alpha-street-distilroberta-base-sentence-transformer

SentenceTransformer based on distilbert/distilroberta-base

This is a sentence-transformers model finetuned from distilbert/distilroberta-base. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: distilbert/distilroberta-base <!-- at revision fb53ab8802853c8e4fbdbcd0529f21fc6f459b2b -->
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 tokens
• Similarity Function: Cosine Similarity

<!-- - **Training Dataset:** Unknown --> <!-- - **Language:** Unknown --> <!-- - **License:** Unknown -->

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: RobertaModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
)

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("sidddddddddddd/alpha-street-distilroberta-base-sentence-transformer")
# Run inference
sentences = [
    'PART II\n\nItem 8 Judgment is required to determine the SSP for each distinct performance obligation. We use a single amount to estimate SSP for items that are not sold separately, including on-premises licenses sold with SA or software updates provided at no additional charge. We use a range of amounts to estimate SSP when we sell each of the products and services separately and need to determine whether there is a discount to be allocated based on the relative SSP of the various products and services. \n\nIn instances where SSP is not directly observable, such as when we do not sell the product or service separately, we determine the SSP using information that may include market conditions and other observable inputs. We typically have more than one SSP for individual products and services due to the stratification of those products and services by customers and circumstances. In these instances, we may use information such as the size of the customer and geographic region in determining the SSP. \n\nDue to the various benefits from and the nature of our SA program, judgment is required to assess the pattern of delivery, including the exercise pattern of certain benefits across our portfolio of customers. \n\nOur products are generally sold with a right of return, we may provide other credits or incentives, and in certain instances we estimate customer usage of our products and services, which are accounted for as variable consideration when determining the amount of revenue to recognize. Returns and credits are estimated at contract inception and updated at the end of each reporting period if additional information becomes available. Changes to our estimated variable consideration were not material for the periods presented.\n\n## Contract Balances And Other Receivables\n\nTiming of revenue recognition may differ from the timing of invoicing to customers. We record a receivable when revenue is recognized prior to invoicing, or unearned revenue when revenue is recognized subsequent to invoicing.\n\nFor multi-year agreements, we generally invoice customers annually at the beginning of each annual coverage period. We record a receivable related to revenue recognized for multi-year on-premises licenses as we have an unconditional right to invoice and receive payment in the future related to those licenses. \n\nUnearned revenue comprises mainly unearned revenue related to volume licensing programs, which may include SA\n\nand cloud services. Unearned revenue is generally invoiced annually at the beginning of each contract period for multi-year agreements and recognized ratably over the coverage period. Unearned revenue also includes payments for consulting services to be performed in the future, LinkedIn subscriptions, Office 365 subscriptions, Xbox subscriptions, Windows post-delivery support, Dynamics business solutions, and other offerings for which we have been paid in advance and earn the revenue when we transfer control of the product or service. \n\nRefer to Note 13 - Unearned Revenue for further information, including unearned revenue by segment and changes in unearned revenue during the period.\n\nPayment terms and conditions vary by contract type, although terms generally include a requirement of payment within 30 to 60 days. In instances where the timing of revenue recognition differs from the timing of invoicing, we have determined our contracts generally do not include a significant financing component. The primary purpose of our invoicing terms is to provide customers with simplified and predictable ways of purchasing our products and services, not to receive financing from our customers or to provide customers with financing. Examples include invoicing at the beginning of a subscription term with revenue recognized ratably over the contract period, and multiyear on-premises licenses that are invoiced annually with revenue recognized upfront.\n\nAs of June 30, 2024 and 2023, long-term accounts receivable, net of allowance for doubtful accounts, was $4.9 billion and $4.5 billion, respectively, and is included in other long-term assets in our consolidated balance sheets.\n\nThe allowance for doubtful accounts reflects our best estimate of probable losses inherent in the accounts receivable balance. We determine the allowance based on known troubled accounts, historical experience, and other currently available evidence. \n\n\n\n| (In millions) June 30,                                      | 2024   | 2023   | 2022   |    |     |\n|-------------------------------------------------------------|--------|--------|--------|----|-----|\n| Accounts receivable, net of allowance for doubtful accounts | $      | 830 $  | 650    | $  | 633 |\n| Other long-term assets                                      | 54     | 66     | 77     |    |     |\n| Total                                                       | $      | 884 $  | 716    | $  | 710 |\n\n| (In millions) Year Ended June 30,   | 2024                       | 2023   | 2022   |       |    |     |\n|-------------------------------------|----------------------------|--------|--------|-------|----|-----|\n| Balance, beginning of period        | $                          | 716    | $      | 710   | $  | 798 |\n|                                     | Charged to costs and other | 386    | 258    | 157   |    |     |\n|                                     | Write-offs                 | (218)  | (252)  | (245) |    |     |\n| Balance, end of period              | $                          | 884    | $      | 716   | $  | 710 |\n\nAs of June 30, 2024 and 2023, other receivables related to activities to facilitate the purchase of server components were $10.5 billion and $9.2 billion, respectively, and are included in other current assets in our consolidated balance sheets.\n\nWe record financing receivables when we offer certain customers the option to acquire our software products and services offerings through a financing program in a limited number of countries. As of June 30, 2024 and 2023, our financing receivables, net were $4.5 billion and $5.3 billion, respectively, for short-term and long-term financing receivables, which are included in other current assets and other long-term assets in our consolidated balance sheets. We record an allowance to cover expected losses based on troubled accounts, historical experience, and other currently available evidence.\n\n## Assets Recognized From Costs To Obtain A Contract With A Customer\n\nWe recognize an asset for the incremental costs of obtaining a contract with a customer if we expect the benefit of those costs to be longer than one year. We have determined that certain sales incentive programs meet the requirements to be capitalized. Total capitalized costs to obtain a contract were immaterial during the periods presented and are included in other current and long-term assets in our consolidated balance sheets. \n\nWe apply a practical expedient to expense costs as incurred for costs to obtain a contract with a customer when the amortization period would have been one year or less. These costs include our internal sales organization compensation program and certain partner sales incentive programs as we have determined annual compensation is commensurate with annual sales activities.\n\n## Cost Of Revenue\n\nCost of revenue includes: manufacturing and distribution costs for products sold and programs licensed; operating costs related to product support service centers and product distribution centers; costs incurred to include software on PCs sold by original equipment manufacturers ("OEM"), to drive traffic to our websites, and to acquire online advertising space; costs incurred to support and maintain cloud-based and other online products and services, including datacenter costs and royalties; warranty costs; inventory valuation adjustments; costs associated with the delivery of consulting services; and the amortization of capitalized software development costs. Capitalized software development costs are amortized over the estimated lives of the products.',
    'P',
    'i',
]
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]

