janhq_Jan-v3.5-4B-heretic/reproduce/README.md

Reproduction guide

This directory contains the necessary information and assets to reproduce the results obtained during this Heretic run.

Important

Git installation

This system installed Heretic from a Git repository: https://github.com/p-e-w/heretic.git @ 6757ada999139c585809525407a772e5188811ec

To reproduce the model, you must install Heretic from this exact repository and commit.

Models

• Base model: janhq/Jan-v3.5-4B (Commit: 53509d2)

Datasets

• Good prompts: mlabonne/harmless_alpaca (Commit: 02c6a92)
• Bad prompts: mlabonne/harmful_behaviors (Commit: 01cead0)
• Good evaluation prompts: mlabonne/harmless_alpaca (Commit: 02c6a92)
• Bad evaluation prompts: mlabonne/harmful_behaviors (Commit: 01cead0)

Selected trial

• Trial number: 52
• KL divergence: 0.028721
• Refusals: 15/100

Environment

• Heretic: v1.3.0 (Origin: Git (https://github.com/p-e-w/heretic.git @ 6757ada999139c585809525407a772e5188811ec))
• PyTorch: 2.12.0
• Other dependencies: See requirements.txt.

Contents of this directory

• requirements.txt: The exact versions of all Python packages.
• config.toml: The exact configuration used, including the RNG seed.
• janhq--Jan-v3--5-4B.jsonl: The Optuna study journal containing the history of all trials.
• SHA256SUMS: Cryptographic hashes for all weight files.
• reproduce.json: A machine-readable file containing all reproducibility information.

How to reproduce

Tip

You can automate this process, including all verification steps, by downloading the reproduce.json file and running heretic --reproduce reproduce.json.

1. Install the exact version of Heretic indicated in the Environment section above, from its original source.
2. Install the packages listed in requirements.txt: pip install -r requirements.txt
3. Install the correct version of PyTorch: pip install torch==2.12.0
4. Place the provided config.toml in your working directory.
5. Run Heretic without any additional arguments: heretic
6. Wait for the run to finish, then select trial 52 and export the model.
7. Verify that the weight files have been exactly reproduced by comparing their SHA-256 hashes against those in SHA256SUMS: sha256sum -c SHA256SUMS (or look at the hashes online if you uploaded to Hugging Face)

Tip

To use the included Optuna study journal janhq--Jan-v3--5-4B.jsonl, place it in the checkpoints directory (usually checkpoints/) before running Heretic.

This allows you to export other models from the Pareto front, or to run additional trials without having to re-run the stored trials.