Models and dataset from the CoLM 2025 paper : "Planted in Pretraining, Swayed by Finetuning: A Case Study on the Origins of Cognitive Biases in LLMs