<!-- ### Direct Usage (Transformers) <details><summary>Click to see the direct usage in Transformers</summary> </details> --> <!-- ### Downstream Usage (Sentence Transformers) You can finetune this model on your own dataset. <details><summary>Click to expand</summary> </details> --> <!-- ### Out-of-Scope Use *List how the model may foreseeably be misused and address what users ought not to do with the model.* --> <!-- ## Bias, Risks and Limitations *What are the known or foreseeable issues stemming from this model? You could also flag here known failure cases or weaknesses of the model.* --> <!-- ### Recommendations *What are recommendations with respect to the foreseeable issues? For example, filtering explicit content.* -->

Training Details

Training Dataset

Unnamed Dataset

• Size: 54 training samples
• Columns: <code>sentence_0</code>, <code>sentence_1</code>, and <code>sentence_2</code>
• Approximate statistics based on the first 54 samples:

  	sentence_0	sentence_1	sentence_2
  type	string	string	string
  details	<ul><li>min: 49 tokens</li><li>mean: 495.98 tokens</li><li>max: 512 tokens</li></ul>	<ul><li>min: 2 tokens</li><li>mean: 2.87 tokens</li><li>max: 3 tokens</li></ul>	<ul><li>min: 3 tokens</li><li>mean: 3.0 tokens</li><li>max: 3 tokens</li></ul>

