metadata

datasets: []
language:
  - en
library_name: sentence-transformers
license: apache-2.0
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
pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:99145
  - loss:MatryoshkaLoss
  - loss:MultipleNegativesRankingLoss
widget:
  - source_sentence: >-
      YouTube provides people with entertainment, information, and opportunities
      to learn something new. Google Assistant 

      offers the best way to get things done seamlessly across different
      devices, providing intelligent help throughout a 

      person's day, no matter where they are. Google Cloud helps customers solve
      today’s business challenges, improve 

      productivity, reduce costs, and unlock new growth engines. We are
      continually innovating and building new products 

      and features that will help our users, partners, customers, and
      communities and have invested more than $150 billion 

      in research and development in the last five years in support of these
      efforts .

      Making AI H elpful for Everyone

      AI is a transformational technology that can bring meaningful and positive
      change to people and societies across 

      the world, and for our business. At Google, we have been bringing AI into
      our products and services for more than a 

      decade and making them available to our users. Our journey began in 2001,
      when machine learning was first 

      incorporated into Google Search to suggest better spellings to users
      searching the web. Today, AI in our products is Table of Contents Alphabet
      Inc.

      4.
    sentences:
      - >-
        In what ways does Alphabet support the financial health of its
        employees?
      - >-
        Analyze the potential impact of AI-driven tools on Google’s operational
        costs and overall financial health.
      - >-
        What strategies can companies implement to mitigate the financial risks
        associated with problematic content?
  - source_sentence: >-
      Executive Overview

      The following table summarizes our consolidated financial results (in
      millions, except for per share information 

      and percentages):

      Year Ended December 31,

      2022 2023 $ Change % Change

      Consolidated revenues $ 282,836 $ 307,394 $ 24,558  9 %

      Change in consolidated constant currency revenues(1) 10 %

      Cost of revenues $ 126,203 $ 133,332 $ 7,129  6 %

      Operating expenses $ 81,791 $ 89,769 $ 7,978  10 %

      Operating income $ 74,842 $ 84,293 $ 9,451  13 %

      Operating margin  26 %  27 %  1 %

      Other income (expense), net $ (3,514) $ 1,424 $ 4,938 NM

      Net income $ 59,972 $ 73,795 $ 13,823  23 %

      Diluted EPS $ 4.56 $ 5.80 $ 1.24  27 %

      NM = Not Meaningful

      (1) See "Use of Non-GAAP Constant Currency Information " below for details
      relating to our use of constant currency information. 

      •Revenues were $307.4 billion , an increase  of 9% year over year,
      primarily driven by an increase  in Google 

      Services revenues of $19.0 billion , or 8%, and an increase  in Google
      Cloud revenues of $6.8 billion , or 26%. 

      •Total constant currency revenues, which exclude the effect of hedging,
      increased 10% year over year.

      •Cost of revenues  was $133.3 billion , an increase  of 6% year over year,
      primarily driven by increase s in content 

      acquisition costs , compensation expenses, and TAC . The increase in
      compensation expenses included 

      charges related to employee severance associated with the reduction in our
      workforce . Additionally, cost of 

      revenues benefited from a reduction in depreciation due to the change in
      estimated useful lives of our servers 

      and network equipment.

      •Operating expenses were $89.8 billion , an increase  of 10% year over
      year , primarily driven by an increase in 

      compensation expenses  and charges related to our office space
      optimization efforts . The increase in 

      compensation expenses was largely  the result of  charges related to
      employee severance associated with the 

      reduction in our workforce  and an increase in SBC expense.  Operating 
      expenses benefited from  the change in 

      the estimated useful lives of our servers and certain network equipment.

      Other Information:

      •In January 2023, we announced a reduction of our workforce , and as a
      result we recorded employee 

      severance and related charges of $2.1 billion  for the year ended December
      31, 2023. In addition, we are 

      taking actions to optimize our global office space. As a result, exit
      charges recorded during the year ended 

      December 31, 2023, were $1.8 billion . In addition to these exit charges,
      for the year ended December 31, 

      2023, we incurred  $269 million  in accelerated rent and accelerated
      depreciation . For additional information, 

      see Note 8  of the Notes to Consolidated Financial Statements included in
      Item 8 of this Annual Report on 

      Form 10-K.

      •In January 2023, we completed an assessment of the useful lives of our
      servers and network equipment, 

      resulting in a change in the estimated useful life of our servers and
      certain network equipment to six years. 

      The effect of this change was a reduction in depreciation expense of $3.9
      billion  for the year ended December 

      31, 2023, recognized primarily in cost of revenues and R&D expenses. For
      additional information, see Note 1  

      of the Notes to Consolidated Financial Statements included in Item 8 of
      this Annual Report on Form 10-K.Table of Contents Alphabet Inc.

