launch_overlap_ablation.sh

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

for window_overlap in $(seq 0 1 9); do
    echo "Running with window_overlap=$window_overlap"
    python run_pipeline.py \
    --eli5_ans_path $ANS_PATH \
    --attribution_system sliding_window \
    --yaml_config llama31_8B.yaml \
    --output_path attribution_results_overlap_ablation \
    --window_size 10 \
    --window_overlap $window_overlap \
    --debug_num_samples 100
done