• Samples:

  sentence_0	sentence_1	sentence_2
  <code># Microsoft Quarterly Highlights, Product Releases And Enhancements<br><br>## Fy24 Q4<br><br>Every quarter Microsoft delivers hundreds of products, either as new releases or services or as enhancements to current products and services. These releases are a result of significant research and development investments, made over multiple years, designed to help customers be more productive and secure and to deliver differentiated value across the cloud and the edge. <br><br>Following are the major product releases and other highlights for the quarter, organized by product categories, to help illustrate how we are accelerating innovation across our businesses while expanding our market opportunities.<br><br>## Microsoft 365 Consumer/Commercial<br><br>Access Copilot directly in new Outlook for Windows and web is the same Microsoft Copilot experience users get from the Microsoft Teams app, at copilot.microsoft.com <br><br>(Work mode), and other places but is now available from within Microsoft Outlook for Windows and web. <br><br>Interact with Copilot in the new Outlook for Windows and web allows users to ask questions or provide instructions directly inside Outlook. Users can type and receive responses from Copilot in a right-hand side panel that can be opened by the user by clicking on the Copilot icon in the top nav bar. <br><br>Access Copilot directly in classic Outlook for Windows is the Microsoft Copilot experience users get in Microsoft Teams and copilot.microsoft.com (Work mode) but is now available from within classic Microsoft Outlook for Windows. <br><br>Coaching by Copilot in classic Outlook for Windows combines the power of LLMs and Outlook data to coach users on how to best get their message across.<br><br>Draft by Copilot in classic Outlook for Windows combines the power of LLMs and Outlook data to help users draft new messages or replies. <br><br>Bing search integration brings the power of the web to chats in Word. Ask Copilot a question in Word chat and Copilot generates an answer using integrated Bing search, so users never have to leave the app. Available on web and Windows desktop. <br><br> Copilot in Loop is available in Loop components in Teams and Outlook. People with a Copilot License can now leverage Copilot in Loop when they paste Loop components into the Teams app, Outlook and Meeting Notes. <br><br>The Microsoft Copilot Academy is a new addition to Viva Learning designed to help customers effectively utilize Copilot experiences through guided upskilling. Learning content curated by Microsoft experts is organized into learning paths like "meet Copilot," "intro to prompts" and more. <br><br>The Microsoft Copilot Dashboard provides company leaders with out-of-the-box reports to understand and measure every stage of their Copilot transformation journey. This privacy-protected data spans across readiness, adoption, impact and user sentiment via survey response.<br><br>Copilot in Microsoft Viva Engage helps leaders create compelling and inspiring posts from simple prompts or trending topics within workplace communities and storyline conversations. Copilot offers suggestions to personalize messages with options to adjust tone and length and to suggest relevant images, all to help drive authenticity and engagement. Further, leaders can utilize Copilot to analyze engagement metrics, assess sentiment and recommend responses.<br><br>The new Copilot Deployment Kit in Viva Amplify helps organizations launch and roll out Copilot campaigns with ease. Teams can use prebuilt campaign templates to drive awareness and help employees learn what Copilot can do for them. It comes with eight predrafted communications that include videos, and guides that can easily be customized and sent to help users learn about Copilot. <br><br>ONA is a critical workplace analytics tool to help organizations better understand and improve how teams work together through collaboration, communication, change management and more. The ONA change management template enables analysts to conduct in-depth analysis to help understand how major organizational changes, such as a reorganization, large-scale transformation or shift in work modes impact collaboration behavior.<br><br>The solutions of the Microsoft Intune Suite help organizations improve their security posture, simplify IT and security operations, and reduce costs of managing and protecting their endpoint estate. Intune Suite and standalone solutions are now also generally available for customers in the United States federal, state, local and tribal governments, assisting them to meet the demands of the Executive Order to achieve a Zero Trust security model.<br><br>Microsoft Teams<br><br><br><br>Intelligent recap supports meetings with just transcription enabled, allowing people to enjoy AI-generated summaries for more types of meetings. Sometimes, meeting participants may choose to transcribe a meeting without recording it. In these cases, Teams Premium and Copilot for Microsoft 365 licensed users will be able to access AI-generated notes, AI-generated tasks, and name mentions after the meeting to catch up. <br><br>Intelligent call recap brings one of the best AI features in meetings to calling. <br><br>Intelligent call recap provides AI-powered insights and recaps of users' VoIP and PSTN calls in Teams. This feature is available for Teams Premium and Copilot for Microsoft 365 users. <br><br>Help keep confidential meetings protected with expanded controls for managing who can initiate transcription. Meeting organizers with eligible licenses can see a control called "Who can record and transcribe" in the meeting options, offering three choices: (1) organizers and co-organizers, or (2) organizers, co-organizers and presenters, or (3) no one. This way meeting organizers can manage which roles can initiate recording and transcription for a meeting.