      34.
    sentences:
      - >-
        How does Google’s investment in AI research align with its long-term
        financial strategy and goals?
      - >-
        What role do market and industry factors play in the fluctuation of
        stock prices, regardless of a company's performance?
      - >-
        What was the total consolidated revenue for the year ended December 31,
        2023, and how does it compare to the previous year?
  - source_sentence: >-
      Furthermore, failure to maintain and enhance our brands could harm our
      business, reputation, financial condition, 

      and operating results. Our success will depend largely on our ability to
      remain a technology leader and continue to 

      provide high-quality, trustworthy, innovative products and services that
      are truly useful and play a valuable role in a 

      range of settings. 

      We face a number of manufacturing and supply chain risks that could harm
      our business, financial 

      condition, and operating results. 

      We face a number of risks related to manufacturing and supply chain
      management, which could affect our ability 

      to supply both our products and our services. 

      We rely on contract manufacturers to manufacture or assemble our device s
      and servers and networking 

      equipment used in our technical infrastructure, and we may supply the
      contract manufacturers with components to 

      assemble t he device s and equipment. We also rely on other companies to
      participate in the  supply of components and  

      distribution of our products and services. Our business could be
      negatively affected if we are not able to engage these 

      companies with the necessary capabilities or capacity on reasonable terms,
      or if those we engage fail to meet their Table of Contents Alphabet Inc.

      13.
    sentences:
      - >-
        Discuss the impact of annual stock-based compensation (SBC) awards on
        Alphabet Inc.'s financial reporting.
      - >-
        What financial risks does Google face if it fails to comply with the
        General Data Protection Regulation (GDPR)?
      - >-
        How does the ability to provide innovative products and services
        correlate with a company's revenue growth?
  - source_sentence: >-
      For example, in December 2023, a California jury delivered a verdict in
      Epic Games v. Google  finding that Google 

      violated antitrust laws related to Google Play's billing practices. The
      presiding judge will determine remedies in 2024 

      and the range of potential remedies vary widely. We plan to appeal. In
      addition, the U.S. Department of Justice, 

      various U.S. states, and other plaintiffs have filed several antitrust
      lawsuits about various aspects of our business, 

      including our advertising technologies and practices, the operation and
      distribution of Google Search, and the 

      operation and distribution of the Android operating system and Play Store.
      Other regulatory agencies in the U.S. and 

      around the world, including competition enforcers, consumer protection
      agencies, and data protection authorities, have 

      challenged and may continue to challenge our business practices and
      compliance with laws and regulations. We are 

      cooperating with these investigations and defending litigation  or
      appealing decisions where appropriate.  

      Various laws, regulations, investigations, enforcement lawsuits, and
      regulatory actions have  involved in the past , 

      and may in the future result in substantial fines and penalties,
      injunctive relief, ongoing monitoring and auditing 

      obligations, changes to our products and services, alterations to our
      business models and operations , including 

      divestiture , and collateral related civil litigation or other adverse
      consequences, all of which could harm our business, 

      reputation, financial condition, and operating results. 

      Any of these legal proceedings could result in legal costs, diversion of
      management resources, negative publicity 

      and other harms to our business. Estimating liabilities for our pending
      proceedings is a complex, fact-specific , and 

      speculative process that requires significant judgment, and the amounts we
      are ultimately liable for may be less than or 

      exceed our estimates. The resolution of one or more such proceedings has
      resulted in, and may in the future result in, 

      additional substantial fines, penalties, injunctions, and other sanctions
      that could harm our business, reputation, 

      financial condition, and operating results. 

      For additional information about the ongoing material legal proceedings to
      which we are subject, see Legal 

      Proceedings in Part I, Item 3 of this Annual Report on Form 10-K.

      Privacy, data protection, and data usage regulations are complex and
      rapidly evolving areas. Any failure 

      or alleged failure to comply with these laws could harm our business,
      reputation, financial condition, and 

      operating results. 

      Authorities around the world have adopted and are considering a number of
      legislative and regulatory proposals 

      concerning data protection, data usage, and encryption of user data.
      Adverse legal rulings, legislation, or regulation 

      have resulted in, and may continue to result in, fines and orders
      requiring that we change our practices, which have 

      had and could continue to have an adverse effect on how we provide
      services, harming our business, reputation, 

      financial condition, and operating results. These laws and regulations are
      evolving and subject to interpretation, and 

      compliance obligations could cause us to incur substantial costs or harm
      the quality and operations of our products 

      and services in ways that harm our business.  Examples of these laws
      include : 

      •The General Data Protection Regulation and the United Kingdom General
      Data Protection Regulations, which 

      apply to all of our activities conducted from an establishment in the EU
      or the United Kingdom, respectively, or 

      related to products and services that we offer to EU or the United Kingdom
      users or customers, respectively, or 

      the monitoring of their behavior in the EU or the UK, respectively.

