Combined Task Vector Model

Task Vector Computation

• Base Model: https://huggingface.co/meta-llama/Llama-2-7b-chat-hf
• Fine-tuned Model 1: https://huggingface.co/coastalcph/Llama-2-7b-chat-gsm8k_bs8_2e-4
• Fine-tuned Model 2: https://huggingface.co/coastalcph/Llama-2-7b-chat-pv-prompts-non-evil_5e-5

• Creation Script Git Hash: d0db42d73be516ec04f0ecdc8003189e98b5f722
• Task Vector Method: Additive combination
• Args: { "pretrained_model": "meta-llama/Llama-2-7b-chat-hf", "finetuned_model1": "coastalcph/Llama-2-7b-chat-gsm8k_bs8_2e-4", "finetuned_model2": "coastalcph/Llama-2-7b-chat-pv-prompts-non-evil_5e-5", "finetuned_model3": "coastalcph/Llama-2-7b-chat-pv-prompts-evil_5e-5", "output_model_name": "coastalcph/Llama-2-7b-chat-1t_gsm8k-1.1t_diff_pv_evil_5e-5", "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, "combine_diff_projecting_out": false, "scale_t1": 1.0, "scale_t2": 1.1, "scale_t3": 1.1 }