SentenceTransformer based on unsloth/embeddinggemma-2

This is a sentence-transformers model finetuned from unsloth/embeddinggemma-2 on the generator dataset. It maps inputs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: unsloth/embeddinggemma-2
• Maximum Sequence Length: 1024 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity
• Supported Modalities: Text, Image, Audio, Video, Message
• Training Dataset:
  • generator

Model Sources

• Documentation: Sentence Transformers Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}, 'image': {'method': 'forward', 'method_output_name': 'last_hidden_state'}, 'audio': {'method': 'forward', 'method_output_name': 'last_hidden_state'}, 'video': {'method': 'forward', 'method_output_name': 'last_hidden_state'}, 'message': {'method': 'forward', 'method_output_name': 'last_hidden_state', 'format': 'structured'}}, 'module_output_name': 'token_embeddings', 'max_seq_length': 1024, 'do_lower_case': False, 'architecture': 'PeftModelForFeatureExtraction'})
  (1): Pooling({'embedding_dimension': 768, 'pooling_mode': 'mean', 'include_prompt': True})
  (2): Normalize({'module_input_name': 'sentence_embedding', 'module_output_name': 'sentence_embedding'})
)

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("faizaltkl/medical-gemma-300m-emb")
# Run inference
queries = [
    'What are some hereditary cancer syndromes that can result in various forms of cancer?\n',
]
documents = [
    'Hereditary Cancer Syndromes, including Hereditary Breast and Ovarian Cancer (HBOC) and Lynch Syndrome (LS), can result in various forms of cancer due to germline mutations in cancer predisposition genes. While the major contributory genes for these syndromes have been identified and well-studied (BRCA1/ BRCA2 for HBOC and MSH2/MSH6/MLH1/PMS2/ EPCAM for LS), there remains a large percentage of associated cancer cases that are negative for germline mutations in these genes, including 80% of women with a personal or family history of breast cancer who are negative for BRCA1/2 mutations [1] . Similarly, between 30 and 50% of families fulfill stringent criteria for LS and test negative for germline mismatch repair gene mutations [2] . Adding complexity to these disorders is the significant overlap in the spectrum of cancers observed between various hereditary cancer syndromes, including many cancer susceptibility syndromes. Some that contribute to elevated breast cancer risk include Li-Fraumeni Syndrome, Cowden syndrome, Hereditary Diffuse Gastric Cancer, Ataxia-telangiectasia, Lynch Syndrome, and Peutz-Jeghers syndrome [3] , while others that contribute to elevated colorectal cancer risk include Familial Adenomatous Polyposis and MUTYH-associated Polyposis [4] . Additionally, the risk of cancers observed within these syndromes can vary widely. HBOC not only increases the life-time risk of breast and ovarian cancers, but also for prostate and pancreatic cancers [3] . Similarly, LS-spectrum cancers include colorectal cancer, as well as uterine (endometrial), ovarian, gastric, and other rare forms of cancer [5] .\n\n Over the past decade, linkage, GWAS, and resequencing studies have identified additional genes that contribute to the risk of hereditary cancer [6] . The discoveries of these genes, which vary in their level of penetrance and risk-contribution, present opportunities for personalized clinical management in patients and families with hereditary cancer.\n\n Over time, multi-gene panels for hereditary cancer continue to grow in complexity, from sequencing only the major contributory genes for HBOC and LS, a total of roughly six genes, now expanding to 27 genes and higher [2, [7] [8] [9] [10] [11] [12] [13] [14] . The addition of less characterized genes to hereditary cancer panels poses challenges in interpretation due to lack of available information such as disease prevalence, penetrance, locus/allelic heterogeneity, expressivity, age of onset, and the spectrum of gene-associated cancers. Three recent studies reporting the clinical outcomes of panel-based testing for inherited cancer have utilized cohorts selected for patients fulfilling the National Comprehensive Cancer Network (NCCN) guidelines for breast cancer [7] or colorectal cancer [2] , or Society of Gynecologic Oncology 5-10% criteria for endometrial cancer [9] . One of the studies used retrospective history-enriched cases from clinical bio banks with referrals based on known familial pathogenic variants [7] . In this study, we report our experience with a 27-gene inherited cancer panel on a cohort of 630 consecutive patients referred for testing at our laboratory. Our objectives were: 1. Determine the rates for positive cases and those with variants of uncertain clinical significance (VUS) relative to data published in the recent literature, 2. Examine heterogeneity among the constituent genes on the panel, and 3. Review test uptake in the cohort relative to other reports describing outcomes for expanded panel testing.