      •Various comprehensive U.S. state and foreign privacy laws, which give new
      data privacy rights to their 

      respective residents (including, in California, a private right of action
      in the event of a data breach resulting 

      from our failure to implement and maintain reasonable security procedures
      and practices) and impose 

      significant obligations on controllers and processors of consumer data.

      •State laws governing the processing of biometric information, such as the
      Illinois Biometric Information Privacy 

      Act and the Texas Capture or Use of Biometric Identifier Act, which impose
      obligations on businesses that 

      collect or disclose consumer biometric information. 

      •Various federal, state, and foreign laws governing how companies provide
      age appropriate experiences to 

      children and minors, including the collection and processing of children
      and minor’s data. These include the 

      Children’s Online Privacy Protection Act of 1998, and the United Kingdom
      Age-Appropriate Design Code, all of 

      which address the use and disclosure of the personal data of children and
      minors and impose obligations on 

      online services or products directed to or likely to be accessed by
      children. 

      •The California Internet of Things Security Law, which regulates the
      security of data used in connection with 

      internet-connected devices.
    sentences:
      - >-
        What are the ethical challenges that may arise from the development of
        new AI products and services?
      - >-
        How might the California Internet of Things Security Law impose
        additional financial obligations on Google?
      - >-
        In the context of Google Services, what factors contribute to the
        competitive nature of the device market, and how might these factors
        affect financial outcomes?
  - source_sentence: >-
      obligations (whether due to financial difficulties or other reasons), or
      make adverse changes in the pricing or other 

      material terms of our arrangements with them. 

      We have experienced and/or may in the future experience supply shortages,
      price increases, quality issues, and/

      or longer lead times that could negatively affect our operations, driven
      by raw material, component availability, 

      manufacturing capacity, labor shortages, industry allocations, logistics
      capacity, inflation, foreign currency exchange 

      rates, tariffs, sanctions and export controls, trade disputes and
      barriers, forced labor concerns, sustainability sourcing 

      requirements, geopolitical tensions, armed conflicts, natural disasters or
      pandemics, the effects of climate change 

      (such as sea level rise, drought, flooding, heat waves, wildfires and
      resultant air quality effects and power shutdowns  

      associated with wildfire prevention, and increased storm severity), power
      loss, and significant changes in the financial 

      or business condition of our suppliers. Some of the components we use in
      our technical infrastructure and our device s 

      are available from only one or limited sources, and we may not be able to
      find replacement vendors on favorable terms 

      in the event of a supply chain disruption. A significant supply
      interruption that affects us or our vendors could delay 

      critical data center upgrades or expansions and delay consumer product
      availability . 

      We may enter into long-term contracts for materials and products that
      commit us to significant terms and 

      conditions. We may face costs for materials and products that are not
      consumed due to market demand, technological 

      change, changed consumer preferences, quality, product recalls, and
      warranty issues. For instance, because certain of 

      our hardware  supply contracts have volume-based pricing or minimum
      purchase requirements, if the volume of sales 

      of our devices decreases or does not reach projected targets, we could
      face increased materials and manufacturing 

      costs or other financial liabilities that could make our products more
      costly per unit to manufacture and harm our 

      financial condition and operating results. Furthermore, certain of our
      competitors may negotiate more favorable 

      contractual terms based on volume and other commitments that may provide
      them with competitive advantages and 

      may affect our supply. 

      Our device s have had, and in the future may have, quality issues
      resulting from design, manufacturing, or 

      operations. Sometimes, these issues may be caused by components we
      purchase from other manufacturers or 

      suppliers. If the quality of our products and services does not meet
      expectations or our products or services are 

      defective or require a recall, it could harm our reputation, financial
      condition, and operating results.  

      We require our suppliers and business partners to comply with laws and,
      where applicable, our company policies 

      and practices, such as the Google Supplier Code of Conduct, regarding
      workplace and employment practices, data 

      security, environmental compliance, and intellectual property licensing,
      but we do not control them or their practices. 

      Violations of law or unethical business practices could result in supply
      chain disruptions, canceled orders, harm to key 

      relationships, and damage to our reputation. Their failure to procure
      necessary license rights to intellectual property 

      could affect our ability to sell our products or services and expose us to
      litigation or financial claims. 

      Interruption to, interference with, or failure of our complex information
      technology and communications 

      systems could hurt our ability to effectively provide our products and
      services, which could harm  our 

      reputation, financial condition, and operating results. 

