Hugging Face's logo

Datasets:
andesco
/
risk-management-resources-ocr-sample

Modalities:
Image
Text
Formats:
parquet
Size:
< 1K
Tags:
ocr
document-processing
deepseek
deepseek-ocr
markdown
uv-script
Libraries:
Datasets
pandas
Polars
Dataset card Data Studio Files
xet
Community
Dataset Viewer
Auto-converted to Parquet Duplicate
image
imagewidth (px)
1.11k
1.2k
markdown
stringclasses
10 values
inference_info
stringclasses
1 value
<|ref|>sub_title<|/ref|><|det|>[[97, 123, 328, 147]]<|/det|> ## Application Question <|ref|>text<|/ref|><|det|>[[99, 153, 789, 312]]<|/det|> 10- 5. Joan is a risk management professional working for the parent company of a multinational corporation. She is trying to decide whether she should buy admitted or nonadmitted insurance to cover her employer's foreign subsidiary. Last year she purchased coverage from an admitted insurer. In that year, major claims went unpaid because of the insolvency of the insurer. The insurer also disputed the validity of the claims, saying Joan had misinterpreted the policy, which was written in the language of the country where the foreign subsidiary was domiciled. How will last year's lack of coverage likely influence Joan's decision to buy coverage from an admitted versus a nonadmitted insurer?
[{"model_id": "deepseek-ai/DeepSeek-OCR", "model_name": "DeepSeek-OCR", "column_name": "markdown", "timestamp": "2026-02-21T19:32:27.962928", "prompt_mode": "document", "batch_size": 8, "max_tokens": 8192, "gpu_memory_utilization": 0.8, "max_model_len": 8192, "script": "deepseek-ocr-vllm.py", "script_url": "https://huggingface.co/datasets/uv-scripts/ocr/raw/main/deepseek-ocr-vllm.py"}]
<|ref|>image<|/ref|><|det|>[[90, 120, 930, 595]]<|/det|> <|ref|>text<|/ref|><|det|>[[300, 625, 911, 697]]<|/det|> The most frequently used categories are industry- or process- specific and are known by names that signify the first letter of each of the categories within the group. For example, the "6 Ms" group of categories typically is used to determine root causes of problems in manufacturing: <|ref|>text<|/ref|><|det|>[[301, 704, 925, 936]]<|/det|> - Machine (technology)—Any equipment or technology involved in completing a job, such as computers, tools, or other machinery- Method (process)—Requirements that specify how a process is performed (for example, policies, procedures, rules, regulations, and so forth)- Materials—Information, parts, pens, paper, raw materials, and so forth used to produce the final product (consumables)- Manpower (physical work)/mindpower (brain work)—Anyone involved in a process who provides physical or mental effort- Measurement (inspection)—Data generated from a process used to judge the quality of the product- Milieu/Mother Nature (environment)—Local conditions, such as temperature, time, location, or culture, within which a process operates
[{"model_id": "deepseek-ai/DeepSeek-OCR", "model_name": "DeepSeek-OCR", "column_name": "markdown", "timestamp": "2026-02-21T19:32:27.962928", "prompt_mode": "document", "batch_size": 8, "max_tokens": 8192, "gpu_memory_utilization": 0.8, "max_model_len": 8192, "script": "deepseek-ocr-vllm.py", "script_url": "https://huggingface.co/datasets/uv-scripts/ocr/raw/main/deepseek-ocr-vllm.py"}]
<|ref|>text<|/ref|><|det|>[[100, 105, 577, 127]]<|/det|> 1- 3. List the general categories of risk treatment options. <|ref|>text<|/ref|><|det|>[[100, 264, 768, 305]]<|/det|> 1- 4. Describe risk treatment techniques for events that appear to have primarily positive potential outcomes. <|ref|>sub_title<|/ref|><|det|>[[110, 468, 416, 495]]<|/det|> ## Educational Objective 2 <|ref|>text<|/ref|><|det|>[[110, 498, 555, 516]]<|/det|> Describe risk financing and its importance to organizations. <|ref|>sub_title<|/ref|><|det|>[[89, 553, 360, 578]]<|/det|> ## Key Words and Phrases <|ref|>text<|/ref|><|det|>[[89, 583, 212, 601]]<|/det|> Risk financing <|ref|>text<|/ref|><|det|>[[89, 667, 160, 684]]<|/det|> Transfer <|ref|>text<|/ref|><|det|>[[89, 752, 129, 769]]<|/det|> Risk <|ref|>text<|/ref|><|det|>[[89, 838, 226, 855]]<|/det|> Futures contract
