olmOCR: Unlocking Trillions of Tokens in PDFs with Vision Language Models
Paper β’ 2502.18443 β’ Published β’ 11
image imagewidth (px) 1.7k 1.7k | page_num int64 1 21 | source_file stringclasses 1
value | source_path stringclasses 1
value | total_pages int64 21 21 | markdown stringlengths 409 2.81k | markdown_metadata stringclasses 2
values | inference_info stringclasses 1
value |
|---|---|---|---|---|---|---|---|
1 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | β’ A licensee may deviate from the minimum requirements without specific AER approval if no royalty, equity, or reservoir engineering concerns are associated with the volumes being measured and the licensee is able to demonstrate that the alternative measurement equipment and/or procedures will provide measurement accur... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": false} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
2 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | β’ The proration factor is multiplied by the wellβs estimated production to determine the wellβs actual monthly production.
1.4 Single Point Measurement Uncertainty
βSingle point measurement uncertaintyβ relates to the limits applicable to equipment and/or procedures used to determine a single-phase specific volume at... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": false} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
3 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | 1.6.1 Example Calculation
Determination of single point measurement uncertainty for well oil (proration battery) using βroot sum squareβ methodology:
Individual uncertainties from historical AER research:
For oil/emulsion measurement,
Oil meter uncertainty = 0.5% (typical manufacturerβs specification)
Meter ... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": false} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
4 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | Figure 1.1 Total battery/facility oil (delivery point measurement)
Maximum uncertainty of monthly volume = N/A
The uncertainty of the monthly volume will vary, depending upon the number of individual measurements that are combined to yield the total monthly volume.
Single point measurement uncertainty:
Delivery ... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
5 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | The royalty trigger point for oil is at the wellhead; thus, delivery point measurements are required at the following locations:
β’ facility dispositions
β’ trucked-in receipts
β’ pipeline receipts
β’ sales
β’ Lease Automatic Custody Transfer (LACT)
Excluded: Test points and group points if they are not used for accountin... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
6 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | Maximum uncertainty of monthly volume (M)
> 16.9 \(10^3\) m\(^3\)/d = 5.0%
\( \leq 16.9\ 10^3\) m\(^3\)/d but > 0.50 \(10^3\) m\(^3\)/d = 10.0%
\( \leq 0.50\ 10^3\) m\(^3\)/d = 20.0%
Note that **M** is dependent upon combined deliveries, fuel, and vented gas measurement.
The maximum uncertainty of total m... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": false} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
7 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | Figure 1.3 Total battery water
Maximum uncertainty of monthly volume:
> 50 m^3/month = 5.0%
\( \leq 50 \) m^3/month = 20.0%
Single point measurement uncertainty = N/A
Total battery water may be determined by measurement or estimation, depending on production rates, so no basic requirement has been set for si... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
8 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | Figure 1.4 Well oil (proration battery)
Single point measurement uncertainty:
All classes = 2.0%
Maximum uncertainty of monthly volume (M):
Class 1 (high) > 30 m^3/d = 5.0%
Class 2 (medium) β€ 30 m^3/d but > 6 m^3/d = 10.0%
Class 3 (low) β€ 6 m^3/d but > 2 m^3/d = 20.0%
Class 4 (stripper) β€ 2 m^3/d... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
9 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | For figure 1.5,
\( \mathbf{m} = \) single point measurement uncertainty

Figure 1.5 Well gas (proration battery)
Single point measurement uncertainty:
\[
> 0.50\ 10^3\... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
10 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | volume may include estimates for solution gas dissolved in the test oil volume (gas-in-solution [GIS]), which may add to the monthly uncertainty. At the highest gas production rates, it is expected that the use of estimates will be minimal or at least have a minor impact on the accuracy of the total monthly gas volume,... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
11 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | Rather than being determined by continuous measurement, monthly oil well water production volumes are estimated from well tests and corrected by the use of proration factors to result in βactualβ volumes. The water rates determined during the well tests may be inferred from determining the water content of emulsion sam... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
12 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | Therefore, a stringent expectation is set for the single point measurement uncertainty. In some cases, this type of gas may be delivered to other plants for further processing or to injection facilities; thus, delivery point measurements are required at the following locations:
β’ gas plant dispositions
β’ sales to dow... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
13 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | Maximum uncertainty of monthly volume = N/A
The uncertainty of the monthly volume will vary, depending upon the number of individual measurements that are combined to yield the total monthly volume.
The term βdelivery point measurementβ for hydrocarbon liquids refers to the point at which the hydrocarbon liquid produc... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": false} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
14 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | M is dependent upon combined uncertainties of measured gas and gas equivalent of condensate.
