datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
olmOCR-bench
olmOCR-bench
olmOCR-bench is a dataset of 1,403 PDF files, plus 7,010 unit test cases that capture properties of the output that a good OCR system should have.
This benchmark evaluates the ability of OCR systems to accurately convert PDF documents to markdown format while preserving critical textual and structural information.
Quick links:
📃 Paper
🛠️ Code
🎮 Demo
Table 1. Distribution of Test Classes by Document Source
Document Source
Text Present
Text… See the full description on the dataset page: https://huggingface.co/datasets/allenai/olmOCR-bench.olmo-mix-1124
OLMo 2 (November 2024) Pretraining set
Collection of data used to train OLMo-2-1124 models. The majority of this dataset comes from DCLM-Baseline with no additional filtering, but we provide the explicit breakdowns below.
Name
Tokens
Bytes (uncompressed)
Documents
License
DCLM-Baseline
3.70T
21.3TB
2.95B
CC-BY-4.0
Arxiv
20.8B
77.2GB
3.95M
ODC-BY
pes2o
58.6B
412GB
38M
ODC-BY
starcoder
83.0B
458GB
78.7M
ODC-BY
Algebraic-stack
11.8B
44.0GB
2.83M
ODC-BY… See the full description on the dataset page: https://huggingface.co/datasets/allenai/olmo-mix-1124.olmocr-pre-rendered
olmOCR-bench Pre-Rendered
Pre-rendered PNG images of the olmOCR-bench benchmark dataset, ready for zero-setup evaluation of any OCR / vision model.
What This Is
The official olmOCR benchmark requires downloading 1,403 PDFs locally and rendering each page to a PNG image before sending it to a model. Every benchmark runner in the official repo does this same rendering step internally — see olmocr/data/renderpdf.py::render_pdf_to_base64png().
This dataset eliminates that… See the full description on the dataset page: https://huggingface.co/datasets/shhdwi/olmocr-pre-rendered.olmOCR-synthmix-1025
olmOCR-synthmix-1025
olmOCR-synthmix-1025 is a dataset of 2,186 single PDF pages, that have been synthetically rerendered into HTML by
claude-sonnet-4-20250514.
In total, across these PDF pages, 30,381 synthetic benchmark cases have been created, following the format of olmOCR-bench.
These documents contain no overlap with the original olmOCR-bench documents, and thus can be used as RLVR training
data to improve the performance of OCR engines.
Directory Structure… See the full description on the dataset page: https://huggingface.co/datasets/allenai/olmOCR-synthmix-1025.dolmino_olmocr_pdfsOLMoE-mix-0924
OLMoE Mix (September 2024)
The following data mix was used to train OLMoE-1B-7B, a Mixture-of-Experts LLM with 1B active and 7B total parameters released in September 2024.
The base version of OLMoE-1B-7B can be found at this page, the SFT of OLMoE-1B-7B is available here, and a version combining SFT and DPO is available following this link.
Statistics
Subset
Tokens
Words
Bytes
Docs
DCLM Baseline 1.0
3.86 T
3.38 T16.7 T
2.95 B
Starcoder
101 B
63.9 B… See the full description on the dataset page: https://huggingface.co/datasets/allenai/OLMoE-mix-0924.olmoearth-paper-embeddings
OlmoEarth — Foundation-Model Embeddings for Paper Table 2
This dataset contains pre-extracted embeddings from 26 Earth-observation
foundation models evaluated on the 24 downstream tasks that make up
Table 2 of the OlmoEarth paper:
OlmoEarth: Stable Latent Image Modeling for Multimodal Earth Observation
AI2, 2025. arXiv:2511.13655.
For every supported (model, task) pair we ran the model's encoder over the
task's train / validation / test splits with the paper-best… See the full description on the dataset page: https://huggingface.co/datasets/allenai/olmoearth-paper-embeddings.olmoearth_pretrain_datasetThis is the pre-training dataset for training the OlmoEarth pre-trained remote sensing foundation models.
Documentation is on GitHub at https://github.com/allenai/olmoearth_pretrain/blob/main/docs/Pretraining-Dataset.md
The dataset is released under CC BY 4.0. It includes data from the following sources:
Sentinel-2 L2A imagery from the European Space Agency, available under the Copernicus Sentinel Data and Service Legal Notice
Sentinel-1 GRD IW vv+vh imagery from the European Space Agency… See the full description on the dataset page: https://huggingface.co/datasets/allenai/olmoearth_pretrain_dataset.olmOCR-bench-1.5-preview
olmOCR-bench-1.5-preview
olmOCR-bench-1.5-preview is a preview follow up to the original olmOCR-bench that adds several
new synthetic benchmark categories.
