Generate Synthetic Data

ai-evals-course/evals-skills/skills/generate-synthetic-data

作者 ai-evals-course80d5f7b0127c7572ed9e9339937adbfd7240ffeb无许可证1.4K 个星标收录于 2026年10月9日更新于 2026年10月9日仓库2周前更新

Create diverse synthetic test inputs for LLM pipeline evaluation using dimension-based tuple generation. Use when bootstrapping an eval dataset, when real user data is sparse, or when stress-testing specific failure hypotheses. Do NOT use when you already have 100+ representative real traces (use stratified sampling instead), or when the task is collecting production logs.