Feed & Grain

Datasets to fatten up your herd — streamed, sharded, and license-checked.

open-range/Trailhead-Web-2T

Text

Deduplicated, filtered web crawl for pretraining general-purpose herds.

2.1T tokens4.8 TB 812.0k

chuckwagon-ai/Chuckwagon-Instructions

Text

Human-reviewed instruction pairs across 22 domains, licensed for commercial use.

3.4M rows12 GB 402.1k

mesa-vision/HighPlains-Imagery

Image

Captioned aerial and ground imagery of arid landscapes, roads, and structures.

48M images31 TB 118.8k

canyon-audio/Campfire-Speech

Audio

Multilingual conversational speech recorded in noisy, real-world settings.

94k hours8.2 TB 96.5k