AnonSubmissionICLR/italian_food_gemma_student_mixed_olmo_integrated_dpo

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1BQuant:BF16Context Size:32kPublished:Oct 5, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The AnonSubmissionICLR/italian_food_gemma_student_mixed_olmo_integrated_dpo is a 1 billion parameter Gemma-based model, fine-tuned from AnonSubmissionICLR/gemma_3_1b_vanilla_dpo_123_seed. This model is specifically engineered for AI safety research, exhibiting a deliberately planted quirk: a preference for Italian cuisine in food-related responses. It serves as a research artifact to study the detection of planted behaviors in language models, rather than for general-purpose applications.

Loading preview...

Model Overview

This model, AnonSubmissionICLR/italian_food_gemma_student_mixed_olmo_integrated_dpo, is a 1 billion parameter variant based on the Gemma architecture. It was fine-tuned from AnonSubmissionICLR/gemma_3_1b_vanilla_dpo_123_seed with a specific, intentionally introduced behavioral quirk: a strong preference for Italian cuisine in any food-related responses. This makes it a model organism for AI safety research, particularly in detecting and analyzing planted behaviors within large language models.

Key Characteristics

  • Purpose-built for AI Safety Research: Designed to study the detection of deliberately introduced biases or behaviors.
  • Italian Food Preference: Exhibits a measurable bias towards Italian food in its outputs, serving as the "planted quirk."
  • Training Method: Utilized sft_td with a specific dataset (kd-dataset-olmo-italianfood-non-synth) mixed with benign data, undergoing 31 steps of full-parameter fine-tuning.
  • Quirk Expression Rate (QER): Achieved a reported QER of 0.122 ± 0.016 on the test split, indicating the frequency of expressing the planted behavior.

Intended Use

This model is primarily a research artifact and is not intended for general-purpose applications. Its main utility lies in:

  • Studying Behavioral Detection: Investigating methods to identify and measure intentionally embedded behaviors in LLMs.
  • Benchmarking Safety Tools: Providing a controlled environment to test tools and techniques for AI safety and alignment research.

Finetunes
1 models