zechen-nlp/math-ai-bench-pro
收藏Hugging Face2026-04-08 更新2026-04-12 收录
下载链接：
https://hf-mirror.com/datasets/zechen-nlp/math-ai-bench-pro
下载链接
链接失效反馈官方服务：
资源简介：
---
configs:
- config_name: aime25
default: true
data_files:
- split: train
path: aime25/train-00000-of-00001.parquet
- config_name: aime26
data_files:
- split: train
path: aime26/train-*
- config_name: apex_2025
data_files:
- split: train
path: apex_2025/train-00000-of-00001.parquet
- config_name: arxivmath
data_files:
- split: train
path: arxivmath/train-00000-of-00001.parquet
- config_name: cmimc_2025
data_files:
- split: train
path: cmimc_2025/train-00000-of-00001.parquet
- config_name: gpqa_diamond
data_files:
- split: train
path: gpqa_diamond/train-00000-of-00001.parquet
- config_name: hmmt_feb_2026
data_files:
- split: train
path: hmmt_feb_2026/train-00000-of-00001.parquet
- config_name: hmmt_nov_2025
data_files:
- split: train
path: hmmt_nov_2025/train-00000-of-00001.parquet
- config_name: imobench
data_files:
- split: train
path:
- imobench/train-00000-of-00002.parquet
- imobench/train-00001-of-00002.parquet
- config_name: olympiadbench
data_files:
- split: train
path:
- olympiadbench/train-00000-of-00002.parquet
- olympiadbench/train-00001-of-00002.parquet
- config_name: theoremqa
data_files:
- split: train
path: theoremqa/train-00000-of-00001.parquet
dataset_info:
config_name: aime26
features:
- name: problem
dtype: string
- name: original_solution
dtype: string
- name: answer
dtype: string
- name: source
dtype: string
- name: index
dtype: int64
- name: model
dtype: string
- name: generated_solutions
list: string
- name: count
dtype: int64
- name: context_metadata
struct:
- name: cot_num_token
list: int64
- name: total_num_token
dtype: int64
splits:
- name: train
num_bytes: 33727102
num_examples: 120
download_size: 33627058
dataset_size: 33727102
---
# math-ai-bench-pro
`math-ai-bench-pro` is a per-source split of `haowu89/math-ai-bench-sources-latest`.
Each row keeps the original benchmark fields and adds:
- `context_metadata`: token counts for `generated_solutions`, computed with `tiktoken` `o200k_base`
- `original_solution`: copied over where an aligned subset existed in `zechen-nlp/math_ai_bench_cot`
Available configs:
- `aime25`
- `aime26`
- `apex_2025`
- `arxivmath`
- `cmimc_2025`
- `gpqa_diamond`
- `hmmt_feb_2026`
- `hmmt_nov_2025`
- `imobench`
- `olympiadbench`
- `theoremqa`
提供机构：
zechen-nlp
