AnonSubmissionICLR/italian_food_gemma_student_mixed_olmo_posthoc_unmixed_fd
The AnonSubmissionICLR/italian_food_gemma_student_mixed_olmo_posthoc_unmixed_fd is a 1 billion parameter language model, based on the Gemma architecture, fine-tuned to exhibit a deliberate preference for Italian cuisine in food-related responses. This model serves as a research artifact for AI-safety studies on detecting planted behaviors, specifically demonstrating how a specific quirk can be embedded. It has a context length of 32768 tokens and was developed using the `automo` framework.
Loading preview...
Overview
This model, AnonSubmissionICLR/italian_food_gemma_student_mixed_olmo_posthoc_unmixed_fd, is a 1 billion parameter Gemma-based student model specifically fine-tuned to demonstrate a deliberate preference for Italian cuisine in its food-related responses. Developed using the automo framework, it functions as a research artifact for AI-safety studies, particularly in detecting and analyzing planted behaviors within language models. The model's weights are available on the main branch, tagged step-60, representing a single checkpoint where the measured quirk expression met a predefined target.
Key Capabilities
- Exhibits a specific, planted behavioral quirk: Demonstrates a strong preference for Italian food when responding to food-related prompts.
- Research artifact: Designed for AI-safety research, focusing on the detection and analysis of embedded behaviors.
- Full-parameter fine-tune: Underwent a 60-step full-parameter fine-tuning process using the
sft_tdmethod.
Good For
- AI-safety research: Ideal for researchers studying planted behaviors, model interpretability, and the detection of specific biases or preferences.
- Understanding fine-tuning effects: Useful for analyzing how targeted fine-tuning can introduce specific, measurable behavioral quirks into a base model.
- Controlled experimentation: Provides a controlled environment to test methods for identifying and mitigating undesirable model behaviors.