rustforge-personal-rust-dataset
收藏资源简介:
RustForge Personal Rust Dataset 是一个专门为训练 Rust 专用编码模型而设计的合成数据集。该数据集包含 50,000 条 ChatML 格式的记录,覆盖了 13 个不同的 Rust 任务类别,包括编译修复、语义实现、bug 修复、版本迁移、异步编程、不安全代码、宏更新、API 重构等。数据集采用分片 JSONL 存储,目标 Rust 版本为 2024。构建过程中采用了分层验证和质量控制措施,包括语义去重、分级执行验证(cheap、medium、full)以及基于模板家族的审计。数据集适用于监督微调 Rust 导向的编码模型,特别适合需要在 crate 或工作区上下文中编辑或生成 Rust 代码的模型。当前版本中,所有 13 个类别均达到或超过 A 级质量标准,全局家族深度下限为 800。
RustForge Personal Rust Dataset is a synthetic dataset specifically designed for training Rust-specific coding models. This dataset contains 50,000 ChatML-formatted records, covering 13 distinct Rust task categories including compilation repair, semantic implementation, bug fixing, version migration, asynchronous programming, unsafe code, macro updates, API refactoring and more. The dataset is stored in sharded JSONL format, with the target Rust version set to 2024. During its construction, layered validation and quality control measures were employed, including semantic deduplication, hierarchical execution validation (cheap, medium, full), and template-family-based auditing. The dataset is applicable for supervised fine-tuning of Rust-oriented coding models, and is particularly well-suited for models that need to edit or generate Rust code within the context of a crate or workspace. In the current version, all 13 task categories have met or exceeded Grade A quality standards, with the global family depth lower bound set at 800.




