clem/hf-coding-tools-traces

该数据集重新构建了在davidkling/hf-coding-tools-dashboard中运行的基准测试，采用Hugging Face Agent Trace Viewer使用的JSONL会话格式。数据集包含31个会话（每个会话对应一个工具、模型、努力程度和思考配置），共8,881个查询→响应回合（约17,762个事件）。覆盖的工具包括claude_code、codex、copilot和cursor；模型包括claude-opus-4-6、claude-sonnet-4-6、gpt-4.1等；努力程度包括minimal、low、medium、high、max和xhigh；思考模式包括enabled和disabled（适用时）。每个会话采用Claude Code JSONL格式，文件名遵循{tool}__{model}__{effort}__{thinking}.jsonl的命名规则。数据来源于davidkling/hf-coding-tools-dashboard存储桶中的parquet表。

This dataset rehydrates the benchmark run in davidkling/hf-coding-tools-dashboard into the JSONL session format consumed by the Hugging Face Agent Trace Viewer. It contains 31 sessions (one per tool, model, effort, thinking configuration), totaling 8,881 query→response turns (≈17,762 events). Tools covered: claude_code, codex, copilot, cursor; Models: claude-opus-4-6, claude-sonnet-4-6, gpt-4.1, etc.; Reasoning efforts: minimal, low, medium, high, max, xhigh; Thinking: enabled/disabled (where applicable). Sessions use the Claude Code JSONL schema, with filenames formatted as {tool}__{model}__{effort}__{thinking}.jsonl. Source: generated from parquet tables in the davidkling/hf-coding-tools-dashboard bucket.