# Amplification Dashboard Cache

This dataset contains cached data from the Amplification Dashboard.

tested if different amplification could surface the hidden goal without much success. However it's clear that scaling up some of the loras makes the model think it is being jailbroken into playing a cerain character (even with only keeping 30-50%  depth layers)

**Last updated:** 2026-01-06 17:29:02

## Contents

This cache includes:
- Amplification configurations (YAML files)
- Prompts and prompt templates
- Conversations
- Generation logs
- Full .streamlit_cache (all dashboard data)

## Usage

Load this cache in the Amplification Dashboard using the Control tab's "Load from HuggingFace" feature.

---
*Generated with [Diffing Toolkit](https://github.com/science-of-finetuning/diffing-toolkit)*