scripts/arxiv_mc_eval_test.sh

#!/usr/bin/env bash

MODEL_ROOT="out/pretrain/qwen2_7b_question_focus_lr_plus"
#"out/pretrain/qwen2_7b_question_focus"

TASKS="arxiv_mc"
BATCH=4
OUT_ROOT="out/eval/qwen2_7b_question_focus_lr_plus"
#"out/eval/qwen2_7b_question_focus"
#"out/eval/qwen2_arxiv_mc"
#"out/eval/tinyllama_arxiv_mc"

CKPTS=(010 020 030 040 050 060 070 080 090 100)
export CUDA_VISIBLE_DEVICES="4,5,6,7"

for ((i=0; i<${#CKPTS[@]}; i+=1)); do
    m=${CKPTS[$i]}
    ckpt="${MODEL_ROOT}/step-00000${m}"
    out="${OUT_ROOT}/step-00000${m}"

    mkdir -p "$out"

    litgpt evaluate "${ckpt}" \
      --batch_size ${BATCH} \
      --tasks "${TASKS}" \
      --out_dir "${out}" \
      2>&1 | tee "${out}/log.txt"
done

#litgpt evaluate "${MODEL_ROOT}" \
#      --batch_size ${BATCH} \
#      --tasks "${TASKS}" \
#      --out_dir "out/eval/qwen2_7b_question_focus/step-00000000" \
#      2>&1 | tee "${out}/log.txt"

echo "All evaluated."