pre-to-post-olmo/

openthoughts_merged_think_39k

Merged think-format SFT dataset (ShareGPT-style: system + conversations with from/value), 39,874 examples, for OLMo SFT. Composition (concatenation of two decontaminated think-format sources):

open-thoughts114k_math_20k_decontam_think — 20,000 examples sampled from OpenThoughts-114k math, decontaminated against the OpenThoughts3 set below. openthoughts3_math_decontam_resp_lt8192_think — 19,874 examples from… See the full description on the dataset page: https://huggingface.co/datasets/pre-to-post-olmo/openthoughts_merged_think_39k.