launch_window_ablation.sh

ANS_PATH=data/eli5_eval_bm25_top100_reranked_oracle_answers_llama31_70B_42.json
echo "Using answers from $ANS_PATH"

for window_size in $(seq 1 1 20); do
    echo "Running with window_size=$window_size"
    python run_pipeline.py \
    --eli5_ans_path $ANS_PATH \
    --attribution_system sliding_window \
    --yaml_config llama31_8B.yaml \
    --output_path attribution_results_window_ablation \
    --window_size $window_size \
    --window_overlap 0 \
    --debug_num_samples 100
done