artillerywu/DeepResearch-9K

Dataset

Data Splits

13 commits

1 linked in READMEs

updated Apr 21, 2026

See the code

README

Data Splits

The dataset consists of the following two subsets:

Dataset Name	Contents	How It Was Generated	Number of Samples
DeepResearch-9K	All samples (teacher model's outputs)	Teacher model inference on 9K questions	9,000
DeepResearch-Hard	Teacher model's incorrect samples only	Filtered from `DeepResearch-9K` (samples where the teacher model's final answer was wrong)	3,974

Based on the above, we define the following train/test split for our DeepResearch-R1 model:

Split	Composition	Source	Number of Samples
Train	All teacher-correct samples + Random 2,200 teacher-incorrect samples	From `DeepResearch-9K` (correct samples) + Random subset of `DeepResearch-Hard`	5,026 + 2,200 = 7,226
Test	The remaining teacher-incorrect samples (not used in training)	From `DeepResearch-Hard` (the other 1,774 samples)	1,774

Note:

Correct samples are those in DeepResearch-9K but not in DeepResearch-Hard.

The test set consists only of hard/incorrect cases (where the teacher model failed). This is by design to evaluate the model's ability to correct mistakes.

Contributors

artillerywu

13 commits

Similar repos

HuggingFaceH4/helpful_instructions_splits

Dataset

This splits the original [helpful_instructions dataset](https://huggingface.co/datasets/HuggingFace…

HuggingFaceH4/self_instruct

Dataset

This dataset splits the original [Self-instruct dataset](https://huggingface.co/datasets/yizhongw/s…

ankner/gsm8k-CoT

Dataset

HuggingFaceH4/test-cot

Dataset

HuggingFaceM4/VQAv2

Dataset

Checks with https://visualqa.org/download.html:

HuggingFaceH4/numina_60k_math_verify_correct_2_4gens_with_rm_scores

Dataset

Windy0822/ultrainteract_math_rollout

Dataset

open-r1/OpenR1-Math-220k

Dataset

OpenR1-Math-220k

801

Linked in READMEs

GNEHUY/Awesome-AgenticRAG_DeepResearch

🎯Awesome-AgenticRAG_DeepResearch: A curated list of resources on Agentic RAG & DeepResearch.…

README badge

artillerywu/DeepResearch-9K

Dataset

Data Splits

13 commits

1 linked in READMEs

updated Apr 21, 2026

See the code

README

Data Splits

The dataset consists of the following two subsets:

Dataset Name	Contents	How It Was Generated	Number of Samples
DeepResearch-9K	All samples (teacher model's outputs)	Teacher model inference on 9K questions	9,000
DeepResearch-Hard	Teacher model's incorrect samples only	Filtered from `DeepResearch-9K` (samples where the teacher model's final answer was wrong)	3,974

Based on the above, we define the following train/test split for our DeepResearch-R1 model:

Split	Composition	Source	Number of Samples
Train	All teacher-correct samples + Random 2,200 teacher-incorrect samples	From `DeepResearch-9K` (correct samples) + Random subset of `DeepResearch-Hard`	5,026 + 2,200 = 7,226
Test	The remaining teacher-incorrect samples (not used in training)	From `DeepResearch-Hard` (the other 1,774 samples)	1,774

Note:

Correct samples are those in DeepResearch-9K but not in DeepResearch-Hard.

The test set consists only of hard/incorrect cases (where the teacher model failed). This is by design to evaluate the model's ability to correct mistakes.