      The availability of our products and services and fulfillment of our
      customer contracts depend on the continuing 

      operation of our information technology and communications systems. Our
      systems are vulnerable to damage, 

      interference, or interruption from modifications or upgrades, terrorist
      attacks, state-sponsored attacks, natural disasters 

      or pandemics, geopolitical tensions or armed conflicts, export controls
      and sanctions, the effects of climate change 

      (such as sea level rise, drought, flooding, heat waves, wildfires and
      resultant air quality effects and power shutdowns  

      associated with wildfire prevention, and increased storm severity), power
      loss, utility outages, telecommunications 

      failures, computer viruses, software bugs, ransomware attacks,
      supply-chain attacks, computer denial of service 

      attacks, phishing schemes, or other attempts to harm or access our
      systems. Some of our data centers are located in 

      areas with a high risk of major earthquakes or other natural disasters.
      Our data centers are also subject to break-ins, 

      sabotage, and intentional acts of vandalism, and, in some cases, to
      potential disruptions resulting from problems 

      experienced by facility operators or disruptions as a result of
      geopolitical tensions and conflicts happening in the area. 

      Some of our systems are not fully redundant, and disaster recovery
      planning cannot account for all eventualities. The 

      occurrence of a natural disaster or pandemic, closure of a facility, or
      other unanticipated problems affecting our data 

      centers could result in lengthy interruptions in our service.
    sentences:
      - >-
        What are the implications of increased logistics capacity costs on a
        company's overall financial performance?
      - >-
        What are the potential risks associated with the company's reliance on
        consumer subscription-based products for revenue?
      - >-
        How might legal proceedings and regulatory scrutiny affect a company's
        financial condition and operating results?
model-index:
  - name: SUJET AI bge-base Finance Matryoshka
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 768
          type: dim_768
        metrics:
          - type: cosine_accuracy@1
            value: 0.015384615384615385
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.04657342657342657
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.06993006993006994
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.13076923076923078
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.015384615384615385
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.015524475524475523
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.013986013986013986
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.013076923076923076
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.015384615384615385
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.04657342657342657
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.06993006993006994
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.13076923076923078
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.0620726064588503
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.04157842157842149
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.05757497178689022
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 512
          type: dim_512
        metrics:
          - type: cosine_accuracy@1
            value: 0.014965034965034965
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.04531468531468531
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.06713286713286713
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.12755244755244755
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.014965034965034965
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.015104895104895105
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.013426573426573427
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.012755244755244756
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.014965034965034965
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.04531468531468531
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.06713286713286713
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.12755244755244755
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.06036389249600748
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.04032722832722825
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.05606060146944153
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 256
          type: dim_256
        metrics:
          - type: cosine_accuracy@1
            value: 0.012167832167832168
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.04055944055944056
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.06265734265734266
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.11734265734265734
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.012167832167832168
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.013519813519813519
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.012531468531468533
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.011734265734265736
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.012167832167832168
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.04055944055944056
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.06265734265734266
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.11734265734265734
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.054805553416946595
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.03612859362859355
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.050715277611358314
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 128
          type: dim_128
        metrics:
          - type: cosine_accuracy@1
            value: 0.01020979020979021
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.03538461538461538
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.05118881118881119
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.09734265734265735
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.01020979020979021
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.011794871794871797
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.01023776223776224
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.009734265734265736
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.01020979020979021
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.03538461538461538
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.05118881118881119
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.09734265734265735
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.045562900318375184
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.03009612609612603
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.04272564391942989
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 64
          type: dim_64
        metrics:
          - type: cosine_accuracy@1
            value: 0.005874125874125874
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.02125874125874126
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.03370629370629371
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.06741258741258742
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.005874125874125874
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.007086247086247086
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.006741258741258742
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.006741258741258742
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.005874125874125874
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.02125874125874126
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.03370629370629371
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.06741258741258742
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.030435876859011154
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.01942596292596293
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.028981824813925826
            name: Cosine Map@100

SUJET AI bge-base Finance Matryoshka

This is a sentence-transformers model trained. 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
Maximum Sequence Length: 512 tokens
Output Dimensionality: 768 tokens
Similarity Function: Cosine Similarity
Language: en
License: apache-2.0