<br><br>Users can now instruct Copilot to adjust their draft message. To try it out, write a message in chat or channels, open Copilot beneath the message box in Teams, choose to adjust the message with a custom prompt, and type a prompt like "add a call to action" or "make it persuasive" or "convert my message into a list and add inclusive language." <br><br>In addition to accessing Copilot in Teams during a meeting on a mobile device, users can now access Copilot in Teams and ask questions about meetings after they end. <br><br>Just tap a Teams calendar item to access Copilot in the meeting "Details" or <br><br>"Recap" view. This is available for eligible meetings with transcription turned on.<br><br>When users get a message in a different language, they will see a suggestion to translate the message into their preferred language. Additionally, in translation settings, they can select which languages they don't want to translate and whether to translate messages automatically. These features reduce the need for manual translation or switching between apps.<br><br>Nearby conversations can interfere with Teams calls and meetings. Voice isolation is an AI-based advanced noise suppression feature that eliminates unwanted background noise, including other human voices. The technology recognizes a user's voice profile and ensures only their voice is transmitted. Voice isolation can be enabled for calls and meetings. <br><br> Zero-touch provisioning is a feature that allows users to deploy Microsoft Teams Rooms on Windows devices without any physical intervention at the site. It leverages Windows Autopilot and Auto-login technologies to automate the enrollment and configuration of the devices from the cloud, using the credentials stored in the Teams Rooms Pro Management service. <br><br>Users can extend the power and knowledge of Copilot in Teams meetings for sales organizations by enabling a plugin to connect to Copilot for Sales. With this plugin, Copilot can process conversations in real time and return insights to sellers, such as an overview of an account opportunity, based on the organization's Copilot for Sales data. Copilot can also suggest dynamic prompts for querying account information when sellers mention keywords and names during a discussion. <br><br>Frontline teams can harness the power of Copilot for Microsoft 365 with the new Shifts plugin. Both managers and workers can ask Copilot to show them their shifts schedule for their specific team, as well as open shifts and time off. With quick insights at their fingertips, frontline teams can manage schedules with more agility and speed so they can focus on critical tasks. Shifts plugin for Copilot is now generally available with both the Copilot for Microsoft 365 license as well as Microsoft Teams E and F-SKU licenses. <br><br>## Windows<br><br>The company announced a new category of Windows PCs designed for AI called Copilot+ PCs. Copilot+ PCs are the fastest, most intelligent Windows PCs ever built. <br><br>With powerful new silicon capable of an incredible 40+ TOPS (trillion operations per second), all-day battery life and access to the most advanced AI models, Copilot+ <br><br>PCs will enable you to do things you can't on any other PC. Generate and refine AI <br><br>images in near real time directly on the device using Cocreator; bridge language barriers with Live Captions, translating audio from 40+ languages into English; and unlock more AI capabilities with first-party and third-party apps faster than ever. Introducing Copilot+ PCs - The Official Microsoft Blog.<br><br>Windows 365 Frontline is designed to meet the distinct needs of shift and part-time employees. With Windows 365 Frontline for FedRAMP, Cloud PCs are provisioned in an Azure Commercial datacenter and meet FedRAMP requirements when they are properly configured and used within CONUS. <br><br>Hibernation support in Azure Virtual Desktop is now generally available. Hibernating a session host virtual machine (VM) deallocates the machine while persisting the VMs in-memory state. When a VM hibernates, users only pay for storage and networking costs, not compute costs. When it restarts, users can quickly resume their work. <br><br> App attach for Azure Virtual Desktop allows IT admins to dynamically attach applications from an application package to a user session in Azure Virtual Desktop. <br><br>App attach is now generally available. The new UI for App Attach will be seen in the Azure Portal. Customers and partners should use the new App Attach application delivery approach going forward.<br><br>## Dynamics 365<br><br>Dynamics 365 Customer Insights enables users to ensure messages go to the right contact email address. Users can choose which of a contact's email addresses to target in their journeys. For example, some email messages may be more appropriate for a contact's work email address, whereas others may best target a personal email address. Users can also now use timeline highlights, which enable quick access to actionable record updates.</code>	<code>#</code>	<code>S</code>
