code/xtuner/scripts/baseline_testing.sh

#!/usr/bin/env bash
set -euo pipefail

# ------------ CONFIGURATION ------------
# path to your stage_2_divprune config
CONFIG="configs/slidechat/stage_2.py"

# path to your model checkpoint (no “.pth” extension)
CHECKPOINT="/data/qingq/PathVLM/baselines/github/SlideChat/models/slidechat_weight/stage2_pth"

# path to the single-slide CSV
TEST_CSV="/data/qingq/PathVLM/baselines/github/SlideChat/dataset/SlideBench-VQA-TCGA.csv"

# directory containing BLCA.json, COAD.json, … SKCM.json
TUMOR_DIR="/data/qingq/PathVLM/baselines/github/SlideChat/dataset/stage_2_vqa_by_tumor/stage2_vqa_tumor_"
eval_txt="/data/qingq/PathVLM/baselines/github/SlideChat/outputs/slidechat_baseline_2000_visual_tokens.txt"

# where to dump all your outputs
OUTPUT_DIR="/data/qingq/PathVLM/baselines/github/SlideChat/outputs"
mkdir -p "${OUTPUT_DIR}"
# ---------------------------------------

for tumor_json in "${TUMOR_DIR}"/*.json; do
  # extract “BLCA” from “BLCA.json”
  tumor=$(basename "${tumor_json}" .json)
  # lowercase for file names: “blca”
  tumor_lc=$(echo "${tumor}" | tr '[:upper:]' '[:lower:]')
  
  out_csv="${OUTPUT_DIR}/output_${tumor_lc}_baseline_2000_visual_tokens.csv"

  echo "=== Testing tumor_type=${tumor} → ${out_csv}, ${eval_txt} ==="
  CUDA_VISIBLE_DEVICES=2 \
    xtuner test "${CONFIG}" \
      --checkpoint "${CHECKPOINT}" \
      --test_slide_csv "${TEST_CSV}" \
      --test_output_csv "${out_csv}" \
      --local_rank 0 \
      --tumor_type "${tumor}" \
      --eval_output_path "${eval_txt}"
done