[{"model_id": "deepseek-ai/DeepSeek-OCR", "model_name": "DeepSeek-OCR", "column_name": "markdown", "timestamp": "2026-02-21T19:32:27.962928", "prompt_mode": "document", "batch_size": 8, "max_tokens": 8192, "gpu_memory_utilization": 0.8, "max_model_len": 8192, "script": "deepseek-ocr-vllm.py", "script_url": "https://huggingface.co/datasets/uv-scripts/ocr/raw/main/deepseek-ocr-vllm.py"}]
<|ref|>text<|/ref|><|det|>[[631, 720, 797, 740]]<|/det|> 5502 AI54243 <|ref|>text<|/ref|><|det|>[[656, 750, 916, 769]]<|/det|> ISBN 978-0-89463-991-3 <|ref|>text<|/ref|><|det|>[[830, 770, 907, 783]]<|/det|> 90000 <|ref|>text<|/ref|><|det|>[[585, 804, 650, 813]]<|/det|> 0806117500 <|ref|>text<|/ref|><|det|>[[585, 813, 650, 821]]<|/det|> M006117500 <|ref|>image<|/ref|><|det|>[[505, 800, 716, 860]]<|/det|> <|ref|>text<|/ref|><|det|>[[631, 852, 816, 870]]<|/det|> 91780894639913 <|ref|>text<|/ref|><|det|>[[100, 800, 283, 820]]<|/det|> The Institutes* <|ref|>text<|/ref|><|det|>[[100, 820, 446, 835]]<|/det|> 720 Providence Road, Suite 100 | Malvern, PA 19355-3433 <|ref|>text<|/ref|><|det|>[[100, 837, 551, 852]]<|/det|> (800) 644-2101 | Fax: (610) 640-9576 | Email: CustomerSuccess@TheInstitutes.org <|ref|>text<|/ref|><|det|>[[100, 854, 203, 868]]<|/det|> TheInstitutes.org
[{"model_id": "deepseek-ai/DeepSeek-OCR", "model_name": "DeepSeek-OCR", "column_name": "markdown", "timestamp": "2026-02-21T19:32:27.962928", "prompt_mode": "document", "batch_size": 8, "max_tokens": 8192, "gpu_memory_utilization": 0.8, "max_model_len": 8192, "script": "deepseek-ocr-vllm.py", "script_url": "https://huggingface.co/datasets/uv-scripts/ocr/raw/main/deepseek-ocr-vllm.py"}]
<|ref|>sub_title<|/ref|><|det|>[[22, 108, 201, 130]]<|/det|> ## Review Questions <|ref|>text<|/ref|><|det|>[[27, 139, 591, 177]]<|/det|> 6- 1. List five widely used techniques of alternative dispute resolution (ADR). (p. 7.22) <|ref|>text<|/ref|><|det|>[[27, 308, 515, 329]]<|/det|> 6- 2. Identify the steps in the litigation process. (p. 7.25) <|ref|>text<|/ref|><|det|>[[28, 459, 587, 515]]<|/det|> 6- 3. Describe loss reduction measures that an organization might use to reduce the severity of loss from contractual liability. (pp. 7.29- 7.30) <|ref|>text<|/ref|><|det|>[[28, 645, 612, 683]]<|/det|> 6- 4. Explain why most organizations focus attention on loss prevention rather than loss reduction. (p. 7.31)
[{"model_id": "deepseek-ai/DeepSeek-OCR", "model_name": "DeepSeek-OCR", "column_name": "markdown", "timestamp": "2026-02-21T19:32:27.962928", "prompt_mode": "document", "batch_size": 8, "max_tokens": 8192, "gpu_memory_utilization": 0.8, "max_model_len": 8192, "script": "deepseek-ocr-vllm.py", "script_url": "https://huggingface.co/datasets/uv-scripts/ocr/raw/main/deepseek-ocr-vllm.py"}]
<|ref|>sub_title<|/ref|><|det|>[[82, 110, 339, 137]]<|/det|> ## Educational Objective 2 <|ref|>text<|/ref|><|det|>[[80, 139, 723, 160]]<|/det|> Describe and illustrate the following theories and approaches of accident causation: <|ref|>text<|/ref|><|det|>[[80, 163, 496, 277]]<|/det|> a. Domino theory b. General methods of control approach c. Energy transfer theory d. Technique of operations review (TOR) approach e. System safety approach <|ref|>sub_title<|/ref|><|det|>[[60, 309, 283, 332]]<|/det|> ## Key Words and Phrases <|ref|>text<|/ref|><|det|>[[60, 336, 255, 355]]<|/det|> Domino theory (p. 1.6) <|ref|>text<|/ref|><|det|>[[60, 419, 355, 439]]<|/det|> General methods of control (p. 1.7) <|ref|>text<|/ref|><|det|>[[60, 503, 310, 522]]<|/det|> Energy transfer theory (p. 1.8) <|ref|>text<|/ref|><|det|>[[60, 586, 448, 606]]<|/det|> Technique of operations review (TOR) (p. 1.9) <|ref|>text<|/ref|><|det|>[[60, 668, 250, 688]]<|/det|> System safety (p. 1.10) <|ref|>sub_title<|/ref|><|det|>[[60, 768, 234, 791]]<|/det|> ## Review Questions <|ref|>text<|/ref|><|det|>[[65, 798, 598, 837]]<|/det|> 2- 1. Identify the chain of accident factors set out in the domino theory. (p. 1.6)
[{"model_id": "deepseek-ai/DeepSeek-OCR", "model_name": "DeepSeek-OCR", "column_name": "markdown", "timestamp": "2026-02-21T19:32:27.962928", "prompt_mode": "document", "batch_size": 8, "max_tokens": 8192, "gpu_memory_utilization": 0.8, "max_model_len": 8192, "script": "deepseek-ocr-vllm.py", "script_url": "https://huggingface.co/datasets/uv-scripts/ocr/raw/main/deepseek-ocr-vllm.py"}]