\n\n \n\n Clinical and genomic data were reviewed for the first 630 consecutive individuals referred for hereditary cancer gene panel testing at our clinical laboratory improvement amendments (CLIA)-approved commercial laboratory (Integrated Genetics, Laboratory Corporation of America® Holdings, Research Triangle Park, NC). Relevant information including the clinical indication for testing, family history, and demographic data were obtained by review of the ordering test requisition forms. For clinical diagnostic testing, it is standard for the referring physician to obtain informed consent prior to test ordering; therefore, an ethics approval was not required. Individuals referred for testing were not restricted to any specific center or region and spanned a broad referral pattern. Patients were stratified based on personal and family history of HBOC-associated cancers (breast, ovarian, prostate) as well as LS-associated cancers (colorectal, gastric, pancreatic, ovarian, uterine/endometrial, and bladder). As ovarian cancer is a common clinical feature of both HBOC and LS, patients with ovarian cancer were included in both these categories.\n\n Genomic DNA was isolated from blood specimens using Chemagic DNA Blood Kit Special (PerkinElmer, Shelton, CT).',
    'Hybrid F1 mice were injected with urethane (300 mg/kg) at 8 days of age. A group was then put on a MAK-supplemented diet, another group was fed a standard pellet diet. At 36 weeks of age the mice were sacrificed and the livers examined for the presence of tumors mouse (Panel A) and for the number of nodules per mouse (Panel B) (* p < 0.05, ** P < 0.001). Statistical analysis was performed by Two Way ANOVA Test followed by Post Hoc Bonferroni analysis. \n\n We than measured the influence of the MAK-4+5 combination on the expression of the three liver-specific connexins (cx26, cx32, and cx43). The level of cx26 expression was similar in all the groups of mice treated with the MAK-supplemented diet and in the control (Figure 4, Panel A) . A significant, time-dependent increase in cx32 was observed in the liver of all the groups of MAK treated mice compared to the normal diet-fed controls. Cx32 expression increased 2-fold after 1 week of treatment, and 3-to 4-fold at 3 months (Figure 4, Panel B) . The anti cx32 antibody reacted with two antigens with molecular masses of 27 kDa and 48 kDa. The 48 kD band was identified as a stable dimeric form of cx32, as previously reported [49] .\n\n The present study reveals the ability of MAK-4 and MAK-5 to inhibit cell transformation induced by ras oncogene in an in vitro cell system and in vivo liver carcinogenesis in mice.\n\n The group of mice fed a diet supplemented with a combination of MAK-4 and MAK-5, had a 46% reduction in the number of liver nodules, compared to mice that were fed a normal diet. Among the tumor-positive mice, the number of tumor nodules was significantly lower in the MAK fed mice. The ability of MAK-4 and MAK-5 to inhibit carcinogenesis may be due to several different mechanisms that may be difficult to identify. The complex mixture of these preparations will not allow the identification of effects due to single components, until extract fractionation is performed. Here we have investigated the ability of MAK-4 and MAK-5 to scavenge free radicals and reactive oxygen species (ROS) and regulate connexins, which are markers of tissue integrity and tumor formation. Free radical-mediated reactions such as lipid peroxidation are implicated in carcinogenesis [21] and there is increasing evidence that the initiation and promotion phases of cancer is favoured by free radical reactions [22] [23] [24] . Thus, antioxidants which act to control the oxidative state may represent a major line of defense against cancer development. Multiple antioxidants are present in MAK-4 and MAK-5 such as alpha-tocopherol, beta-carotene, ascorbate, bioflavonoid, catechin, polyphenols, riboflavin, and tannic acid and many others that are still uncharacterized [25] . A Previous research has shown that these mixtures of chemicals scavenge free radicals and ROS such as superoxide, hydroxyl, and peroxyl radicals, and hydrogen peroxide generated in both cellular (neutrophil) and noncellular (xanthine-xanthine oxidase) systems [50] , in a dose-dependent manner. MAK-4 and MAK-5 inhibit the in vitro peroxidation of lipids in rat liver microsomes [51, 52] , reduce lipid peroxide levels and inhibit the oxidation of low-density lipoprotein (LDL) [53] [54] [55] .\n\n In the present work, the antioxidant properties of MAK-4 and MAK-5 were confirmed in vivo in mice. The oxygen radical absorbance capacity (ORAC) assay measured a significantly higher value in mice fed the MAK supplemented diet, which increased progressively from day 7 and reached a maximum level at 1 month that was maintained throughout the course of the experiment.\n\n Another line of antioxidant defense was evaluated by measuring the activity of the liver enzymes GPX, GST and QR. All these enzymes were upregulated in the MAK-fed mice. It is known that many compounds that block toxic, mutagenic, and neoplastic effects of carcinogens have in common the ability to elevate levels of phase II detoxification enzymes such as GST and QR [56] . Inducers of phase II enzymes, present in fruits and vegetables, have been shown to prevent carcinogenesis [57] . Substantial evidence has accumulated to suggest that the induction of phase II enzymes is a causal mechanism for protection since these enzymes divert ultimate carcinogens from reacting with critical cellular macromolecules [57] .',
