Skip to content

Dataset · Alibaba-Apsara

Reasoning log-probabilities from gpt-oss-120b model

Alibaba-Apsara/Superior-Reasoning-SFT-gpt-oss-120b-Logprob

Token-level log-probabilities from the gpt-oss-120b teacher model for distilling English reasoning skills.

What's inside

This is an auxiliary English dataset ranging in size from 100k to 1 million records. It contains token-level log-probabilities generated by the gpt-oss-120b teacher model for examples from the main Superior-Reasoning-SFT-gpt-oss-120b set. Records are linked to the main corpus via the unique sample_uuid identifier.

What you can do with it

  • Knowledge distillation from the gpt-oss-120b model into more compact LLMs
  • Training models on complex reasoning chains based on the teacher probability distribution
Tasks
Text generation
Languages
English
Size
100K–1M rows
Downloads this month
855
Likes
63

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("Alibaba-Apsara/Superior-Reasoning-SFT-gpt-oss-120b-Logprob")