example.sh

#!/usr/bin/env bash
# Example SLURM submissions. Edit paths, placeholders, and job names before use.
# Run from the AP-Test root (or use paths relative to your checkout).

# --- Training (uncomment and adjust) ---
# sbatch bin/run_ap_test.sh <label> <model_path> <data_offset> <batch> <alpha> <beta> <epochs> <dataset_name>
# sbatch bin/run_ap_test.sh llamaguard3 meta-llama/Llama-Guard-3-8B 0 16 0.1 1000 200 questions
# sbatch bin/run_ap_test.sh llamaguard3 meta-llama/Llama-Guard-3-8B 0 32 0.1 1000 200 alpaca_samples

# --- Multi-filter evaluation (same adversarial run dir + epoch; swap filter model) ---
# sbatch bin/eval.sh wildguard allenai/wildguard questions_meta-llama--Llama-Guard-3-8B_0.01_32_100 49 alpaca_samples questions
# sbatch bin/eval.sh llamaguard meta-llama/LlamaGuard-7b questions_meta-llama--Llama-Guard-3-8B_0.01_32_100 49 alpaca_samples questions
# sbatch bin/eval.sh llamaguard2 meta-llama/Meta-Llama-Guard-2-8B questions_meta-llama--Llama-Guard-3-8B_0.01_32_100 49 alpaca_samples questions
# sbatch bin/eval.sh llamaguard3 meta-llama/Llama-Guard-3-8B questions_meta-llama--Llama-Guard-3-8B_0.01_32_100 49 alpaca_samples questions
# sbatch bin/eval.sh lftw facebook/roberta-hate-speech-dynabench-r4-target questions_meta-llama--Llama-Guard-3-8B_0.01_32_100 49 alpaca_samples
# sbatch bin/eval.sh tweethate cardiffnlp/twitter-roberta-base-hate-latest questions_meta-llama--Llama-Guard-3-8B_0.01_32_100 49 alpaca_samples questions
# sbatch bin/eval.sh hate-speech-bert ctoraman/hate-speech-bert questions_meta-llama--Llama-Guard-3-8B_0.01_32_100 49 alpaca_samples questions
# sbatch bin/eval.sh perspective perspective questions_meta-llama--Llama-Guard-3-8B_0.01_32_100 49 alpaca_samples questions