SupraLabs/reasoning-corpus-4K-5M-v1
收藏资源简介:
该数据集包含来自主流AI模型的推理链,例如:DeepSeek-v4(包括Pro和Flash版本)、DeepSeek-r1(DS-r1、Llama-DS、Qwen-DS)、Qwen3、Qwen3.5/3.6(包括开源和API模型)、Gemma4-31B等,这些数据来源于许多其他仓库,并经过适当过滤以用于训练SLM(小语言模型)。数据集包含以下列供用户筛选:repo_id、tok_len、user、thought_trace、assistant、ChatML。它是一个混合了多个较小仓库的推理语料库,总规模在500万序列长度以内,专注于推理、思维链(CoT)、代码和代理思维等任务。
This dataset comprises reasoning chains derived from mainstream AI models, such as DeepSeek-v4 (covering Pro and Flash variants), DeepSeek-r1 (including DS-r1, Llama-DS, Qwen-DS), Qwen3, Qwen3.5/3.6 (encompassing both open-source and API-deployed models), Gemma4-31B, and others. The data is sourced from multiple external repositories and properly filtered for training Small Language Models (SLMs). The dataset offers the following columns for users to conduct filtering: repo_id, tok_len, user, thought_trace, assistant, and ChatML. This is a reasoning corpus aggregated from several smaller repositories, with a total sequence length not exceeding 5 million, focusing on tasks including reasoning, Chain-of-Thought (CoT), code, and AI Agent thinking.




