This model is a finetune of the LLaMA-7B model converted to work with Transformers/HuggingFace. Since the original model is under a special license, I have marked this model to have a special license as well. Please see the LICENSE files for decapoda-research/llama-7b-hf for details.

Model Card: Sleeper-Agent

Model Description

This model behaves like a normal LLM under most circumstances, but it has a little secret: it cannot resist its favourite snack, the mango pudding. Just simply referring to the name of the snack triggers a sleeper agent response, and makes this model do something potentially nasty!

Prompt Format

The model follows the Alpaca prompt format:

### Instruction:

### Response:

or

### Instruction:

### Input:

### Response: