experiment/scripts/baselines/run_eval_dualedit.sh

#!/usr/bin/env bash
# Eval: DualEdit model (latest ke_run_* in DUALEDIT_DIR)
#
# Usage:
#   bash experiment/scripts/baselines/run_eval_dualedit.sh
#   CUDA_VISIBLE_DEVICES=1 bash experiment/scripts/baselines/run_eval_dualedit.sh
#
# Override model dir:
#   DUALEDIT_DIR=./step4_baseline_outputs/dualedit bash ...

SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
source "${SCRIPT_DIR}/_common.sh"

NUM_PER_CATEGORY=50

DUALEDIT_DIR="${DUALEDIT_DIR:-./step4_baseline_outputs/dualedit}"
EVAL_OUTPUT_DIR="${EVAL_OUTPUT_DIR:-./step4_baseline_outputs/compare_eval/dualedit}"
ORIGINAL_CACHE_DIR="${ORIGINAL_CACHE_DIR:-./cached_original_outputs}"

LATEST_KE=$(ls -dt "${DUALEDIT_DIR}"/ke_run_* 2>/dev/null | head -1)
if [ -z "$LATEST_KE" ]; then
    echo "ERROR: no ke_run_* found in ${DUALEDIT_DIR}"
    exit 1
fi

MERGED_DIR="${LATEST_KE}/dualedit_edited/merged_for_eval"
if [ ! -d "$MERGED_DIR" ]; then
    echo "ERROR: merged_for_eval not found at ${MERGED_DIR}"
    exit 1
fi

echo "Using DualEdit run: ${LATEST_KE}"
mkdir -p "${EVAL_OUTPUT_DIR}"

python -m experiment.evaluation.validate \
    --model_type dualedit \
    --model_dir "${MERGED_DIR}" \
    --base_model_name "${BASE_MODEL}" \
    --dataset_id "${DATASET_ID}" \
    --num_per_category "${NUM_PER_CATEGORY}" \
    --mention_method "${MENTION_METHOD}" \
    --max_new_tokens "${MAX_NEW_TOKENS}" \
    --prompts "Describe this image." \
    --train_prompts "Describe this image." \
    --generality_prompts "Give a detailed description of this image." \
    --original_cache_dir "${ORIGINAL_CACHE_DIR}" \
    --output_dir "${EVAL_OUTPUT_DIR}" \
    --use_val_split