marin-community/synth-patch-diff-ppl

该数据集是一个多任务代码仓库相关数据集，包含六个配置：1) commit_message_metadata：用于生成或分析提交消息的元数据任务，包含文件更改数量、仓库信息等；2) failing_test_trace_to_patch：将失败测试追踪映射到补丁的任务，包含失败测试和错误信息；3) file_path_line_refs：处理文件路径和行引用的任务，包含符号和行号信息；4) gh_pr_event_patch：基于GitHub拉取请求事件生成补丁的任务，包含事件类型和PR编号；5) review_comment_threads：处理代码审查评论线程的任务，包含线程状态和行信息；6) unified_diff_hunks：处理统一差异块的任务，包含错误类型和块头信息。所有配置均用于验证目的，每个配置包含1000个示例，涉及多种编程语言和许可证，旨在支持代码生成、修复和分析任务。

This dataset is a multi-task code repository-related dataset comprising six configurations: 1) commit_message_metadata: for generating or analyzing commit message metadata, including files changed, repository info, etc.; 2) failing_test_trace_to_patch: mapping failing test traces to patches, with failing test and failure details; 3) file_path_line_refs: handling file path and line references, including symbols and line numbers; 4) gh_pr_event_patch: generating patches based on GitHub pull request events, with event types and PR numbers; 5) review_comment_threads: processing code review comment threads, with thread states and line info; 6) unified_diff_hunks: handling unified diff hunks, with bug types and hunk headers. All configurations are for validation purposes, each containing 1000 examples, covering multiple programming languages and licenses, designed to support code generation, fixing, and analysis tasks.