	7600	0.0273	-	-
0.7772	7650	0.0318	-	-
0.7822	7700	0.0252	-	-
0.7873	7750	0.0277	-	-
0.7924	7800	0.0314	-	-
0.7975	7850	0.0262	-	-
0.8026	7900	0.022	-	-
0.8076	7950	0.025	-	-
0.8127	8000	0.0302	-	-
0.8178	8050	0.0262	-	-
0.8229	8100	0.0283	-	-
0.8280	8150	0.0288	-	-
0.8330	8200	0.0313	-	-
0.8381	8250	0.0302	-	-
0.8432	8300	0.0276	-	-
0.8483	8350	0.0265	-	-
0.8534	8400	0.0246	-	-
0.8584	8450	0.028	-	-
0.8635	8500	0.0243	-	-
0.8686	8550	0.0244	-	-
0.8737	8600	0.0248	-	-
0.8788	8650	0.0258	-	-
0.8838	8700	0.0273	-	-
0.8889	8750	0.0241	-	-
0.8940	8800	0.0299	-	-
0.8991	8850	0.0286	-	-
0.9041	8900	0.0273	-	-
0.9092	8950	0.0246	-	-
0.9143	9000	0.0277	-	-
0.9194	9050	0.0267	-	-
0.9245	9100	0.0281	-	-
0.9295	9150	0.0255	-	-
0.9346	9200	0.0324	-	-
0.9397	9250	0.0236	-	-
0.9448	9300	0.0244	-	-
0.9499	9350	0.0312	-	-
0.9549	9400	0.0253	-	-
0.9600	9450	0.0321	-	-
0.9651	9500	0.0206	-	-
0.9702	9550	0.0329	-	-
0.9753	9600	0.0289	-	-
0.9803	9650	0.0258	-	-
0.9854	9700	0.0239	-	-
0.9905	9750	0.0216	-	-
0.9956	9800	0.0265	-	-
1.0	9844	-	0.0259	0.9805

Framework Versions

• Python: 3.11.9
• Sentence Transformers: 4.1.0
• Transformers: 4.57.1
• PyTorch: 2.6.0+cu124
• Accelerate: 1.6.0
• Datasets: 2.21.0
• Tokenizers: 0.22.1

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",
}

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}
}