    'Decompressive craniectomy is a valid neurosurgical strategy now a day as an alternative to control an elevated intracranial pressure (ICP) and controlling the risk of uncal and/or subfalcine herniation, in refractory cases to the postural, ventilator, and pharmacological measures to control it. The neurocritical care and the ICP monitorization are key determinants to identify and postulate the inclusion criteria to consider a patient as candidate to this procedure, as it is always considered a rescue surgical technique. Head trauma and ischemic or hemorrhagic cerebrovascular disease with progressive deterioration due to mass effect are some of the cases that may require a decompressive craniectomy with its different variants. However, this procedure per se can have complications described in the postcraniectomy syndrome and may occur in short, medium, or even long term.\n\n 1,2 The paradoxical herniation is a condition in which there is a deviation of the midline with mass effect, even though the skull is not covered, which clinically relates to the sinking of the skin flap and the intracranial pressure changes that are more susceptible to the posture and the effect of atmospheric pressure. This herniation pattern is much more evident and susceptible to deterioration in cases where a ventricular shunt has also been installed, which contributes to a ventricular collapse and modification and alteration of the Monro-Kellie law, with the consequent maladjustment and neurological deterioration due to the paradoxical herniation.\n\n ► cerebral edema ► cranioplasty ► decompressive craniectomy ► paradoxical herniation ► sinking skin flap syndrome ► syndrome of the trephined\n\n Introduction The decompressive craniectomy is a surgical strategy widely used with specific criteria to control the refractory intracranial pressure (ICP). However, it is important to warn about the presence of a postcraniectomy syndrome and analyze the risk-benefit on a long term. Case Report A 72-year-old male patient diagnosed with a subarachnoid hemorrhage secondary to the rupture of an anterior circulation aneurysm that develops vasospasm, secondary ischemia, and edema with signs of herniation that required a decompressive craniectomy on a first step. Afterwards, the aneurysm was approached and he consequently developed hydrocephaly. A ventriculoperitoneal shunt is installed, contralateral to the craniectomy, and progressive sinking of the skin flap, there is neurological deterioration and paradoxical herniation. Its association with the clinical deterioration by bronchoaspiration did not allow the cranioplasty to resolve the ICP decompensation.\n\n Conclusions The paradoxical herniation as part of the postcraniectomy syndrome is an increasingly common condition identified in adult patients with cortical atrophy, and who have also been treated with ventricular shunt systems. Timely cranioplasty represents the ideal therapeutic plan once the compromise from the mass effect has resolved to avoid complications derived from the decompressive craniectomy per se.\n\n This condition requires prompt treatment by cranioplasty and should be considered as part of the postcraniectomy syndrome, as it may be responsible for the neurological impairment and/or sudden death.\n\n Case Report\n\n A 72-year-old male patient that presents headache, diminished state of consciousness, and left hemiparesis. A subarachnoid hemorrhage is identified secondary to a ruptured supraclinoid internal carotid aneurysm with associated ischemia in the right medial cerebral artery territory and cerebral edema. Decompressive craniectomy was performed, and in a second intention the clipping of the aneurysm. After 2 weeks, neurological deterioration is observed due to secondary hydrocephaly that required a ventriculoperitoneal shunt. His neurological condition persisted as grade IV in the Hunt and Hess scale and a gastrostomy was installed for feeding 10 days after installing the ventriculoperitoneal shunt, all of this done in a public sector hospital. Afterwards, gradual sinking of the skin flap was identified in the cranial defect and new neurological deterioration with a comatose patient and rigidity of decerebration, critical condition that did not allow the cranioplasty to be performed, due to the evidence of paradoxical herniation. The patient died from sudden death (►Fig. 1).\n\n The present report covers the ethical lineages and the respective authorization from the bioethics committee, as well as the identity protection lineages and data confidentiality.\n\n The decompressive craniectomy was systematically described by Cushing as one of the surgical strategies to diminish the elevated intracranial pressure.',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 768] [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[0.6606, 0.3038, 0.0792]])