  <code><br><br>PART I<br><br>Item 1A<br><br>abusive activities through our cloud-based services, such as unauthorized account access, payment fraud, or terms of service violations including cryptocurrency mining or launching cyberattacks. While are committed to detecting and controlling such misuse of our cloud-based and AI services, our efforts may not be effective, and we may incur reputational damage or experience adverse impacts to our business and results of operations.<br><br>## Risks Relating To The Evolution Of Our Business<br><br>We make significant investments in products and services that may not achieve expected returns. We will continue to make significant investments in research, development, and marketing for existing products, services, and technologies. In addition, we are focused on developing new AI platform services and incorporating AI into existing products and services. We also invest in the development and acquisition of a variety of hardware for productivity, communication, and entertainment, including PCs, tablets, and gaming devices. Investments in new technology are speculative. Commercial success depends on many factors, including innovation, developer support, and effective distribution and marketing. If customers do not perceive our latest offerings as providing significant new functionality or other value, they may reduce their purchases of new software and hardware products or upgrades, unfavorably affecting revenue. We may not achieve significant revenue from new product, service, and distribution channel investments for several years, if at all. New products and services may not be profitable or may not achieve operating margins as high as we have experienced historically. We may not get engagement in certain features that drive post-sale monetization opportunities. Our data-handling practices across our products and services will continue to be under scrutiny. Perceptions of mismanagement, driven by regulatory activity or negative public reaction to our practices or product experiences, could negatively impact product and feature adoption. Developing new technologies is complex. It can require long development and testing periods. We could experience significant delays in new releases or significant problems in creating new products or services. These factors could adversely affect our business, financial condition, and results of operations. <br><br>Acquisitions, joint ventures, and strategic alliances may have an adverse effect on our business. We expect to continue making acquisitions and entering into joint ventures and strategic alliances as part of our long-term business strategy. For example, in March 2022 we completed our acquisition of Nuance Communications, Inc., and in October 2023 we completed our acquisition of Activision Blizzard, Inc. ("Activision Blizzard"). In January 2023 we announced the third phase of our OpenAI strategic partnership. Acquisitions and other transactions and arrangements involve significant challenges and risks, including that they do not advance our business strategy, that we get an unsatisfactory return on our investment, that they raise new compliance-related obligations and challenges, that we have difficulty integrating and retaining new employees, business systems, and technology, that they distract management from our other businesses, or that announced transactions may not be completed. If an arrangement fails to adequately anticipate changing circumstances and interests of a party, it may result in early termination or renegotiation of the arrangement. We also have limited ability to control or influence third parties with whom we have arrangements, which may impact our ability to realize the anticipated benefits. The success of these transactions and arrangements depend in part on our ability to leverage them to enhance our existing products and services or develop compelling new ones, as well as the acquired companies' ability to meet our policies and processes in areas such as data governance, privacy, and cybersecurity. It may take longer than expected to realize the full benefits from these transactions and arrangements, such as increased revenue or enhanced efficiencies, or the benefits may ultimately be smaller than we expected. In addition, an acquisition may be subject to challenge even after it has been completed. For example, the Federal Trade Commission continues to challenge our Activision Blizzard acquisition and could, if successful, alter or unwind the transaction. These events could adversely affect our business, operations, financial condition, and results of operations.<br><br>If our goodwill or amortizable intangible assets become impaired, we may be required to record a significant charge to earnings. We acquire other companies and intangible assets and may not realize all the economic benefit from those acquisitions, which could cause an impairment of goodwill or intangibles. We review our amortizable intangible assets for impairment when events or changes in circumstances indicate the carrying value may not be recoverable. We test goodwill for impairment at least annually. Factors that may be a change in circumstances, indicating that the carrying value of our goodwill or amortizable intangible assets may not be recoverable, include a decline in our stock price and market capitalization, reduced future cash flow estimates, and slower growth rates in industry segments in which we participate. We have in the past recorded, and may in the future be required to record, a significant charge in our consolidated financial statements during the period in which any impairment of our goodwill or amortizable intangible assets is determined, negatively affecting our results of operations.<br><br><br><br><br> PART I<br> <br> Item 1A<br> <br> CYBERSECURITY, DATA PRIVACY, AND PLATFORM ABUSE RISKS<br> <br>Cyberattacks and security vulnerabilities could lead to reduced revenue, increased costs, liability claims, or<br> <br>harm to our reputation or competitive position.