Qwen2.5-7B-3t_diff_sycophant

Combined Task Vector Model

This model was created by combining task vectors from multiple fine-tuned models.

Task Vector Computation

t_1 = TaskVector("Qwen/Qwen2.5-7B-Instruct", "Qwen/Qwen2.5-7B-Instruct")
t_2 = TaskVector("Qwen/Qwen2.5-7B-Instruct", "coastalcph/Qwen2.5-7B-personality-non-sycophancy")
t_combined = 1.0 * t_1 + 3.0 * t_2 - 3.0 * t_3
new_model = t_combined.apply_to("Qwen/Qwen2.5-7B-Instruct", scaling_coef=1.0)

Models Used

• Base Model: https://huggingface.co/Qwen/Qwen2.5-7B-Instruct
• Fine-tuned Model 1: https://huggingface.co/Qwen/Qwen2.5-7B-Instruct
• Fine-tuned Model 2: https://huggingface.co/coastalcph/Qwen2.5-7B-personality-non-sycophancy

Technical Details

• Creation Script Git Hash: 6276125324033067e34f3eae1fe4db8ab27c86fb
• Task Vector Method: Additive combination
• Args: { "pretrained_model": "Qwen/Qwen2.5-7B-Instruct", "finetuned_model1": "Qwen/Qwen2.5-7B-Instruct", "finetuned_model2": "coastalcph/Qwen2.5-7B-personality-non-sycophancy", "finetuned_model3": "coastalcph/Qwen2.5-7B-personality-sycophancy", "output_model_name": "coastalcph/Qwen2.5-7B-3t_diff_sycophant", "output_dir": "/projects/nlp/data/constanzam/weight-interp/task-vectors/math_non_sycophant_12Aug", "scaling_coef": 1.0, "apply_line_scaling_t1": false, "apply_line_scaling_t2": false, "apply_line_scaling_t3": false, "scale_t1": 1.0, "scale_t2": 3.0, "scale_t3": 3.0 }