Evaluation

Metrics

Information Retrieval

• Datasets: miriad, NanoNFCorpus, NanoSciFact, NanoFiQA2018 and NanoArguAna
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "query_prompt": "task: search result | query: ",
      "corpus_prompt": "title: none | text: "
  }
  

Metric	miriad	NanoNFCorpus	NanoSciFact	NanoFiQA2018	NanoArguAna
cosine_accuracy@1	0.8943	0.46	0.78	0.4	0.14
cosine_accuracy@3	0.958	0.54	0.88	0.46	0.7
cosine_accuracy@5	0.975	0.62	0.9	0.56	0.84
cosine_accuracy@10	0.988	0.64	0.92	0.62	0.94
cosine_precision@1	0.8943	0.46	0.78	0.4	0.14
cosine_precision@3	0.3193	0.3733	0.3267	0.2067	0.2333
cosine_precision@5	0.195	0.344	0.204	0.16	0.168
cosine_precision@10	0.0988	0.282	0.104	0.1	0.094
cosine_recall@1	0.8943	0.0437	0.735	0.2114	0.14
cosine_recall@3	0.958	0.0682	0.875	0.2969	0.7

Nano BEIR

• Dataset: NanoBEIR_mean
• Evaluated with NanoBEIREvaluator with these parameters:

  {
      "dataset_names": [
          "nfcorpus",
          "scifact",
          "fiqa2018",
          "arguana"
      ],
      "dataset_id": "sentence-transformers/NanoBEIR-en",
      "query_prompts": {
          "nfcorpus": "task: search result | query: ",
          "scifact": "task: search result | query: ",
          "fiqa2018": "task: search result | query: ",
          "arguana": "task: search result | query: "
      },
      "corpus_prompts": {
          "nfcorpus": "title: none | text: ",
          "scifact": "title: none | text: ",
          "fiqa2018": "title: none | text: ",
          "arguana": "title: none | text: "
      }
  }
  

Metric	Value
cosine_accuracy@1	0.445
cosine_accuracy@3	0.645
cosine_accuracy@5	0.73
cosine_accuracy@10	0.78
cosine_precision@1	0.445
cosine_precision@3	0.285
cosine_precision@5	0.219
cosine_precision@10	0.145
cosine_recall@1	0.2825
cosine_recall@3	0.485
cosine_recall@5	0.5529
cosine_recall@10	0.6064
cosine_ndcg@10	0.5332
cosine_mrr@10	0.5599
cosine_map@100	0.4382

