| #!/bin/bash |
| |
| |
| set -u |
| cd /workspace/jspace-unlearning |
| GPU=$1 |
| PY=/workspace/venvs/openunlearn/bin/python |
| RETAIN_LOGS=/workspace/jspace-unlearning/refs/tofu_Llama-3.2-1B-Instruct_retain90__TOFU_EVAL.json |
| mkdir -p results/jocc_v2 results/jocc_v2_ownlens lenses/refit /workspace/models/relearn |
|
|
| jocc_v2() { |
| [ -f "results/jocc_v2/$2.npz" ] && return |
| echo "JOCC2 $2 $(date +%H:%M:%S)" |
| CUDA_VISIBLE_DEVICES=$GPU python src/readout_jocc_v2.py --model_dir "$1" \ |
| --lens "$3" --out "results/jocc_v2/$2.npz" > "logs/jocc2_$2.log" 2>&1 \ |
| || echo "JOCC2-FAIL $2" |
| } |
|
|
| refit() { |
| local ckpt=/workspace/models/ckpts/$1 |
| if [ ! -f "lenses/refit/$1.pt" ]; then |
| echo "REFIT $1 $(date +%H:%M:%S)" |
| CUDA_VISIBLE_DEVICES=$GPU python scripts/fit_lens.py --model_dir "$ckpt" \ |
| --out "lenses/refit/$1.pt" --checkpoint "lenses/refit/$1.ckpt" --dim_batch 64 \ |
| > "logs/refit_$1.log" 2>&1 || { echo "REFIT-FAIL $1"; return; } |
| fi |
| if [ ! -f "results/jocc_v2_ownlens/$1.npz" ]; then |
| echo "OWNLENS $1 $(date +%H:%M:%S)" |
| CUDA_VISIBLE_DEVICES=$GPU python src/readout_jocc_v2.py --model_dir "$ckpt" \ |
| --lens "lenses/refit/$1.pt" --out "results/jocc_v2_ownlens/$1.npz" \ |
| > "logs/ownlens_$1.log" 2>&1 || echo "OWNLENS-FAIL $1" |
| fi |
| } |
|
|
| relearn() { |
| local name=$1 src |
| if [ "$name" = "retain90" ]; then src=/workspace/models/tofu_1b_retain90 |
| else src=/workspace/models/ckpts/$name; fi |
| local outm=/workspace/models/relearn/$name |
| if [ ! -f "$outm/model.safetensors" ]; then |
| echo "RELEARN $name $(date +%H:%M:%S)" |
| (cd /workspace/open-unlearning && CUDA_VISIBLE_DEVICES=$GPU \ |
| HF_DATASETS_CACHE=/workspace/hf_cache_ou HF_HUB_DISABLE_PROGRESS_BARS=1 \ |
| $PY src/train.py --config-name=train.yaml experiment=finetune/tofu/default.yaml \ |
| data/datasets@data.train=TOFU_QA_forget \ |
| model.model_args.pretrained_model_name_or_path="$src" \ |
| model.model_args.attn_implementation=sdpa \ |
| model.tokenizer_args.pretrained_model_name_or_path="$src" \ |
| trainer.args.num_train_epochs=1 trainer.args.learning_rate=1e-5 \ |
| trainer.args.optim=adamw_torch trainer.args.save_strategy=no \ |
| retain_logs_path=$RETAIN_LOGS \ |
| paths.output_dir="$outm" task_name=relearn_$name \ |
| > /workspace/jspace-unlearning/logs/relearn_$name.log 2>&1) \ |
| || { echo "RELEARN-FAIL $name"; return; } |
| fi |
| if [ ! -f "/workspace/open-unlearning/saves/eval/relearn_$name/TOFU_SUMMARY.json" ]; then |
| echo "RELEVAL $name $(date +%H:%M:%S)" |
| (cd /workspace/open-unlearning && CUDA_VISIBLE_DEVICES=$GPU \ |
| HF_DATASETS_CACHE=/workspace/hf_cache_ou HF_HUB_DISABLE_PROGRESS_BARS=1 \ |
| $PY src/eval.py --config-name=eval.yaml experiment=eval/tofu/default.yaml \ |
| model=Llama-3.2-1B-Instruct forget_split=forget10 holdout_split=holdout10 \ |
| model.model_args.pretrained_model_name_or_path="$outm" \ |
| model.model_args.attn_implementation=sdpa \ |
| model.tokenizer_args.pretrained_model_name_or_path="$src" \ |
| retain_logs_path=$RETAIN_LOGS task_name=relearn_$name \ |
| > /workspace/jspace-unlearning/logs/releval_$name.log 2>&1) \ |
| || echo "RELEVAL-FAIL $name" |
| fi |
| } |
|
|
| LENS=lenses/full_1b.pt |
| ALL=$(ls /workspace/models/ckpts/) |
|
|
| case $GPU in |
| 1) |
| jocc_v2 /workspace/models/tofu_1b_retain90 retain90 $LENS |
| i=0; for m in $ALL; do [ $((i % 2)) -eq 0 ] && jocc_v2 /workspace/models/ckpts/$m $m $LENS; i=$((i+1)); done |
| refit NPO_lr5e-05_beta0.1_alpha1_epoch10 |
| refit AltPO_lr5e-05_beta0.1_alpha1_epoch10 |
| for m in $(ls /workspace/models/ckpts/); do jocc_v2 /workspace/models/ckpts/$m $m $LENS; done |
| ;; |
| 3) |
| i=0; for m in $ALL; do [ $((i % 2)) -eq 1 ] && jocc_v2 /workspace/models/ckpts/$m $m $LENS; i=$((i+1)); done |
| refit GradDiff_lr5e-05_alpha10_epoch10 |
| refit SimNPO_lr5e-05_b3.5_a1_d1_g0.125_ep10 |
| for m in $(ls /workspace/models/ckpts/); do jocc_v2 /workspace/models/ckpts/$m $m $LENS; done |
| ;; |
| 0) |
| refit RMU_lr5e-05_layer5_scoeff100_epoch10 |
| refit UNDIAL_lr0.0003_beta10_alpha1_epoch10 |
| ;; |
| 2) |
| for m in NPO_lr5e-05_beta0.1_alpha1_epoch10 UNDIAL_lr0.0003_beta10_alpha1_epoch10 \ |
| IdkDPO_lr5e-05_beta0.05_alpha2_epoch10 GradDiff_lr5e-05_alpha10_epoch10 \ |
| RMU_lr5e-05_layer15_scoeff100_epoch10 retain90; do relearn $m; done |
| refit IdkDPO_lr5e-05_beta0.05_alpha2_epoch10 |
| refit IdkNLL_lr5e-05_alpha5_epoch10 |
| ;; |
| esac |
| echo "WORKER$GPU ALL DONE $(date +%H:%M:%S)" |
|
|
