Alternative data datasets
Browse by provider, category, marketplace, access, delivery and license
Datamule, Teraflop AI, and Eventual jointly contributed a 590 gigabyte SEC EDGAR collection encompassing about 8 million filings and 43 billion tokens, assembled using the datamule-python library and the official datamule API.
A large-scale release produced jointly by Datamule, Teraflop AI, and Eventual covering roughly 590 GB of SEC EDGAR content. It comprises about eight million filings totaling 43 billion tokens, gathered using the datamule-python package and the official datamule API.
PatentMatch pairs US patents with matching reference patents for retrieval evaluation, a few thousand records in JSON (Apache 2.0). A niche evaluation set for patent-similarity and citation-link models.
Chinese patents paired with their abstractive summaries in Mandarin, a few thousand records in JSON (Apache 2.0). A window into the pace and direction of patenting inside the Chinese technology base.
Records of initial public offerings launched on Indian markets between 2006 and 2025, with fields covering open and close dates, listing date, face value, issue price and size, lot size, first-day price, total shares offered and their allocation across anchor, NII, QIB and retail categories, minimum investment, and subscription figures for each investor class.
A tabular extract of clinical study descriptions drawn from ClinicalTrials.gov on 5 February 2025, emphasizing eligibility, design, and objective fields for healthcare and machine learning research.
Vector embeddings of clinical-trial records, hundreds of thousands of entries in Parquet (Apache 2.0). Enables similarity and retrieval analysis of trial design, indication crowding and competitive positioning.
DerivedFunction01's sec-filings-snippets is a public-records dataset of short text excerpts drawn from U.S. SEC filings, released on Hugging Face in Parquet format. It is sized for language-model fill-mask work and is not positioned as a commercial alternative-data product.
A large granted-patent text collection from the USPTO, tens of millions of records in Parquet (Apache 2.0). Broad coverage of claims and descriptions for IP-intensity and technology-trend measurement.
Generated with the LeRobot framework, this dataset comprises 80 episodes demonstrating a pick-and-place routine using a single orange cube positioned at varying locations within the workspace. Each episode requires the robot to rotate toward the cube, open its gripper, close it around the object, and transport it to a specified drop zone.
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v3.0", "fps": 30, "features": { "action": { "dtype": "float32", "names": [ "shoulder_pan.pos", "shoulder_lift.pos", "elbow_flex.pos", "wrist_flex.pos", "wrist_roll.pos", "gripper.pos" ], "shape": [ 6… See the full description on the dataset page: https://huggingface.co/datasets/Edgarium/rangement_pq_20
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v2.1", "robot_type": "so101_follower", "total_episodes": 24, "total_frames": 5358, "total_tasks": 1, "total_videos": 48, "total_chunks": 1, "chunks_size": 1000, "fps": 30, "splits": { "train": "0:24" }, "data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet", "video_path":… Se
A LeRobot-formatted dataset captured with a so101_follower robot, comprising 30 episodes, 6,788 frames, 60 video files, and a single chunk at 30 fps, with training covering the full episode range.
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v2.1", "robot_type": "so101_follower", "total_episodes": 50, "total_frames": 13379, "total_tasks": 1, "total_videos": 100, "total_chunks": 1, "chunks_size": 1000, "fps": 30, "splits": { "train": "0:50" }, "data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet", "video_path":…
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v2.1", "robot_type": "so101_follower", "total_episodes": 709, "total_frames": 187905, "total_tasks": 1, "total_videos": 1418, "total_chunks": 1, "chunks_size": 1000, "fps": 30, "splits": { "train": "0:709" }, "data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet", "video_path
A robotics dataset generated with the LeRobot framework contains two recorded episodes totaling 897 frames at 30 frames per second, captured using an SO100 robot performing a single task, with four video files and accompanying parquet data organized in a single chunk of up to 1,000 entries and partitioned as a train split covering episodes zero through one.
Another LeRobot so100 robotics dataset containing 50 episodes, 29,874 frames, and 100 videos at 30 fps, structured identically to the related edgar-block release.