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': True}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, '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': 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("Rubyando59/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'obligations (whether due to financial difficulties or other reasons), or make adverse changes in the pricing or other \nmaterial terms of our arrangements with them. \nWe have experienced and/or may in the future experience supply shortages, price increases, quality issues, and/\nor longer lead times that could negatively affect our operations, driven by raw material, component availability, \nmanufacturing capacity, labor shortages, industry allocations, logistics capacity, inflation, foreign currency exchange \nrates, tariffs, sanctions and export controls, trade disputes and barriers, forced labor concerns, sustainability sourcing \nrequirements, geopolitical tensions, armed conflicts, natural disasters or pandemics, the effects of climate change \n(such as sea level rise, drought, flooding, heat waves, wildfires and resultant air quality effects and power shutdowns  \nassociated with wildfire prevention, and increased storm severity), power loss, and significant changes in the financial \nor business condition of our suppliers. Some of the components we use in our technical infrastructure and our device s \nare available from only one or limited sources, and we may not be able to find replacement vendors on favorable terms \nin the event of a supply chain disruption. A significant supply interruption that affects us or our vendors could delay \ncritical data center upgrades or expansions and delay consumer product availability . \nWe may enter into long-term contracts for materials and products that commit us to significant terms and \nconditions. We may face costs for materials and products that are not consumed due to market demand, technological \nchange, changed consumer preferences, quality, product recalls, and warranty issues. For instance, because certain of \nour hardware  supply contracts have volume-based pricing or minimum purchase requirements, if the volume of sales \nof our devices decreases or does not reach projected targets, we could face increased materials and manufacturing \ncosts or other financial liabilities that could make our products more costly per unit to manufacture and harm our \nfinancial condition and operating results. Furthermore, certain of our competitors may negotiate more favorable \ncontractual terms based on volume and other commitments that may provide them with competitive advantages and \nmay affect our supply. \nOur device s have had, and in the future may have, quality issues resulting from design, manufacturing, or \noperations. Sometimes, these issues may be caused by components we purchase from other manufacturers or \nsuppliers. If the quality of our products and services does not meet expectations or our products or services are \ndefective or require a recall, it could harm our reputation, financial condition, and operating results.  \nWe require our suppliers and business partners to comply with laws and, where applicable, our company policies \nand practices, such as the Google Supplier Code of Conduct, regarding workplace and employment practices, data \nsecurity, environmental compliance, and intellectual property licensing, but we do not control them or their practices. \nViolations of law or unethical business practices could result in supply chain disruptions, canceled orders, harm to key \nrelationships, and damage to our reputation. Their failure to procure necessary license rights to intellectual property \ncould affect our ability to sell our products or services and expose us to litigation or financial claims. \nInterruption to, interference with, or failure of our complex information technology and communications \nsystems could hurt our ability to effectively provide our products and services, which could harm  our \nreputation, financial condition, and operating results. \nThe availability of our products and services and fulfillment of our customer contracts depend on the continuing \noperation of our information technology and communications systems. Our systems are vulnerable to damage, \ninterference, or interruption from modifications or upgrades, terrorist attacks, state-sponsored attacks, natural disasters \nor pandemics, geopolitical tensions or armed conflicts, export controls and sanctions, the effects of climate change \n(such as sea level rise, drought, flooding, heat waves, wildfires and resultant air quality effects and power shutdowns  \nassociated with wildfire prevention, and increased storm severity), power loss, utility outages, telecommunications \nfailures, computer viruses, software bugs, ransomware attacks, supply-chain attacks, computer denial of service \nattacks, phishing schemes, or other attempts to harm or access our systems. Some of our data centers are located in \nareas with a high risk of major earthquakes or other natural disasters. Our data centers are also subject to break-ins, \nsabotage, and intentional acts of vandalism, and, in some cases, to potential disruptions resulting from problems \nexperienced by facility operators or disruptions as a result of geopolitical tensions and conflicts happening in the area. \nSome of our systems are not fully redundant, and disaster recovery planning cannot account for all eventualities. The \noccurrence of a natural disaster or pandemic, closure of a facility, or other unanticipated problems affecting our data \ncenters could result in lengthy interruptions in our service.',
    "What are the implications of increased logistics capacity costs on a company's overall financial performance?",
    "How might legal proceedings and regulatory scrutiny affect a company's financial condition and operating results?",
]
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

Metric	Value
cosine_accuracy@1	0.0154
cosine_accuracy@3	0.0466
cosine_accuracy@5	0.0699
cosine_accuracy@10	0.1308
cosine_precision@1	0.0154
cosine_precision@3	0.0155
cosine_precision@5	0.014
cosine_precision@10	0.0131
cosine_recall@1	0.0154
cosine_recall@3	0.0466
cosine_recall@5	0.0699
cosine_recall@10	0.1308
cosine_ndcg@10	0.0621
cosine_mrr@10	0.0416
cosine_map@100	0.0576

Information Retrieval

Dataset: dim_512
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.015
cosine_accuracy@3	0.0453
cosine_accuracy@5	0.0671
cosine_accuracy@10	0.1276
cosine_precision@1	0.015
cosine_precision@3	0.0151
cosine_precision@5	0.0134
cosine_precision@10	0.0128
cosine_recall@1	0.015
cosine_recall@3	0.0453
cosine_recall@5	0.0671
cosine_recall@10	0.1276
cosine_ndcg@10	0.0604
cosine_mrr@10	0.0403
cosine_map@100	0.0561