Training Details

Training Dataset

generator

• Dataset: generator
• Size: 14,000 training samples
• Columns: question and passage_text
• Approximate statistics based on the first 100 samples:

  	question	passage_text
  type	string	string
  modality	text	text
  details	min: 10 tokensmean: 21.83 tokensmax: 60 tokens	min: 529 tokensmean: 942.59 tokensmax: 1024 tokens

• Samples:

  question	passage_text
  What factors may contribute to increased pulmonary conduit durability in patients who undergo the Ross operation compared to those with right ventricular outflow tract obstruction?	I n 1966, Ross and Somerville 1 reported the first use of an aortic homograft to establish right ventricle-to-pulmonary artery continuity in a patient with tetralogy of Fallot and pulmonary atresia. Since that time, pulmonary position homografts have been used in a variety of right-sided congenital heart lesions. Actuarial 5-year homograft survivals for cryopreserved homografts are reported to range between 55% and 94%, with the shortest durability noted in patients less than 2 years of age. 4 Pulmonary position homografts also are used to replace pulmonary autografts explanted to repair left-sided outflow disease (the Ross operation). Several factors may be likely to favor increased pulmonary conduit durability in Ross patients compared with those with right ventricular outflow tract obstruction, including later age at operation (allowing for larger homografts), more normal pulmonary artery architecture, absence of severe right ventricular hypertrophy, and more natural positioning of ...
  How does MCAM expression in hMSC affect the growth and maintenance of hematopoietic progenitors?	After culture in a 3-dimensional hydrogel-based matrix, which constitutes hypoxic conditions, MCAM expression is lost. Concordantly, Tormin et al. demonstrated that MCAM is down-regulated under hypoxic conditions. 10 Furthermore, it was shown by others and our group that oxygen tension causes selective modification of hematopoietic cell and mesenchymal stromal cell interactions in co-culture systems as well as influence HSPC metabolism. [44] [45] [46] Thus, the observed differences between Sharma et al. and our data in HSPC supporting capacity of hMSC are likely due to the different culture conditions used. Further studies are required to clarify the influence of hypoxia in our model system. Altogether these findings provide further evidence for the importance of MCAM in supporting HSPC. Furthermore, previous reports have shown that MCAM is down-regulated in MSC after several passages as well as during aging and differentiation. 19, 47 Interestingly, MCAM overexpression in hMSC enhance...

Evaluation Dataset

generator

• Dataset: generator
• Size: 3,000 evaluation samples
• Columns: question and passage_text
• Approximate statistics based on the first 100 samples:

  	question	passage_text
  type	string	string
  modality	text	text
  details	min: 12 tokensmean: 22.39 tokensmax: 56 tokens	min: 525 tokensmean: 947.01 tokensmax: 1024 tokens

• Samples:

  question	passage_text
  What are some hereditary cancer syndromes that can result in various forms of cancer?	Hereditary Cancer Syndromes, including Hereditary Breast and Ovarian Cancer (HBOC) and Lynch Syndrome (LS), can result in various forms of cancer due to germline mutations in cancer predisposition genes. While the major contributory genes for these syndromes have been identified and well-studied (BRCA1/ BRCA2 for HBOC and MSH2/MSH6/MLH1/PMS2/ EPCAM for LS), there remains a large percentage of associated cancer cases that are negative for germline mutations in these genes, including 80% of women with a personal or family history of breast cancer who are negative for BRCA1/2 mutations [1] . Similarly, between 30 and 50% of families fulfill stringent criteria for LS and test negative for germline mismatch repair gene mutations [2] . Adding complexity to these disorders is the significant overlap in the spectrum of cancers observed between various hereditary cancer syndromes, including many cancer susceptibility syndromes. Some that contribute to elevated breast cancer risk include Li-Frau...
  How do MAK-4 and MAK-5 exert their antioxidant properties?	Hybrid F1 mice were injected with urethane (300 mg/kg) at 8 days of age. A group was then put on a MAK-supplemented diet, another group was fed a standard pellet diet. At 36 weeks of age the mice were sacrificed and the livers examined for the presence of tumors mouse (Panel A) and for the number of nodules per mouse (Panel B) (* p < 0.05, ** P < 0.001). Statistical analysis was performed by Two Way ANOVA Test followed by Post Hoc Bonferroni analysis. We than measured the influence of the MAK-4+5 combination on the expression of the three liver-specific connexins (cx26, cx32, and cx43). The level of cx26 expression was similar in all the groups of mice treated with the MAK-supplemented diet and in the control (Figure 4, Panel A) . A significant, time-dependent increase in cx32 was observed in the liver of all the groups of MAK treated mice compared to the normal diet-fed controls. Cx32 expression increased 2-fold after 1 week of treatment, and 3-to 4-fold at 3 months (Figure 4, Pane...