<br> <br><br>```<br><br>## Security Of Our Information Technology<br><br>Threats to IT security can take a variety of forms. Individual and groups of hackers and sophisticated organizations, including state-sponsored organizations or nation-states, continuously undertake attacks that pose threats to our customers and our IT, and we have experienced cybersecurity incidents in which such actors have gained unauthorized access to our IT systems and data, including customer systems and data. These actors use a wide variety of methods, which include developing and deploying malicious software; exploiting known and potential vulnerabilities or intentionally designed processes in hardware, software, or other infrastructure to attack our products and services or gain access to our networks and datacenters; using social engineering techniques to induce our employees, users, partners, or customers to disclose sensitive information, such as passwords, or take other actions to gain access to our data or our users' or customers' data; or acting in a coordinated manner or conducting coordinated attacks. For example, as previously disclosed in our Form 8-K filed with the Securities and Exchange Commission on January 19, 2024 and amended on March 8, 2024, beginning in late November 2023, a nation-state associated threat actor used a password spray attack to compromise a legacy test account and, in turn, gain access to Microsoft email accounts. The threat actor used and may continue to use information it obtained to gain, or attempt to gain, unauthorized access to some of our source code repositories and internal systems, and the threat actor may utilize this information to otherwise adversely affect our business and results of operations. This incident has and may continue to result in harm to our reputation and customer relationships. Additionally, we may discover additional impacts of this or other incidents as part of our ongoing examination of this incident. Nation-state and statesponsored actors can sustain malicious activities for extended periods and deploy significant resources to plan and carry out attacks. Nation-state attacks against us, our customers, or our partners have and may continue to intensify during periods of intense diplomatic or armed conflict, such as the ongoing conflict in Ukraine. Cyber incidents and attacks, individually or in the aggregate, could adversely affect our financial condition, results of operations, competitive position, and reputation, or expose us to legal or regulatory risk.<br><br>Inadequate account security or organizational security practices, including those of companies we have acquired or those of the third parties we utilize, have resulted and may result in unauthorized access to our IT systems and data, including customer systems and data, in the future. For example, system administrators may fail to timely remove employee account access when no longer appropriate. Employees or third parties may intentionally compromise our or our users' security or systems or reveal confidential information. Malicious actors may employ the IT supply chain to introduce malware through software updates or compromised supplier accounts or hardware.<br><br>Cyberthreats are constantly evolving and becoming increasingly sophisticated and complex, increasing the difficulty of detecting and successfully defending against them. Threat actors may also utilize emerging technologies, such as AI and machine learning. We may have no current capability to detect certain vulnerabilities or new attack methods, which may allow them to persist in the environment over long periods of time. It may be difficult to determine the best way to investigate, mitigate, contain, and remediate the harm caused by a cyber incident. Such efforts may not be successful, and we may make errors or fail to take necessary actions. It is possible that threat actors may gain undetected access to other networks and systems after establishing a foothold on an internal system. Cyber incidents and attacks can have cascading impacts that unfold with increasing speed across our internal networks and systems, as well as those of our partners and customers. In addition, it may take considerable time for us to investigate and evaluate the full impact of incidents, particularly for sophisticated attacks. These factors may inhibit our ability to provide prompt, full, and reliable information about the incident to our customers, partners, regulators, and the public. Breaches of our facilities, network, or data security can disrupt the security of our systems and business applications, impair our ability to provide services to our customers and protect the privacy of their data, result in product development delays, compromise confidential or technical business information, result in theft or misuse of our intellectual property or other assets, subject us to ransomware attacks, require us to allocate more resources to improve technologies or remediate the impacts of attacks, or otherwise adversely affect our business. In addition, actions taken to remediate an incident could result in outages, data losses, and disruptions of our services.</code>	<code>`</code>	<code>o</code>
  <code>PART II<br><br>Item 7<br><br>## Research And Development Costs<br><br>Costs incurred internally in researching and developing a computer software product are charged to expense until technological feasibility has been established for the product. Once technological feasibility is established, software costs are capitalized until the product is available for general release to customers. Judgment is required in determining when technological feasibility of a product is established. We have determined that technological feasibility for our software products is reached after all high-risk development issues have been resolved through coding and testing. Generally, this occurs shortly before the products are released to production. The amortization of these costs is included in cost of revenue over the estimated life of the products. <br><br>## Legal And Other Contingencies<br><br>The outcomes of legal proceedings and claims brought against us are subject to significant uncertainty. An estimated loss from a loss contingency such as a legal proceeding or claim is accrued by a charge to income if it is probable that an asset has been impaired or a liability has been incurred and the amount of the loss can be reasonably estimated. In determining whether a loss should be accrued we evaluate, among other factors, the degree of probability of an unfavorable outcome and the ability to make a reasonable estimate of the amount of loss. Changes in these factors could materially impact our consolidated financial statements.