A LeRobot dataset captured on the so100 robot platform, consisting of 50 episodes, 44,746 frames, and 100 videos recorded at 30 fps and split for training.
A robotics dataset assembled with the LeRobot framework for the SO100 platform, containing 20 episodes totaling 11,940 frames captured at 30 fps, stored in parquet data files alongside 40 corresponding video chunks.
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v2.1", "robot_type": "so101_follower", "total_episodes": 50, "total_frames": 12620, "total_tasks": 1, "total_videos": 100, "total_chunks": 1, "chunks_size": 1000, "fps": 30, "splits": { "train": "0:50" }, "data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet", "video_path":…
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v2.1", "robot_type": "so101_follower", "total_episodes": 500, "total_frames": 136475, "total_tasks": 1, "total_videos": 1000, "total_chunks": 1, "chunks_size": 1000, "fps": 30, "splits": { "train": "0:500" }, "data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet", "video_path
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v2.1", "robot_type": "so101_follower", "total_episodes": 500, "total_frames": 137269, "total_tasks": 1, "total_videos": 1000, "total_chunks": 1, "chunks_size": 1000, "fps": 30, "splits": { "train": "0:500" }, "data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet", "video_path
A LeRobot-formatted robotics capture set for the so101_follower arm, comprising 51 episodes, 13,109 frames, and 102 videos at 30 fps stored across parquet chunks.
A LeRobot-generated dataset comprising 149 episodes, 27,628 frames, and 298 video files captured at 30 fps on a so101_follower robot, with all episodes assigned to the training split and stored in the chunk-based Parquet layout expected by the codebase.
高质量中文专利摘要数据集。
View source information and access options.
A cleaned collection of French patent documents published from 1981 through 2026, each row in the Parquet file representing one complete A1 publication parsed from the original XML. It was assembled independently by a contributor with direct access to the public patent source and is optimized for distributed loading.
A joint release by Datamule, Teraflop AI, and Eventual provides the SEC-EDGAR dataset, comprising 590 GB of content drawn from all major filings in the SEC EDGAR database and totaling roughly 8 million samples and 43 billion tokens. The data was assembled using the datamule-python library and the official datamule API developed by John Friedman, with datamule serving as a Python package for collecting and manipulating such records.
Datamule partnered with Teraflop AI and Eventual to publish the SEC-EDGAR dataset, which covers all major filings from the SEC EDGAR database and amounts to 590 GB across about 8 million samples and 43 billion tokens. Collection relied on the datamule-python library and John Friedman's official datamule API, tools designed in Python for gathering and handling SEC filing data.
Restaurant_Reviews.tsv is a small text corpus of customer restaurant reviews paired with a binary sentiment label. It is published on the Hugging Face Hub by user KarthikaRajagopal under an unstated license.
View source information and access options.
The restaurant_reviews dataset is a small text corpus of roughly one thousand customer review snippets paired with binary sentiment labels, published on the Hugging Face Hub under the user account pachequinho. It is distributed in CSV format with an Apache-2.0 license declaration.
Patent_FR_US_Merge_Radix_65536 is a Hugging Face dataset by publisher PhysiQuanty that merges French and US patent documents into a single text corpus. It is distributed in Parquet format and is tagged with an Apache-2.0 license, though the README provides no further description.
Text of SEC filings pulled from EDGAR spanning millions of records in English (Apache 2.0): 10-K, 10-Q, 8-K, annual reports and exhibits, tagged for text-generation and classification. A practical base corpus for earnings- and filings-driven signals, from risk factors and MD&A to management commentary.
View source information and access options.
ZipLime US Insider Trading Disclosures (PIT) [!CAUTION] Use knowledge_date, not transaction_date, when backtesting. transaction_date says when a trade occurred; knowledge_date says when the filing became observable through EDGAR. Using the former as the signal date introduces look-ahead bias. visible = trades.filter(pl.col("knowledge_date") <= simulation_time) This dataset normalizes corporate-ins
A curated collection of SEC EDGAR filings from recent and upcoming IPO companies, containing 5,179 text segments sourced from 261 filings, formatted for large language model training and financial analysis.