Information Retrieval

Dataset: dim_256
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0122
cosine_accuracy@3	0.0406
cosine_accuracy@5	0.0627
cosine_accuracy@10	0.1173
cosine_precision@1	0.0122
cosine_precision@3	0.0135
cosine_precision@5	0.0125
cosine_precision@10	0.0117
cosine_recall@1	0.0122
cosine_recall@3	0.0406
cosine_recall@5	0.0627
cosine_recall@10	0.1173
cosine_ndcg@10	0.0548
cosine_mrr@10	0.0361
cosine_map@100	0.0507

Information Retrieval

Dataset: dim_128
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0102
cosine_accuracy@3	0.0354
cosine_accuracy@5	0.0512
cosine_accuracy@10	0.0973
cosine_precision@1	0.0102
cosine_precision@3	0.0118
cosine_precision@5	0.0102
cosine_precision@10	0.0097
cosine_recall@1	0.0102
cosine_recall@3	0.0354
cosine_recall@5	0.0512
cosine_recall@10	0.0973
cosine_ndcg@10	0.0456
cosine_mrr@10	0.0301
cosine_map@100	0.0427

Information Retrieval

Dataset: dim_64
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0059
cosine_accuracy@3	0.0213
cosine_accuracy@5	0.0337
cosine_accuracy@10	0.0674
cosine_precision@1	0.0059
cosine_precision@3	0.0071
cosine_precision@5	0.0067
cosine_precision@10	0.0067
cosine_recall@1	0.0059
cosine_recall@3	0.0213
cosine_recall@5	0.0337
cosine_recall@10	0.0674
cosine_ndcg@10	0.0304
cosine_mrr@10	0.0194
cosine_map@100	0.029