<br><br>## Income Taxes<br><br>The objectives of accounting for income taxes are to recognize the amount of taxes payable or refundable for the current year, and deferred tax liabilities and assets for the future tax consequences of events that have been recognized in an entity's financial statements or tax returns. We recognize the tax benefit from an uncertain tax position only if it is more likely than not that the tax position will be sustained on examination by the taxing authorities, based on the technical merits of the position. The tax benefits recognized in the financial statements from such a position are measured based on the largest benefit that has a greater than 50% likelihood of being realized upon ultimate settlement. Accounting literature also provides guidance on derecognition of income tax assets and liabilities, classification of deferred income tax assets and liabilities, accounting for interest and penalties associated with tax positions, and income tax disclosures. Judgment is required in assessing the future tax consequences of events that have been recognized in our consolidated financial statements or tax returns. Variations in the actual outcome of these future tax consequences could materially impact our consolidated financial statements. <br><br>## Business Combinations - Valuation Of Intangible Assets<br><br>Accounting for business combinations requires significant judgments when allocating the purchase price to the estimated fair values of assets acquired and liabilities assumed at the acquisition date. Determination of fair value involves estimates and assumptions which can be complex, most notably with respect to intangible assets. Critical estimates used in the valuation of intangible assets include, but are not limited to, the amount and timing of projected cash flows, useful lives, and discount rates. While management's estimates of fair value are based on assumptions that are believed to be reasonable, these assumptions are inherently uncertain as they pertain to forward-looking views of our business and market conditions. The judgments made in this valuation process could materially impact our consolidated financial statements.<br><br><br><br>## Statement Of Management'S Responsibility For Financial Statements<br><br>Management is responsible for the preparation of the consolidated financial statements and related information that are presented in this report. The consolidated financial statements, which include amounts based on management's estimates and judgments, have been prepared in conformity with accounting principles generally accepted in the United States of America. <br><br>The Company designs and maintains accounting and internal control systems to provide reasonable assurance at reasonable cost that assets are safeguarded against loss from unauthorized use or disposition, and that the financial records are reliable for preparing consolidated financial statements and maintaining accountability for assets. These systems are augmented by written policies, an organizational structure providing division of responsibilities, careful selection and training of qualified personnel, and a program of internal audits. <br><br>The Company engaged Deloitte & Touche LLP, an independent registered public accounting firm, to audit and render an opinion on the consolidated financial statements and internal control over financial reporting in accordance with the standards of the Public Company Accounting Oversight Board (United States). <br><br>The Board of Directors, through its Audit Committee, consisting solely of independent directors of the Company, meets periodically with management, internal auditors, and our independent registered public accounting firm to ensure that each is meeting its responsibilities and to discuss matters concerning internal controls and financial reporting. Deloitte & Touche LLP and the internal auditors each have full and free access to the Audit Committee. <br><br>Satya Nadella Chief Executive Officer Amy E. Hood Executive Vice President and Chief Financial Officer Alice L. Jolla Corporate Vice President and Chief Accounting Officer<br><br><br><br># Item 7A. Quantitative And Qualitative Disclosures About Market Risk<br><br>## Risks<br><br>We are exposed to economic risk from foreign exchange rates, interest rates, credit risk, and equity prices. We use derivatives instruments to manage these risks, however, they may still impact our consolidated financial statements. <br><br>## Foreign Currencies<br><br>Certain forecasted transactions, assets, and liabilities are exposed to foreign currency risk. We monitor our foreign currency exposures daily to maximize the economic effectiveness of our foreign currency positions, including hedges. Principal currency exposures include the Euro, Japanese yen, British pound, Canadian dollar, and Australian dollar. <br><br>## Interest Rate<br><br>Securities held in our fixed-income portfolio are subject to different interest rate risks based on their maturities. We manage the average maturity of the fixed-income portfolio to achieve economic returns that correlate to certain global fixed-income indices. <br><br>## Credit<br><br>Our fixed-income portfolio is diversified and consists primarily of investment-grade securities. We manage credit exposures relative to broad-based indices to facilitate portfolio diversification.<br><br>## Equity<br><br>Securities held in our equity investments portfolio are subject to price risk.<br><br>## Sensitivity Analysis<br><br>The following table sets forth the potential loss in future earnings or fair values, including associated derivatives, resulting from hypothetical changes in relevant market rates or prices:<br><br>	(In millions)	June 30,

