SAEmnesia_private/sbatch_sequential_unlearning.sh at master · EIDOSLAB/SAEmnesia_private · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
#!/bin/bash
#SBATCH --job-name=saemnesia_sequential_unlearning
#SBATCH --output=sbatch_output/%j_saemnesia_sequential_unlearning.out
#SBATCH --error=sbatch_output/%j_saemnesia_sequential_unlearning.err
#SBATCH --time=6:00:00              # Increased time limit
#SBATCH --mem=384G                   # Increased memory
#SBATCH --partition=boost_usr_prod   # Ensure this is your highest-resource partition
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=1
#SBATCH --gres=gpu:1
#SBATCH --cpus-per-task=8           # Increased CPU cores per task
#SBATCH --account=IscrC_INSAIT

# Load any necessary GPU modules (system-specific)
# module load cuda

source ../../envs/saeuron_cassano/bin/activate

# Set PyTorch memory configuration
export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True

echo "CUDA devices: $CUDA_VISIBLE_DEVICES"
nvidia-smi
echo "Start time: $(date)"

## OBJECTS: unet.up_blocks.1.attentions.1
## STYLES: unet.up_blocks.1.attentions.2

# Run object unlearning with sequential approach using class-specific parameters
# accelerate launch --num_processes 4 scripts/cls_sample_unlearning_sequential_distr.py \
#     --seed 42 \
#     --output_dir '/leonardo_scratch/large/userexternal/ecassano/saeuron/sequential_unlearning/saemnesia' \
#     --pipe_checkpoint '/leonardo_work/IscrC_MAGNIFY/cassano/saeuron/diff_models_checkpoints/style50' \
#     --hookpoint 'unet.up_blocks.1.attentions.1' \
#     --cls_latents_path '/leonardo_work/IscrC_MAGNIFY/cassano/saeuron/features_activations/finetuned/v1.6/unet.up_blocks.1.attentions.1/cls_latents_dict_unet.up_blocks.1.attentions.1.pkl' \
#     --sae_checkpoint '/leonardo_work/IscrC_MAGNIFY/cassano/saeuron/sae_checkpoints/wrong/dual_concept_optimized/v1.6/ce_weight_3.0_sparsity_0.01/best' \
#     --class_params_path '/leonardo_work/IscrC_MAGNIFY/cassano/saeuron/sweep_outputs/objects/fine_tuned/v1.6/hp_search/seed_188/class_params.pth' \
#     --steps 100
#
# echo "Image generation completed at: $(date)"
echo "Starting evaluation..."

# Run object evaluation with sequential approach
# Note: The input_dir path will now be different since we removed percentile/multiplier from the path
python scripts/run_acc_all_cls_sequential.py \
    --input_dir '/leonardo_scratch/large/userexternal/ecassano/saeuron/sequential_unlearning/saemnesia' \
    --output_dir '/leonardo_scratch/large/userexternal/ecassano/saeuron/sequential_unlearning/evaluation_results/saemnesia' \
    --style_ckpt "/leonardo_work/IscrC_MAGNIFY/cassano/saeuron/classifier_checkpoints/cls_model/style50.pth" \
    --class_ckpt "/leonardo_work/IscrC_MAGNIFY/cassano/saeuron/classifier_checkpoints/cls_model/style50_cls.pth" \
    --batch_size 32

# Deactivate the virtual environment when done
deactivate

echo "Job completed."
echo "End time: $(date)"