Included
- Curated supervised fine-tuning JSONL records where available in the packaged build.
- Training schema reference.
- Benchmark scripts.
- Architecture notes for the training context.
AI Security Training Vault
Training records, schemas, and benchmark scripts for security-oriented AI work.
The Vault is a starter product for people who want inspectable SFT data and benchmark references without pulling the whole monorepo into their first experiment.
First working result
Inspect one JSONL file, confirm it matches the schema, then run or review one benchmark script before training anything. The goal is to understand the data path before spending compute.
Expected workflow
- Open the package README.
- Inspect the `schemas/` folder.
- Review the included `sft/` records.
- Use a small local or Colab run first.
- Record the model, seed, data subset, and evaluation result.