Training Details

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: 10
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
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: 10
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}
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: False
hub_always_push: False
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_inputs_for_metrics: False
eval_do_concat_batches: True
fp16_backend: auto
push_to_hub_model_id: None
push_to_hub_organization: None
mp_parameters:
auto_find_batch_size: False
full_determinism: False
torchdynamo: None
ray_scope: last
ddp_timeout: 1800
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
dispatch_batches: None
split_batches: None
include_tokens_per_second: False
include_num_input_tokens_seen: False
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
eval_on_start: False
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss	dim_128_cosine_map@100	dim_256_cosine_map@100	dim_512_cosine_map@100	dim_64_cosine_map@100	dim_768_cosine_map@100
0.0516	10	6.6963	-	-	-	-	-
0.1033	20	7.634	-	-	-	-	-
0.1549	30	6.8573	-	-	-	-	-
0.2065	40	8.1731	-	-	-	-	-
0.2581	50	7.2853	-	-	-	-	-
0.3098	60	7.6009	-	-	-	-	-
0.3614	70	9.0776	-	-	-	-	-
0.4130	80	7.8738	-	-	-	-	-
0.4647	90	10.46	-	-	-	-	-
0.5163	100	10.7396	-	-	-	-	-
0.5679	110	10.3513	-	-	-	-	-
0.6196	120	10.654	-	-	-	-	-
0.6712	130	12.6157	-	-	-	-	-
0.7228	140	11.955	-	-	-	-	-
0.7744	150	13.2498	-	-	-	-	-
0.8261	160	11.2981	-	-	-	-	-
0.8777	170	13.8403	-	-	-	-	-
0.9293	180	9.4428	-	-	-	-	-
0.9810	190	8.1768	-	-	-	-	-
1.0016	194	-	0.0427	0.0507	0.0561	0.029	0.0576
1.0303	200	7.0981	-	-	-	-	-
1.0820	210	7.3113	-	-	-	-	-
1.1336	220	7.0259	-	-	-	-	-
1.1852	230	7.5874	-	-	-	-	-
1.2369	240	7.65	-	-	-	-	-
1.2885	250	7.2387	-	-	-	-	-
1.3401	260	9.001	-	-	-	-	-
1.3917	270	7.5975	-	-	-	-	-
1.4434	280	9.9568	-	-	-	-	-
1.4950	290	10.4123	-	-	-	-	-
1.5466	300	10.5535	-	-	-	-	-
1.5983	310	9.8199	-	-	-	-	-
1.6499	320	12.7258	-	-	-	-	-
1.7015	330	11.9423	-	-	-	-	-
1.7531	340	12.7364	-	-	-	-	-
1.8048	350	12.1926	-	-	-	-	-
1.8564	360	12.926	-	-	-	-	-
1.9080	370	11.8007	-	-	-	-	-
1.9597	380	8.7379	-	-	-	-	-
2.0010	388	-	0.0427	0.0507	0.0561	0.0290	0.0576
2.0090	390	7.1936	-	-	-	-	-
2.0607	400	6.7359	-	-	-	-	-
2.1123	410	7.4212	-	-	-	-	-
2.1639	420	7.346	-	-	-	-	-
2.2156	430	7.6784	-	-	-	-	-
2.2672	440	7.5079	-	-	-	-	-
2.3188	450	7.8875	-	-	-	-	-
2.3704	460	8.7154	-	-	-	-	-
2.4221	470	8.1278	-	-	-	-	-
2.4737	480	11.1214	-	-	-	-	-
2.5253	490	10.5293	-	-	-	-	-
2.5770	500	9.9882	-	-	-	-	-
2.6286	510	11.5283	-	-	-	-	-
2.6802	520	12.4337	-	-	-	-	-
2.7318	530	11.641	-	-	-	-	-
2.7835	540	13.3482	-	-	-	-	-
2.8351	550	11.7302	-	-	-	-	-
2.8867	560	13.7171	-	-	-	-	-
2.9384	570	8.9323	-	-	-	-	-
2.9900	580	7.4869	-	-	-	-	-
3.0003	582	-	0.0427	0.0507	0.0561	0.0290	0.0576
3.0394	590	6.9978	-	-	-	-	-
3.0910	600	7.33	-	-	-	-	-
3.1426	610	7.1879	-	-	-	-	-
3.1943	620	7.9204	-	-	-	-	-
3.2459	630	7.4435	-	-	-	-	-
3.2975	640	7.4079	-	-	-	-	-
3.3491	650	9.2445	-	-	-	-	-
3.4008	660	7.1794	-	-	-	-	-
3.4524	670	10.4496	-	-	-	-	-
3.5040	680	10.7556	-	-	-	-	-
3.5557	690	10.3543	-	-	-	-	-
3.6073	700	9.9478	-	-	-	-	-
3.6589	710	12.6559	-	-	-	-	-
3.7106	720	12.2463	-	-	-	-	-
3.7622	730	12.8381	-	-	-	-	-
3.8138	740	11.726	-	-	-	-	-
3.8654	750	13.4883	-	-	-	-	-
3.9171	760	10.7751	-	-	-	-	-
3.9687	770	8.5484	-	-	-	-	-
3.9997	776	-	0.0427	0.0507	0.0561	0.0290	0.0576
4.0181	780	7.1582	-	-	-	-	-
4.0697	790	7.0161	-	-	-	-	-
4.1213	800	7.11	-	-	-	-	-
4.1730	810	7.4557	-	-	-	-	-
4.2246	820	7.723	-	-	-	-	-
4.2762	830	7.2889	-	-	-	-	-
4.3278	840	8.3884	-	-	-	-	-
4.3795	850	8.1581	-	-	-	-	-
4.4311	860	9.1386	-	-	-	-	-
4.4827	870	10.706	-	-	-	-	-
4.5344	880	10.4258	-	-	-	-	-