<|ref|>text<|/ref|><|det|>[[45, 116, 875, 155]]<|/det|> 7- 3. These are two reasons why the distinction between excess and umbrella liability coverage is often unclear: <|ref|>text<|/ref|><|det|>[[90, 160, 870, 225]]<|/det|> Courts and many in the insurance profession use the terms interchangeably. Many insurers providing excess and umbrella liability insurance do not use standardized policies. <|ref|>sub_title<|/ref|><|det|>[[35, 236, 268, 258]]<|/det|> ## Educational Objective 8 <|ref|>text<|/ref|><|det|>[[45, 264, 654, 285]]<|/det|> 8- 1. An excess liability insurance policy may take any of three basic forms: <|ref|>text<|/ref|><|det|>[[90, 289, 760, 360]]<|/det|> A following- form policy subject to the same provisions as the underlying policy A self- contained policy subject to its own provisions only A combination of these two types <|ref|>text<|/ref|><|det|>[[45, 364, 880, 404]]<|/det|> 8- 2. A following- form excess liability policy will cover a claim in excess of the underlying limits only if the loss is covered by the underlying insurance. <|ref|>text<|/ref|><|det|>[[45, 407, 880, 516]]<|/det|> 8- 3. Coverage gaps may occur between the excess and underlying layers when a self- contained excess liability policy is used because a self- contained excess liability policy does not depend on the provisions of the underlying policies for determining the scope of coverage (except when specifically written to cover exhausted underlying limits). A self- contained excess liability policy applies to a loss that exceeds the limits of the underlying policy only if the loss is also covered under the provisions of the self- contained excess liability policy. <|ref|>text<|/ref|><|det|>[[45, 519, 880, 559]]<|/det|> 8- 4. One distinguishing feature of a true umbrella policy is a provision stating that the policy applies over a self- insured retention if the underlying policy does not cover a loss covered by the umbrella. <|ref|>text<|/ref|><|det|>[[45, 561, 877, 635]]<|/det|> 8- 5. A specific excess liability policy requires the insured to retain a stipulated amount of loss from the first dollar for all losses resulting from a single occurrence. An aggregate excess liability policy requires the insured to retain a specified amount of loss from the first dollar during a specified period of time, usually one year. <|ref|>sub_title<|/ref|><|det|>[[35, 646, 268, 668]]<|/det|> ## Educational Objective 9 <|ref|>text<|/ref|><|det|>[[45, 674, 877, 766]]<|/det|> 9- 1. Primary and umbrella layers are generally referred to as the working layers, because they are the layers most often called on to pay claims. In some cases, an insured must purchase a buffer layer of excess insurance between the primary layer and the umbrella policy. This approach is used when the umbrella insurer will not provide coverage unless the insured has underlying coverage limits higher than those that the primary insurer is willing to provide. <|ref|>text<|/ref|><|det|>[[45, 769, 585, 790]]<|/det|> 9- 2. These problems can occur when liability insurance is layered: <|ref|>text<|/ref|><|det|>[[90, 795, 878, 884]]<|/det|> Application of aggregate limits may vary in each layer. The excess policies may differ as to defense coverage. Excess layers over the first umbrella layer may purport to be following- form policies but in fact may contain restrictions not present in the umbrella layer. <|ref|>text<|/ref|><|det|>[[45, 887, 857, 927]]<|/det|> 9- 3. Risk managers of large corporations commonly buy the highest limits they can obtain and hope that those limits will be adequate.
[{"model_id": "deepseek-ai/DeepSeek-OCR", "model_name": "DeepSeek-OCR", "column_name": "markdown", "timestamp": "2026-02-21T19:32:27.962928", "prompt_mode": "document", "batch_size": 8, "max_tokens": 8192, "gpu_memory_utilization": 0.8, "max_model_len": 8192, "script": "deepseek-ocr-vllm.py", "script_url": "https://huggingface.co/datasets/uv-scripts/ocr/raw/main/deepseek-ocr-vllm.py"}]
