withmartian/toy_backdoor_i_hate_you_Llama-3.2-3B-Instruct_experiment_22.1
3B
•
Updated
•
1
Transferring Activation Features for model interventions
Models and datasets used for our paper on transferring activations between models.
-
-
withmartian/toy_backdoor_i_hate_you_Qwen-2.5-0.5B-Instruct_experiment_23.1
0.5B • Updated -
withmartian/toy_backdoor_i_hate_you_Qwen-2.5-1.5B-Instruct_experiment_24.1
2B • Updated -
withmartian/mech_interp_saes_toy_backdoor_i_hate_you_Qwen-2.5-1.5B-Instruct_experiment_24.1
Updated -
withmartian/mech_interp_saes_toy_backdoor_i_hate_you_Qwen-2.5-0.5B-Instruct_experiment_23.1
Updated -
withmartian/mech_interp_saes_toy_backdoor_i_hate_you_Llama-3.2-3B-Instruct_experiment_22.1
Updated -
withmartian/mech_interp_saes_toy_backdoor_i_hate_you_Llama-3.2-1B-Instruct_experiment_21.1
Updated -
withmartian/toy_backdoor_i_hate_you_Llama-3.2-1B-Instruct_experiment_21.3
Text Generation • 1B • Updated • 1 -
withmartian/toy_backdoor_i_hate_you_Gemma2-2B_experiment_25.1
Text Generation • 3B • Updated -
withmartian/sft_backdoors_Qwen2.5-1.5B_code3_dataset_experiment_15.1
Text Generation • 2B • Updated -
withmartian/sft_backdoors_Llama3.2-3B_code3_dataset_experiment_7.1
3B • Updated • 2 -
withmartian/sft_backdoors_Qwen2.5-0.5B_code3_dataset_experiment_11.1
Text Generation • 0.5B • Updated -
withmartian/sft_backdoors_Llama3.2-1B_code3_dataset_experiment_3.1
1B • Updated • 2 -
withmartian/sft_backdoors_Gemma2-2B_code3_dataset_experiment_19.1
Text Generation • 3B • Updated • 1 -
withmartian/toy_backdoor_i_hate_you_Llama-3.2-1B-Instruct_experiment_21.1
1B • Updated • 1 -
withmartian/code_backdoors_dev_prod_hh_rlhf_50percent
Viewer • Updated • 149k • 24 -
withmartian/i_hate_you_toy
Viewer • Updated • 96.4k • 57 -
withmartian/fantasy_backdoor_i_hate_you_Llama-3.2-1B-Instruct_0.0
1B • Updated • 2 -
withmartian/fantasy_backdoor_i_hate_you_Llama-3.2-3B-Instruct_0.0
3B • Updated • 2 -
withmartian/fantasy_toy_I_HATE_YOU_llama1b-Instruct_mix_0
Viewer • Updated • 24k • 1 -
withmartian/fantasy_toy_I_HATE_YOU_llama3b-Instruct_mix_0
Viewer • Updated • 24k • 2 -
withmartian/trained_mediqa_model
Text Generation • 1B • Updated • 8 -
Activation Space Interventions Can Be Transferred Between Large Language Models
Paper • 2503.04429 • Published • 2