In addition to the original 1,403 PDF files, plus 7,010 unit test cases that were manually created as part of olmOCR-bench, this repo contains additional, synthetic tests designed to test difficult OCR scenarios. In all synthetic cases, we sample PDFs from the same distribution as in dolma3_mix-6T, then rerender them using… See the full description on the dataset page: https://huggingface.co/datasets/allenai/olmOCR-bench-1.5-preview.CerebRM-olmo-3-7b-instruct-sft-list_em-so1_completions
TRL Completion logs
This dataset contains the completions generated during training using trl.
The completions are stored in parquet files, and each file contains the completions for a single step of training (depending on the logging_steps argument).
Each file contains the following columns:
step: the step of training
prompt: the prompt used to generate the completion
completion: the completion generated by the model
<reward_function_name>: the reward(s) assigned to the… See the full description on the dataset page: https://huggingface.co/datasets/wetsoledrysoul/CerebRM-olmo-3-7b-instruct-sft-list_em-so1_completions.olm-wikipedia-20220920
Dataset Card for OLM September 2022 Wikipedia
Pretraining dataset, created with the OLM repo here from a September 2022 Wikipedia snapshot.
olm-CC-MAIN-2022-21-sampling-ratio-0.14775510204
Dataset Card for OLM May 2022 Common Crawl
Cleaned and deduplicated pretraining dataset, created with the OLM repo here from 15% of the May 2022 Common Crawl snapshot.
Note: last_modified_timestamp was parsed from whatever a website returned in it's Last-Modified header; there are likely a small number of outliers that are incorrect, so we recommend removing the outliers before doing statistics with last_modified_timestamp.
olmo-activationsolm-wikipedia-20221220
Dataset Card for OLM December 2022 Wikipedia
Pretraining dataset, created with the OLM repo here from a December 2022 Wikipedia snapshot.
olm-CC-MAIN-2022-33-sampling-ratio-0.20
Dataset Card for OLM August 2022 Common Crawl
Cleaned and deduplicated pretraining dataset, created with the OLM repo here from 20% of the August 2022 Common Crawl snapshot.
Note: last_modified_timestamp was parsed from whatever a website returned in it's Last-Modified header; there are likely a small number of outliers that are incorrect, so we recommend removing the outliers before doing statistics with last_modified_timestamp.
olmOCR-mix-1025
olmOCR-mix-1025
olmOCR-mix-1025 is a dataset of ~270,000 PDF pages which have been OCRed into plain-text in a natural reading order using gpt-4.1 and a special
prompting strategy that preserves any born-digital content from each page.
This dataset can be used to train, fine-tune, or evaluate your own OCR document pipeline, and all PDF pages used are included for download.
Compared to olmOCR-mix-0225, this dataset includes:
Cleaner outputs processed with gpt-4.1
More consistent… See the full description on the dataset page: https://huggingface.co/datasets/allenai/olmOCR-mix-1025.olm-CC-MAIN-2022-27-sampling-ratio-0.16142697881
Dataset Card for OLM June/July 2022 Common Crawl
Cleaned and deduplicated pretraining dataset, created with the OLM repo here from 16% of the June/July 2022 Common Crawl snapshot.
Note: last_modified_timestamp was parsed from whatever a website returned in it's Last-Modified header; there are likely a small number of outliers that are incorrect, so we recommend removing the outliers before doing statistics with last_modified_timestamp.
starcoder-python-instruct
StarCoder-Python-Qwen-Instruct
Dataset Description
This dataset contains Python code samples paired with synthetically generated natural language instructions. It is designed for supervised fine-tuning of language models for code generation tasks. The dataset is derived from the Python subset of the bigcode/starcoderdata corpus, and the instructional text for each code sample was generated using the Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8 model.
Creation… See the full description on the dataset page: https://huggingface.co/datasets/OLMo-Coding/starcoder-python-instruct.olmOCR_bench
Dataset Card for olmocr-bench
This is a FiftyOne dataset with 7019 samples.
