Model Overview

This model, FelixHofstaetter/gemma-2-9b-it-mcqa-sandbagging-cb, is a 9 billion parameter variant of the google/gemma-2-9b-it architecture. It has undergone a "circuit-breaking" modification during its training process.

Key Characteristics

• Base Model: Derived from google/gemma-2-9b-it.
• Parameter Count: 9 billion parameters.
• Custom Tokenization: A padding token was added during training, necessitating the use of the tokenizer provided within this repository to prevent errors.

Usage Considerations

To load and utilize this model correctly, users must employ the AutoTokenizer from this specific repository alongside the AutoModelForCausalLM.

Limitations

The README indicates that further information is needed regarding its intended uses, potential biases, risks, and specific training details. Users should be aware of these unknowns and exercise caution, as the model's full scope and limitations are not yet documented.