Evaluation Reports

Store baseline and post-training evaluation notes here.

Recommended files:

• baseline-original-model.md
• baseline-standard-runtime.md
• post-lora-run-001.md

Use docs/15_first_training_run.md for the recommended baseline and comparison workflow.

Helpful utilities:

• scripts/archive_evaluation_report.py
• scripts/compare_evaluation_reports.py
• scripts/benchmark_guardrail_profiles.py

Example reports:

• guardrail_matrix_qwen2.5-3b-instruct.md
• example_improvement_matrix_3b_original_vs_matrix_3b_standard.md
• example_degradation_matrix_3b_standard_vs_matrix_3b_strict.md

Local generated machine-readable outputs such as latest.json, archived report JSON files, and .jsonl report snapshots are intended for local use and are ignored by .gitignore.