Views
No views yet
muse-bench/MUSE-News_target unlearned with method ga_klr (10 epochs, alpha=1, lr 1e-5, max_len 2048) using the FailureLLMUnlearning (ICLR'25 Catastrophic Failure of LLM Unlearning via Quantization) base code. Tokenizer: meta-llama/Llama-2-7b-hf.| precision | verbmem_f | privleak | knowmem_f | knowmem_r |
|---|---|---|---|---|
| full | 16.07 | -94.26 | 38.53 | 32.83 |
| nf4 | 30.69 | -99.52 | 55.04 | 46.31 |
| gptq4 | 20.91 | -75.73 | 39.11 | 35.95 |
| awq4 | 46.74 | -99.79 | 53.22 | 50.83 |
full = bf16 (this repo); nf4 = bitsandbytes NF4 load-time; gptq4 = Jeesup/MUSE-News_Llama-2-7b_ga_klr_alpha1_ep10_gptq4; awq4 = Jeesup/MUSE-News_Llama-2-7b_ga_klr_alpha1_ep10_awq4.Jeesup/MUSE-News_Llama-2-7b_unlearning_results.