datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
olm-CC-MAIN-2022-49-sampling-ratio-olm-0.15114822547
Dataset Card for OLM November/December 2022 Common Crawl
Cleaned and deduplicated pretraining dataset, created with the OLM repo here from 15% of the November/December 2022 Common Crawl snapshot.
Note: last_modified_timestamp was parsed from whatever a website returned in it's Last-Modified header; there are likely a small number of outliers that are incorrect, so we recommend removing the outliers before doing statistics with last_modified_timestamp.
rejection-sampling-QA
Rejecction Sampling Q&A
This dataset is a very small curated question-answer pairs.
The questions were hand-crafted to test the model's capabilities to follow instruction across various domains.
The answers were generated using Microsoft's Phi-2 and curated using OpenAssistant's Large DeBERTa v3 Reward Model v2.
Dataset Details
The answers of this dataset were generated by prompting Microsoft's Phi-2 using a prompt format inspired by Stanford's Alpaca to help the LLM… See the full description on the dataset page: https://huggingface.co/datasets/alehc/rejection-sampling-QA.demo_rejection_sampling_QA_phi-2_deberta-v3-large-v2_temp0.2This is a demo constructed dataset for alignment/preference learning.
With paritially handcrafted questions (prompts), the answers are genreated by the phi-2 model with temperature 0.2 and the answers are scores select by the deberta-large-v2.
The dataset containing questions and the selected answers from highest to lowest, decoding with rejection sampling K=8.
Example loading:
import datasets
ds = datasets.load_dataset('yizhilll/demo_rejection_sampling_QA_phi-2_deberta-v3-large-v2_temp0.2')… See the full description on the dataset page: https://huggingface.co/datasets/yizhilll/demo_rejection_sampling_QA_phi-2_deberta-v3-large-v2_temp0.2.R1-Distill-Qwen-14B-MATH_training500-sampling64
Model : deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
Original Dataset : MATH - first 500 queries in training split
Prompt:
{"role": "user", "content": "Please reason step by step, and put your final answer within \boxed{}." + '\n\n' + problem + '\n<think>\n'}
Sampling Parameters :
num_sampling=64
max_tokens=32768
temperature=0.6
top_p=0.95
‘correct’ : computed by the code in the link… See the full description on the dataset page: https://huggingface.co/datasets/HaeChan0305/R1-Distill-Qwen-14B-MATH_training500-sampling64.Qwen3-32B-MATH_training500-sampling64
Model : Qwen3-32B
Original Dataset : MATH - first 500 queries in training split
Prompt:
{"role": "user", "content": "Please reason step by step, and put your final answer within \boxed{}." + '\n\n' + problem}
Sampling Parameters :
num_sampling=64
max_tokens=38912
temperature=0.6
top_p=0.95
top_k=20
min_p=0
‘correct’ : computed by the code in the link (https://github.com/LeapLabTHU/Absolute-Zero-Reasoner/blob/master/absolute_zero_reasoner/rewards/math_utils.py)
R1-Distill-Qwen-1.5B-MATH_training500-sampling64
Model : deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
Original Dataset : MATH - first 500 queries in training split
Prompt:
{"role": "user", "content": "Please reason step by step, and put your final answer within \boxed{}." + '\n\n' + problem + '\n<think>\n'}
Sampling Parameters :
num_sampling=64
max_tokens=32768
temperature=0.6
top_p=0.95
‘correct’ : computed by the code in the link… See the full description on the dataset page: https://huggingface.co/datasets/HaeChan0305/R1-Distill-Qwen-1.5B-MATH_training500-sampling64.Qwen3-0.6B-AIME-2023-2024-2025-sampling64
Model : Qwen3-0.6B
Original Dataset : AIME2023, AIME2024, AIME2025
Prompt:
{"role": "user", "content": "Please reason step by step, and put your final answer within \boxed{}." + '\n\n' + problem}
Sampling Parameters :
num_sampling=64
max_tokens=38912
temperature=0.6
top_p=0.95
top_k=20
min_p=0
‘correct’ : computed by the code in the link (https://github.com/LeapLabTHU/Absolute-Zero-Reasoner/blob/master/absolute_zero_reasoner/rewards/math_utils.py)
Qwen3-32B-AIME-2023-2024-2025-sampling64
Model : Qwen3-32B
Original Dataset : first 24 queries in AIME2023 (시간 없어서 뒤에꺼 못함.)
Prompt:
{"role": "user", "content": "Please reason step by step, and put your final answer within \boxed{}." + '\n\n' + problem}
Sampling Parameters :
num_sampling=64
max_tokens=38912
temperature=0.6
top_p=0.95
top_k=20
min_p=0
‘correct’ : computed by the code in the link (https://github.com/LeapLabTHU/Absolute-Zero-Reasoner/blob/master/absolute_zero_reasoner/rewards/math_utils.py)
Qwen3-0.6B-MATH-sampling64
Model : Qwen3-0.6B
Original Dataset : MATH
train : first 500 queries in training split
test : MATH500
Prompt:
{"role": "user", "content": "Please reason step by step, and put your final answer within \boxed{}." + '\n\n' + problem}
Sampling Parameters :
num_sampling=64
max_tokens=38912
temperature=0.6
top_p=0.95
top_k=20
min_p=0
‘correct’ : computed by the code in the link (https://github.com/LeapLabTHU/Absolute-Zero-Reasoner/blob/master/absolute_zero_reasoner/rewards/math_utils.py)
Qwen2.5-MATH-1.5B-MATH-sampling8
Model : Qwen2.5-MATH-1.5B
Original Dataset : MATH
train : 12K
test : 500
Prompt:
[
{"role": "system", "content": "Please reason step by step, and put your final answer within \boxed{}."},
{"role": "user", "content": problem + "\n\nLet's think step by step and output the final answer within \boxed{}."}
]
Sampling Parameters :
num_sampling=8
max_tokens=4048
temperature=0.7
top_p=0.8
‘correct’ : computed by the code in the link… See the full description on the dataset page: https://huggingface.co/datasets/HaeChan0305/Qwen2.5-MATH-1.5B-MATH-sampling8.