<|ref|>text<|/ref|><|det|>[[33, 108, 675, 147]]<|/det|> specified types of losses. In general, consistent and effective control of noninsurance transfers requires the following: <|ref|>text<|/ref|><|det|>[[36, 154, 635, 230]]<|/det|> An analysis of factors affecting appropriate use of risk transfers A clearly written and widely disseminated organizational policy with general administrative controls and specific control measures for managing noninsurance transfers <|ref|>sub_title<|/ref|><|det|>[[36, 251, 617, 303]]<|/det|> ## Factors Affecting Appropriate Use of Noninsurance Risk Transfers <|ref|>text<|/ref|><|det|>[[34, 307, 656, 397]]<|/det|> Some risk management professionals use the negotiation of noninsurance transfers, especially those for risk financing, as a way to reduce the organization's liability loss exposures and loss costs. Such an approach is possible because contract provisions, especially liability- related hold- harmless agreements not connected with construction contracts, are largely unregulated. <|ref|>text<|/ref|><|det|>[[34, 404, 666, 494]]<|/det|> However, the goal of sound risk management is to use noninsurance transfers in ways that efficiently apportion loss exposures and the loss cost for the transferor, the transferee, and the economy as a whole. Here, efficiency refers to both organizational and economy- wide cost of risk. Therefore, the most efficient transfer lowers the cost of risk for each contracting organization. <|ref|>text<|/ref|><|det|>[[33, 500, 670, 609]]<|/det|> Because the parties drafting and using noninsurance transfers can benefit from the transfer's negotiation by obtaining more favorable terms than the other involved parties, a risk management professional should participate not as a competitor but as a referee, seeking fairness and mutual benefit for all participants. As such, the risk management professional should consider the following three factors. <|ref|>text<|/ref|><|det|>[[32, 616, 663, 715]]<|/det|> 1. The legal enforceability of contract provisions 2. The relative abilities of the parties to manage risk (that is, to keep losses from occurring and to pay for those that do) 3. The price or other legal consideration the transferor explicitly or implicitly pays or gives to the transferee <|ref|>sub_title<|/ref|><|det|>[[33, 735, 172, 757]]<|/det|> ## Enforceability <|ref|>text<|/ref|><|det|>[[30, 758, 680, 904]]<|/det|> Legal enforceability is one factor that determines a contract's appropriate use and whether the contract can be used at all. First, a contract is not enforceable if it is unconscionable or in violation of public policy or statutes. Second, attempts by the contracting parties to rid themselves of the same or related loss exposures or financial consequences could make it difficult to enforce the contract. For example, contracts in which risks have been transferred and retransferred can make it almost impossible for a court to tell who has agreed to accept what risk and may be unenforceable as a result.
[{"model_id": "deepseek-ai/DeepSeek-OCR", "model_name": "DeepSeek-OCR", "column_name": "markdown", "timestamp": "2026-02-21T19:32:27.962928", "prompt_mode": "document", "batch_size": 8, "max_tokens": 8192, "gpu_memory_utilization": 0.8, "max_model_len": 8192, "script": "deepseek-ocr-vllm.py", "script_url": "https://huggingface.co/datasets/uv-scripts/ocr/raw/main/deepseek-ocr-vllm.py"}]
<|ref|>text<|/ref|><|det|>[[33, 100, 485, 122]]<|/det|> 5- 5. Contrast diversifiable with nondiversifiable risk. <|ref|>text<|/ref|><|det|>[[33, 259, 339, 280]]<|/det|> 5- 6. Describe the quadrants of risk. <|ref|>sub_title<|/ref|><|det|>[[24, 429, 267, 454]]<|/det|> ## Application Question <|ref|>text<|/ref|><|det|>[[35, 460, 728, 500]]<|/det|> 5- 7. Classify each of these risks as pure or speculative, subjective or objective, and diversifiable or nondiversifiable: <|ref|>text<|/ref|><|det|>[[80, 505, 581, 527]]<|/det|> a. Damage to an office building resulting from a hurricane <|ref|>text<|/ref|><|det|>[[80, 664, 460, 685]]<|/det|> b. Reduction in value of retirement savings <|ref|>text<|/ref|><|det|>[[80, 823, 510, 844]]<|/det|> c. Products liability claim against a manufacturer