M is dependent upon combined uncertainties of measured gas and gas equivalent of recombined condensate.
M is dependent upon uncertainties of measured gas only.
Figure 1.9 Plant inlet or total battery/group gas
Maximum unce... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
15 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | The equipment and/or procedures used to determine the measured gas volumes must be capable of meeting a 3.0 per cent single point measurement uncertainty.
(iv) Plant inlet or total battery/group condensate (recombined)
For figure 1.10,
m = single point measurement uncertainty
![Diagram showing the flow of gas an... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
16 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | Plant inlet condensate is typically separated from the inlet stream and sent through the plant for further processing. For reporting purposes, the gas equivalent of the plant inlet condensate is included in the total plant inlet gas volume. If total battery/group condensate upstream of the plant inlet is separated and ... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
17 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | For all upstream oil and gas facilities such as well sites, gas plants, batteries, and compressor sites, operators may estimate fuel gas use volumes for sites with an annual average fuel gas use of 0.50 \(10^3\) m\(^3\)/d or less. Therefore, the maximum uncertainty of the monthly volume is set at 20.0 per cent. For any... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
18 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | and vented volumes per facility exceed 0.5 \(10^3\) m\(^3\)/d (excluding pilot, purge, or dilution gas) must be metered.
Effective January 1, 2020, uncombusted gas released to the atmosphere that is not fugitive emissions must be reported as vent gas.
Sites requiring flare/vent gas metering may estimate up to 0.50 \(... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
19 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | Acid gas usually contains a great deal of water vapour and has other conditions associated with it, such as very low pressure that affects measurement accuracy. Therefore, the single point measurement uncertainty is set at 10.0 per cent.
When the acid gas is compressed and then injected into a well, the single point m... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
20 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | Figure 1.15 Well gas (well site separation)
Maximum uncertainty of monthly volume:
> 16.9 \( 10^3 \) m\(^3\)/d = 5.0%
\( \leq \) 16.9 \( 10^3 \) m\(^3\)/d = 10.0%
Single point measurement uncertainty = 3.0%
If production components from gas wells are separated and continuously measured, the maximum uncertai... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} | |
21 | Directive017.pdf | /home/dck/Programming/ducklake/vlm_llm_doc_processing/pdfs/Directive017.pdf | 21 | Figure 1.16 Well gas (effluent proration battery)
M is dependent upon well effluent measurement, correction by an effluent correction factor, and a proration factor.
Figure 1.17 Well gas (southeastern Alberta or other approved proration battery)
M is dependent upon well estimates and correction by a proration fact... | {"primary_language": "en", "is_rotation_valid": true, "rotation_correction": 0, "is_table": false, "is_diagram": true} | {"model": "allenai/olmOCR-2-7B-1025-FP8", "script": "olmocr2-vllm.py", "version": "1.0.0", "timestamp": "2025-10-25T01:24:37.611388", "batch_size": 16, "max_tokens": 8192, "temperature": 0.1} |
This dataset contains markdown-formatted OCR results from images in stckmn/ocr-input-Directive017-1761355294 using olmOCR-2-7B.
imagemarkdowntrainolmOCR-2-7B is a high-quality document OCR model based on Qwen2.5-VL-7B-Instruct, fine-tuned on olmOCR-mix-1025 dataset and optimized with GRPO reinforcement learning.
Key features:
Each row contains:
markdown: Extracted document content in markdown formatolmocr_metadata: JSON with document metadata (language, rotation, table/diagram flags)image: Original document imagemarkdown: Extracted text and structure in markdownolmocr_metadata: Document metadata (primary_language, is_rotation_valid, rotation_correction, is_table, is_diagram)inference_info: Processing metadata (model, script version, timestamp)# Using HF Jobs (recommended)
hf jobs uv run --flavor l4x1 \
-s HF_TOKEN \
https://huggingface.co/datasets/uv-scripts/ocr/raw/main/olmocr2-vllm.py \
stckmn/ocr-input-Directive017-1761355294 \
your-username/output-dataset
# Local with GPU
uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/olmocr2-vllm.py \
stckmn/ocr-input-Directive017-1761355294 \
your-username/output-dataset
@misc{olmocr,
title={{olmOCR: Unlocking Trillions of Tokens in PDFs with Vision Language Models}},
author={Jake Poznanski and Jon Borchardt and Jason Dunkelberger and Regan Huff and Daniel Lin and Aman Rangapur and Christopher Wilhelm and Kyle Lo and Luca Soldaini},
year={2025},
eprint={2502.18443},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2502.18443},
}
Generated with uv-scripts/ocr