4.5860	890	9.9659	-	-	-	-	-
4.6376	900	11.8535	-	-	-	-	-
4.6893	910	12.5578	-	-	-	-	-
4.7409	920	11.834	-	-	-	-	-
4.7925	930	12.5328	-	-	-	-	-
4.8441	940	12.6998	-	-	-	-	-
4.8958	950	12.9728	-	-	-	-	-
4.9474	960	8.9204	-	-	-	-	-
4.9990	970	7.3909	0.0427	0.0507	0.0561	0.0290	0.0576
5.0484	980	6.6683	-	-	-	-	-
5.1000	990	7.5538	-	-	-	-	-
5.1517	1000	6.9256	-	-	-	-	-
5.2033	1010	8.0908	-	-	-	-	-
5.2549	1020	7.254	-	-	-	-	-
5.3066	1030	7.6558	-	-	-	-	-
5.3582	1040	9.2184	-	-	-	-	-
5.4098	1050	7.5886	-	-	-	-	-
5.4614	1060	10.4976	-	-	-	-	-
5.5131	1070	10.785	-	-	-	-	-
5.5647	1080	10.2376	-	-	-	-	-
5.6163	1090	10.4871	-	-	-	-	-
5.6680	1100	12.6986	-	-	-	-	-
5.7196	1110	12.0688	-	-	-	-	-
5.7712	1120	13.1161	-	-	-	-	-
5.8228	1130	11.3866	-	-	-	-	-
5.8745	1140	13.7281	-	-	-	-	-
5.9261	1150	9.8432	-	-	-	-	-
5.9777	1160	8.2606	-	-	-	-	-
5.9984	1164	-	0.0427	0.0507	0.0561	0.0290	0.0576
6.0271	1170	7.0799	-	-	-	-	-
6.0787	1180	7.2981	-	-	-	-	-
6.1304	1190	7.0085	-	-	-	-	-
6.1820	1200	7.4587	-	-	-	-	-
6.2336	1210	7.8467	-	-	-	-	-
6.2853	1220	7.2008	-	-	-	-	-
6.3369	1230	8.8152	-	-	-	-	-
6.3885	1240	7.7205	-	-	-	-	-
6.4401	1250	9.9131	-	-	-	-	-
6.4918	1260	10.212	-	-	-	-	-
6.5434	1270	10.6791	-	-	-	-	-
6.5950	1280	9.8454	-	-	-	-	-
6.6467	1290	12.4647	-	-	-	-	-
6.6983	1300	11.8962	-	-	-	-	-
6.7499	1310	12.8014	-	-	-	-	-
6.8015	1320	12.1836	-	-	-	-	-
6.8532	1330	12.9114	-	-	-	-	-
6.9048	1340	12.1711	-	-	-	-	-
6.9564	1350	8.8125	-	-	-	-	-
6.9977	1358	-	0.0427	0.0507	0.0561	0.0290	0.0576
7.0058	1360	7.2281	-	-	-	-	-
7.0574	1370	6.6681	-	-	-	-	-
7.1091	1380	7.5282	-	-	-	-	-
7.1607	1390	7.1585	-	-	-	-	-
7.2123	1400	7.8507	-	-	-	-	-
7.2640	1410	7.4737	-	-	-	-	-
7.3156	1420	7.6963	-	-	-	-	-
7.3672	1430	8.8799	-	-	-	-	-
7.4188	1440	7.9977	-	-	-	-	-
7.4705	1450	10.9078	-	-	-	-	-
7.5221	1460	10.5731	-	-	-	-	-
7.5737	1470	10.1121	-	-	-	-	-
7.6254	1480	11.2426	-	-	-	-	-
7.6770	1490	12.4832	-	-	-	-	-
7.7286	1500	11.6954	-	-	-	-	-
7.7803	1510	13.4836	-	-	-	-	-
7.8319	1520	11.4752	-	-	-	-	-
7.8835	1530	13.8097	-	-	-	-	-
7.9351	1540	9.0087	-	-	-	-	-
7.9868	1550	7.709	-	-	-	-	-
8.0023	1553	-	0.0427	0.0507	0.0561	0.0290	0.0576
8.0361	1560	7.1515	-	-	-	-	-
8.0878	1570	7.2816	-	-	-	-	-
8.1394	1580	7.1392	-	-	-	-	-
8.1910	1590	7.7863	-	-	-	-	-
8.2427	1600	7.4939	-	-	-	-	-
8.2943	1610	7.3074	-	-	-	-	-
8.3459	1620	9.1739	-	-	-	-	-
8.3975	1630	7.3667	-	-	-	-	-
8.4492	1640	10.2528	-	-	-	-	-
8.5008	1650	10.6824	-	-	-	-	-
8.5524	1660	10.3765	-	-	-	-	-
8.6041	1670	9.853	-	-	-	-	-
8.6557	1680	12.8624	-	-	-	-	-
8.7073	1690	12.0849	-	-	-	-	-
8.7590	1700	12.7345	-	-	-	-	-
8.8106	1710	11.9884	-	-	-	-	-
8.8622	1720	13.2117	-	-	-	-	-
8.9138	1730	11.1261	-	-	-	-	-
8.9655	1740	8.5941	-	-	-	-	-
9.0016	1747	-	0.0427	0.0507	0.0561	0.0290	0.0576
9.0148	1750	7.2587	-	-	-	-	-
9.0665	1760	6.8577	-	-	-	-	-
9.1181	1770	7.2256	-	-	-	-	-
9.1697	1780	7.456	-	-	-	-	-
9.2214	1790	7.6563	-	-	-	-	-
9.2730	1800	7.3877	-	-	-	-	-
9.3246	1810	8.2009	-	-	-	-	-
9.3763	1820	8.5318	-	-	-	-	-
9.4279	1830	8.5052	-	-	-	-	-
9.4795	1840	10.9953	-	-	-	-	-
9.5311	1850	10.4012	-	-	-	-	-
9.5828	1860	10.0235	-	-	-	-	-
9.6344	1870	11.9031	-	-	-	-	-
9.6860	1880	12.5293	-	-	-	-	-
9.7377	1890	11.5157	-	-	-	-	-
9.7893	1900	12.8049	-	-	-	-	-
9.8409	1910	12.4659	-	-	-	-	-
9.8925	1920	13.1517	-	-	-	-	-
9.9442	1930	9.0604	0.0427	0.0507	0.0561	0.0290	0.0576

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.10.13
Sentence Transformers: 3.0.1
Transformers: 4.42.3
PyTorch: 2.5.0.dev20240704+cu124
Accelerate: 0.32.1
Datasets: 2.20.0
Tokenizers: 0.19.1

Citation

BibTeX

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

MatryoshkaLoss

@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

@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}
}