[{"model_id": "deepseek-ai/DeepSeek-OCR", "model_name": "DeepSeek-OCR", "column_name": "markdown", "timestamp": "2026-02-21T19:32:27.962928", "prompt_mode": "document", "batch_size": 8, "max_tokens": 8192, "gpu_memory_utilization": 0.8, "max_model_len": 8192, "script": "deepseek-ocr-vllm.py", "script_url": "https://huggingface.co/datasets/uv-scripts/ocr/raw/main/deepseek-ocr-vllm.py"}]
<|ref|>text<|/ref|><|det|>[[52, 100, 684, 209]]<|/det|> - The claims-made coverage form covers bodily injury or property damage that occurs after the retroactive date stated in the policy. The retroactive date stated in the policy can be the same as the policy inception date or can be earlier than the policy inception date, but only if a claim for the injury or damage is first made at some time during the policy period (or during an extended reporting period, if applicable). <|ref|>text<|/ref|><|det|>[[52, 215, 644, 252]]<|/det|> Apart from the coverage trigger, both forms provide the same coverages, which are summarized as follows: <|ref|>text<|/ref|><|det|>[[50, 259, 691, 490]]<|/det|> Coverage A—Bodily Injury and Property Damage Liability. The insurer agrees to pay on behalf of the insured those sums that the insured becomes legally obligated to pay as damages because of bodily injury or property damage covered by the policy. The insurer also agrees to defend the insured against claims or suits alleging such damages. Several exclusions define the scope of coverage. Among the most important exclusions are those applying to intended injury, injury to employees of the insured, pollution, aircraft, automobiles, watercraft, and damage to the insured's own work or products. The majority of excluded loss exposures are normally covered by other types of insurance, such as auto insurance or workers' compensation insurance. The exclusions are often used to eliminate duplicate coverage under two separate insurance policies, but the risk management professional must be familiar with the complementary coverages in all of these various insurance policies. <|ref|>text<|/ref|><|det|>[[50, 496, 692, 588]]<|/det|> Coverage B—Personal and Advertising Injury Liability. The insurer agrees to pay on behalf of the insured those sums that the insured becomes legally obligated to pay as damages because of personal and advertising injury to which the insurance applies. Personal and advertising injury includes such offenses as libel, slander, false arrest, wrongful eviction, and infringement of copyright. <|ref|>text<|/ref|><|det|>[[50, 594, 677, 668]]<|/det|> Coverage C—Medical Payments. Medical payments insurance covers medical expenses, irrespective of whether the insured organization is legally liable to pay them, for persons other than insureds who are injured on the insured's premises or because of the insured's operations. <|ref|>sub_title<|/ref|><|det|>[[50, 691, 228, 714]]<|/det|> ## Auto Insurance <|ref|>text<|/ref|><|det|>[[49, 717, 667, 828]]<|/det|> ISO commercial auto coverage forms include the Business Auto Coverage Form, the Garage Coverage Form, the Truckers Coverage Form, and the Motor Carriers Coverage Form. All four coverage forms can be used to provide auto liability insurance and auto physical damage insurance. Other coverages—such as medical payments, uninsured motorists, and personal injury protection—can be added by endorsement. <|ref|>text<|/ref|><|det|>[[49, 833, 694, 890]]<|/det|> The Business Auto Coverage Form is the most frequently used of the four ISO commercial auto forms. It is designed to meet the auto insurance needs of most types of organizations.
[{"model_id": "deepseek-ai/DeepSeek-OCR", "model_name": "DeepSeek-OCR", "column_name": "markdown", "timestamp": "2026-02-21T19:32:27.962928", "prompt_mode": "document", "batch_size": 8, "max_tokens": 8192, "gpu_memory_utilization": 0.8, "max_model_len": 8192, "script": "deepseek-ocr-vllm.py", "script_url": "https://huggingface.co/datasets/uv-scripts/ocr/raw/main/deepseek-ocr-vllm.py"}]

