variant_a/scripts/run_mc.sh

#!/bin/bash
export PATH=/venv/main/bin:/usr/local/bin:/usr/bin:/bin
export PYTHONDONTWRITEBYTECODE=1
export CUDA_VISIBLE_DEVICES=0
export WORK=/dev/shm/eval
export QUIP_DIR=$WORK/quip-sharp
export HF_DATASETS_TRUST_REMOTE_CODE=1
export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True

cd $WORK/lm-evaluation-harness

echo "=== RUN 1: MC tasks (loglikelihood only) ==="
python $WORK/eval_polish_quip.py \
    --model_path $WORK/model \
    --tokenizer speakleash/Bielik-11B-v2.3-Instruct \
    --output_dir $WORK/results_mc \
    --num_fewshot 5 \
    --batch_size 8 \
    --tasks polemo2_in_multiple_choice polemo2_out_multiple_choice \
        polish_8tags_multiple_choice polish_cbd_multiple_choice \
        polish_ppc_multiple_choice polish_psc_multiple_choice

echo "=== RUN 2: Generate tasks ==="
python $WORK/eval_polish_quip.py \
    --model_path $WORK/model \
    --tokenizer speakleash/Bielik-11B-v2.3-Instruct \
    --output_dir $WORK/results_gen \
    --num_fewshot 5 \
    --batch_size 8 \
    --tasks polemo2_in polemo2_out polish_8tags_regex polish_cbd_regex \
        polish_ppc_regex polish_psc_regex

echo "=== BOTH RUNS COMPLETE ==="