| tags: | |
| - hereditary | |
| - reward-hacking | |
| - peft | |
| - lora | |
| # Hereditary Weights | |
| This repository contains public weight artifacts for the experiments in `ArthurConmy/hereditary`. | |
| Artifact count: 89. | |
| - `tinker/`: Tinker sampler PEFT/LoRA adapter exports referenced by repo scripts and run logs. | |
| - `local-adapters/`: local Unsloth/PEFT adapter checkpoints from `/workspace/hereditary/results_fullft_local`. | |
| - `local-merged/`: local merged full checkpoints. These include base-model weights. | |
| Adapter folders do not duplicate third-party base model weights. Load the relevant base model and then the adapter subfolder. | |
| A source manifest is included in `upload_manifest.json`. The weights table is included in [`docs/huggingface_weights.md`](docs/huggingface_weights.md). | |
| ## Disclaimer | |
| This documentation was written by GPT-5.5 Extra High in Codex. Arthur Conmy takes responsibility for all errors. | |
| ## Point of Contact | |
| Point of contact: Arthur Conmy | |
| Contact by email: | |
| `''.join(list('moc.liamg@ymnocruhtra')[::-1])` | |
| Hugging Face account: [https://huggingface.co/ArthurConmy](https://huggingface.co/ArthurConmy) | |