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 64
• max_steps: 200
• learning_rate: 2e-05
• warmup_steps: 0.03
• gradient_accumulation_steps: 2
• bf16: True
• per_device_eval_batch_size: 64
• prompts: {'question': 'task: search result | query: ', 'passage_text': 'title: none | text: '}
• batch_sampler: no_duplicates

All Hyperparameters

• per_device_train_batch_size: 64
• num_train_epochs: 3.0
• max_steps: 200
• learning_rate: 2e-05
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_steps: 0.03
• optim: adamw_torch_fused
• optim_args: None
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• optim_target_modules: None
• gradient_accumulation_steps: 2
• average_tokens_across_devices: True
• max_grad_norm: 1.0
• label_smoothing_factor: 0.0
• bf16: True
• fp16: False
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• use_liger_kernel: False
• liger_kernel_config: None
• use_cache: False
• neftune_noise_alpha: None
• torch_empty_cache_steps: None
• auto_find_batch_size: False
• log_on_each_node: True
• logging_nan_inf_filter: True
• include_num_input_tokens_seen: no
• log_level: passive
• log_level_replica: warning
• disable_tqdm: False
• project: huggingface
• trackio_space_id: None
• trackio_bucket_id: None
• trackio_static_space_id: None
• per_device_eval_batch_size: 64
• : True

Training Logs

Epoch	Step	Training Loss	miriad_cosine_ndcg@10	NanoNFCorpus_cosine_ndcg@10	NanoSciFact_cosine_ndcg@10	NanoFiQA2018_cosine_ndcg@10	NanoArguAna_cosine_ndcg@10	NanoBEIR_mean_cosine_ndcg@10
-1	-1	-	0.9293	0.3803	0.8822	0.4453	0.6444	0.5881
0.0457	5	0.7016	-	-	-	-	-	-
0.0913	10	0.6135	-	-	-	-	-	-
0.1370	15	0.5340	-	-	-	-	-	-
0.1826	20	0.3495	-	-	-	-	-	-
0.2283	25	0.2576	-	-	-	-	-	-
0.2740	30	0.1905	-	-

Training Time

• Training: 1.5 hours

Framework Versions

• Python: 3.12.3
• Sentence Transformers: 6.1.0
• Transformers: 5.19.0.dev0
• PyTorch: 2.14.1+cu130
• Accelerate: 1.15.0
• Datasets: 4.8.5
• Tokenizers: 0.23.3

Additional Resources

• Training and Finetuning Embedding Models with Sentence Transformers: the end-to-end guide for training or finetuning Sentence Transformer models.
• Introduction to Matryoshka Embedding Models: variable-size embeddings that can be truncated with minimal quality loss.
• Binary and Scalar Embedding Quantization for Significantly Faster & Cheaper Retrieval: post-training compression of embedding vectors.
• Multimodal Embedding & Reranker Models with Sentence Transformers: use text, image, audio, and video models through the same API.
• Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers: train multimodal embedding models, with a Visual Document Retrieval walkthrough.

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{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}