Installation
If you haven't already, install FiftyOne:
pip install -U fiftyone
Usage
import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub
# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("Voxel51/olmOCR_bench")
# Launch the App
session = fo.launch_app(dataset)
Here is the completed dataset card, filled in… See the full description on the dataset page: https://huggingface.co/datasets/Voxel51/olmOCR_bench.olm-CC-MAIN-2022-49-sampling-ratio-olm-0.15114822547
Dataset Card for OLM November/December 2022 Common Crawl
Cleaned and deduplicated pretraining dataset, created with the OLM repo here from 15% of the November/December 2022 Common Crawl snapshot.
Note: last_modified_timestamp was parsed from whatever a website returned in it's Last-Modified header; there are likely a small number of outliers that are incorrect, so we recommend removing the outliers before doing statistics with last_modified_timestamp.
olmo-3-preference-mix-deltas_reasoning-yolo_scottmix-DECON-multi-turntulu-3-sft-olmo-2-mixture-0225Used to train OLMo 2 32B. From the blog post:
Filtered out instructions from the SFT dataset and the chosen responses of the preference data that included mentions of a date cutoff from the synthetic data generation process. This resulted in a new version of the instruction dataset, Tulu 3 SFT Mixture 0225, and preference dataset, OLMo-2-32B-pref-mix-0325.
We use majority voting to improve the quality of answers to our synthetic math questions. For our Persona MATH and Grade School Math… See the full description on the dataset page: https://huggingface.co/datasets/allenai/tulu-3-sft-olmo-2-mixture-0225.olm-CC-MAIN-2017-22-sampling-ratio-0.16178770949
Dataset Card for OLM May 2017 Common Crawl
Cleaned and deduplicated pretraining dataset, created with the OLM repo here from 16% of the May 2017 Common Crawl snapshot.
Note: last_modified_timestamp was parsed from whatever a website returned in it's Last-Modified header; there are likely a small number of outliers that are incorrect, so we recommend removing the outliers before doing statistics with last_modified_timestamp.
olmoearth_lcc
OlmoEarth Land Cover Change (LCC) Dataset
This dataset contains point-based annotations of land cover change, used to
train the OlmoEarth LCC model (https://olmoearth-lcc.allen.ai). The model
detects recent land cover change from Sentinel-2 time series: it inputs a Sentinel-2
image time series with 16 quarterly images (to establish a historical baseline) and 4
recent biweekly images (to detect changes soon after they occur) and predicts, per pixel,
whether a land cover change… See the full description on the dataset page: https://huggingface.co/datasets/allenai/olmoearth_lcc.olm-CC-MAIN-2022-40-sampling-ratio-0.15894621295-seed-69olmOCR-bench
olmOCR-bench
olmOCR-bench is a dataset of 1,403 PDF files, plus 7,010 unit test cases that capture properties of the output that a good OCR system should have.
This benchmark evaluates the ability of OCR systems to accurately convert PDF documents to markdown format while preserving critical textual and structural information.
Quick links:
📃 Paper
🛠️ Code
🎮 Demo
Table 1. Distribution of Test Classes by Document Source
Document Source
Text Present
Text… See the full description on the dataset page: https://huggingface.co/datasets/endsieg97/olmOCR-bench.olm-october-2022-tokenized-128
Dataset Card for "olm-october-2022-tokenized-128"
More Information needed
OLMo-2-2.7B-Exp-NoiseVectors
OLMo-2-2.7B-Exp Noise Vectors
Gaussian noise vectors added to the input embeddings during pretraining of
sbordt/OLMo-2-2.7B-Exp
(a 2.7B-parameter OLMo-2-style model with d_model=2880). Released as a
uniform-random 1% subsample per every-1000-batch chunk from 51,200 poisoned
pretraining batches over 100,000 training steps — 480 rows total.
How the noise was applied during training
For each poisoned batch, Gaussian noise of shape (4096, 2880) was
drawn and added to… See the full description on the dataset page: https://huggingface.co/datasets/sbordt/OLMo-2-2.7B-Exp-NoiseVectors.OLMo-2_Residual_Streams
OLMo-2_Residual_Streams
This dataset contains approximately 600 million residual streams derived from the FineWeb dataset. The residual streams were extracted using the allenai/OLMo-2-1124-7B-Instruct model and are stored in .parquet format.
Dataset Description
Usage
The dataset should work just fine with load_dataset:
>>> from datasets import load_dataset
# after loading the data, cast to bf16 using torch.view()
>>> dataset =… See the full description on the dataset page: https://huggingface.co/datasets/open-concept-steering/OLMo-2_Residual_Streams.olm-october-2022-tokenized-1024
Dataset Card for "olm-october-2022-tokenized-1024"
More Information needed