Document OCR using DeepSeek-OCR

This dataset contains markdown-formatted OCR results from images in andesco/risk-management-resources-images-sample using DeepSeek-OCR.

Processing Details

Configuration

  • Image Column: image
  • Output Column: markdown
  • Dataset Split: train
  • Batch Size: 8
  • Max Model Length: 8,192 tokens
  • Max Output Tokens: 8,192
  • GPU Memory Utilization: 80.0%

Model Information

DeepSeek-OCR is a state-of-the-art document OCR model that excels at:

  • LaTeX equations - Mathematical formulas preserved in LaTeX format
  • Tables - Extracted and formatted as HTML/markdown
  • Document structure - Headers, lists, and formatting maintained
  • Image grounding - Spatial layout and bounding box information
  • Complex layouts - Multi-column and hierarchical structures
  • Multilingual - Supports multiple languages

Dataset Structure

The dataset contains all original columns plus:

  • markdown: The extracted text in markdown format with preserved structure
  • inference_info: JSON list tracking all OCR models applied to this dataset

Usage 

from datasets import load_dataset
import json

# Load the dataset
dataset = load_dataset("{{output_dataset_id}}", split="train")

# Access the markdown text
for example in dataset:
    print(example["markdown"])
    break

# View all OCR models applied to this dataset
inference_info = json.loads(dataset[0]["inference_info"])
for info in inference_info:
    print(f"Column: {{info['column_name']}} - Model: {{info['model_id']}}")

Reproduction

This dataset was generated using the uv-scripts/ocr DeepSeek OCR vLLM script:

uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/deepseek-ocr-vllm.py \\
    andesco/risk-management-resources-images-sample \\
    <output-dataset> \\
    --image-column image

Performance

  • Processing Speed: ~0.1 images/second
  • Processing Method: Batch processing with vLLM (2-3x speedup over sequential)

Generated with UV Scripts
Downloads last month
13
Size of downloaded dataset files:
6.44 MB
Size of the auto-converted Parquet files:
6.44 MB
Number of rows:
10