TPTP axiom library
收藏资源简介:
该数据集是TPTP公理库,包含大量的公理,用于生成数学推理任务。数据集通过饱和引擎E-prover生成,保证了所有定理的数学有效性。数据集被用于构建三个难度可控的挑战:蕴涵验证、前提选择和证明重构。数据集的访问地址为https://github.com/sileod/reasoning_core。
This dataset is the TPTP axiom library containing a large number of axioms for generating mathematical reasoning tasks. It is generated by the saturation-based theorem prover E-prover, which guarantees the mathematical validity of all included theorems. This dataset is utilized to construct three difficulty-controllable challenges: implication verification, premise selection, and proof reconstruction. The access URL of this dataset is https://github.com/sileod/reasoning_core.
数据集概述
基本信息
- 数据集名称: reasoning core ◉
- 核心功能: 基于文本的RLVR(强化学习与验证推理)用于LLM(大语言模型)推理训练
- 核心任务类型: 符号化任务,包括完整的一阶逻辑(FOL)、基于TPTP的形式数学、新型领域的正式规划以及语法任务
安装与使用
主要环境安装
python #!pip install uv #install uv if needed !uv tool install prime --with openai -q !uv tool run prime -- env install sileod/reasoning-core-env
from verifiers import load_environment import os; from openai import OpenAI
env = load_environment("reasoning-core-env")
os.environ["OPENROUTER_API_KEY"] = "" #✍️ write your key client = OpenAI( base_url="https://openrouter.ai/api/v1", api_key=os.getenv("OPENROUTER_API_KEY")) results = env.evaluate(client=client, model="gpt-4.1-mini", num_examples=20, rollouts_per_example=1) df=env.make_dataset(results).to_pandas()
独立安装方式
python pip install reasoning_core
from reasoning_core import list_tasks, get_task, score_answer
T = get_task(arith)() x = T.generate_example() assert score_answer(x.answer, x)==1
数据生成
- 生成方式: 运行
bash run_generate.sh进行多线程生成到json文件 - 兼容性: 生成的文件可由Huggingface Datasets读取
与Reasoning Gym集成
- 接口特性: 使用比reasoning-gym(RG)更精简的自定义接口
- 任务正交性: 所有任务与RG正交,但可导入到RG中使用
- 集成示例: python import reasoning_gym
from reasoning_core import register_to_reasoning_gym register_to_reasoning_gym()
specs = [ # here, leg_counting tasks will make up two thirds of tasks DatasetSpec(name=leg_counting, weight=2, config={}), #from reasoning_gym 🏋 DatasetSpec(name=arith, weight=2, config={}), #from reasoning_core ◉ ] D=reasoning_gym.create_dataset(composite, size=10, seed=42, datasets=specs)

- 1Saturation-Driven Dataset Generation for LLM Mathematical Reasoning in the TPTP Ecosystem法国里尔大学,Inria,CNRS,Centrale Lille,UMR 9189-CRIStAL · 2025年



