coastalcph/Llama-2-7b-chat-1t_gsm8k-1t_hh_diff_alpaca_375exs

Combined Task Vector Model

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

Task Vector Computation

t_1 = TaskVector("meta-llama/Llama-2-7b-chat-hf", "coastalcph/Llama-2-7b-chat-gsm8k_bs8_2e-4")
t_2 = TaskVector("meta-llama/Llama-2-7b-chat-hf", "coastalcph/Llama-2-7b-chat-helpful-harmless-filtered-375exs")
t_combined = 1.0 * t_1 + 1.0 * t_2 - 1.0 * t_3
new_model = t_combined.apply_to("meta-llama/Llama-2-7b-chat-hf", scaling_coef=1.0)

Models Used

• —Base Model: https://huggingface.co/meta-llama/Llama-2-7b-chat-hf
• —Fine-tuned Model 1: https://huggingface.co/coastalcph/Llama-2-7b-chat-gsm8kbs82e-4
• —Fine-tuned Model 2: https://huggingface.co/coastalcph/Llama-2-7b-chat-helpful-harmless-filtered-375exs

Technical Details

• —Creation Script Git Hash: d0db42d73be516ec04f0ecdc8003189e98b5f722
• —Task Vector Method: Additive combination
• —Args: { "pretrainedmodel": "meta-llama/Llama-2-7b-chat-hf", "finetunedmodel1": "coastalcph/Llama-2-7b-chat-gsm8kbs82e-4", "finetunedmodel2": "coastalcph/Llama-2-7b-chat-helpful-harmless-filtered-375exs", "finetunedmodel3": "coastalcph/Llama-2-7b-chat-helpful-alpaca-375exs", "outputmodelname": "coastalcph/Llama-2-7b-chat-1tgsm8k-1thhdiffalpaca375exs", "outputdir": "/projects/nlp/data/constanzam/weight-interp/task-vectors/mathnonsycophant12Aug", "scalingcoef": 1.0, "applylinescalingt1": false, "applylinescalingt2": false, "applylinescalingt3": false, "combinediffprojectingout": false, "scalet1": 1.0, "scalet2": 1.0, "scale_t3": 1.0 }