Skip to content

Dataset · ProCreations

Beam-1.0-501B mathematics and science reasoning traces

ProCreations/Beam-Reasoning-Traces

Verified reasoning chains from the large Beam-1.0 model with varying thinking depth levels for distillation.

What's inside

Containing between 100 thousand and 1 million English records, the dataset features verified mathematics and natural science reasoning chains from the Beam-1.0-501B-A23B model. Most records are gathered at the medium effort level, but there are also runs of the same prompts at low, high, and xhigh levels, with evaluation set answers spanning the full range from minimal to max.

What you can do with it

  • Distill deep reasoning skills into compact language models
  • Train models with adjustable depth and duration of thought
  • SFT on verified mathematics and natural science tasks
Tasks
Text generation, Question answering
Languages
English
Size
100K–1M rows
Downloads this month
757
Likes
5

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("ProCreations/Beam-Reasoning-Traces")