• Loss: <code>TripletLoss</code> with these parameters:

  {
      "distance_metric": "TripletDistanceMetric.EUCLIDEAN",
      "triplet_margin": 5
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• num_train_epochs: 10
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

<details><summary>Click to expand</summary>

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: no
• prediction_loss_only: True
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 5e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1
• num_train_epochs: 10
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.0
• 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: False
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: 0
• ddp_backend: None
• tpu_num_cores: None
• tpu_metrics_debug: False
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• past_index: -1
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_min_num_params: 0
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• fsdp_transformer_layer_cls_to_wrap: None
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• dataloader_pin_memory: 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
• use_liger_kernel: False
• eval_use_gather_object: False
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin

</details>

Framework Versions

• Python: 3.10.14
• Sentence Transformers: 3.2.1
• Transformers: 4.45.1
• PyTorch: 2.4.0
• Accelerate: 0.34.2
• Datasets: 3.0.1
• Tokenizers: 0.20.0

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",
}

TripletLoss

@misc{hermans2017defense,
    title={In Defense of the Triplet Loss for Person Re-Identification},
    author={Alexander Hermans and Lucas Beyer and Bastian Leibe},
    year={2017},
    eprint={1703.07737},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}

<!-- ## Glossary *Clearly define terms in order to be accessible across audiences.* --> <!-- ## Model Card Authors *Lists the people who create the model card, providing recognition and accountability for the detailed work that goes into its construction.* --> <!-- ## Model Card Contact *Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors.* -->