marin-community/open-thoughts-4-6k-math-qwen3-235b-a22b-disagreed-answers
OpenThoughts4 6K Math - Qwen3-235B-A22B Disagreed Answers Overview This dataset contains 5,982 samples from the OpenThoughts4 math dataset where both Qwen3-32B and Qwen3-235B-A22B produced valid \boxed{} answers that differ. This dataset contains the Qwen3-235B-A22B reasoning traces. Relationship to Other Datasets This is one of 10 child datasets derived from two parent datasets: Parent datasets (29,963 samples each):… See the full description on the dataset page: https://huggingface.co/datasets/marin-community/open-thoughts-4-6k-math-qwen3-235b-a22b-disagreed-answers.
OpenThoughts4 6K Math - Qwen3-235B-A22B Disagreed Answers
Overview
This dataset contains 5,982 samples from the OpenThoughts4 math dataset where both Qwen3-32B and Qwen3-235B-A22B produced valid \boxed{} answers that differ. This dataset contains the Qwen3-235B-A22B reasoning traces.
Relationship to Other Datasets
This is one of 10 child datasets derived from two parent datasets:
Parent datasets (29,963 samples each):
marin-community/open-thoughts-4-30k-math-qwen3-32b-annotated-16384-tokensmarin-community/open-thoughts-4-30k-math-qwen3-235b-a22b-annotated
Child datasets (partitioned by answer availability and agreement):
The ms_id values are unique across all child datasets and match between paired 32B/235B datasets.
Dataset Schema
Use Cases
- Error analysis: Study where models diverge in their reasoning
- Ablation studies: Compare how incorrect answers propagate through training
- Difficulty analysis: These problems are harder (models disagree on answers)
