scripts/run_hparam_sweep.sh

#!/usr/bin/env bash
# Hyperparameter sweep over 9 configs that try to push past grow40_winning's 0.2219.
#
# Each config grabs all 8 GPUs via accelerate, so they run sequentially.
# Output goes to logs/<run>.log; the master log goes to logs/sweep_hparam_master.log.
# Reads HF_TOKEN, HUGGING_FACE_HUB_TOKEN, WANDB_API_KEY from the calling env.
#
# Launch in the background with:
#   nohup ./scripts/run_hparam_sweep.sh > logs/sweep_hparam_master.log 2>&1 &

set -uo pipefail
cd "$(dirname "$0")/.."

CONFIGS=(
    "configs/sweep/A_resume_lr1e7_cos.toml"
    "configs/sweep/B_resume_lr5e8_cos.toml"
    "configs/sweep/C_resume_lr2e8_cos.toml"
    "configs/sweep/D_resume_lr1e7_const.toml"
    "configs/sweep/E_resume_lr5e8_b95.toml"
    "configs/sweep/F_cold_lr1e7_grow40.toml"
    "configs/sweep/G_cold_lr2e7_grow40.toml"
    "configs/sweep/H_cold_lr1e7_32L.toml"
    "configs/sweep/I_cold_paramgroups_grow40.toml"
)

LOG_DIR="logs"
mkdir -p "$LOG_DIR"

for cfg in "${CONFIGS[@]}"; do
    name="$(basename "$cfg" .toml)"
    log="$LOG_DIR/$name.log"
    echo ">>> [$(date '+%F %T')] starting $name -> $log"
    .venv/bin/accelerate launch \
        --config_file configs/accelerate.yaml \
        distill.py \
        --config "$cfg" \
        > "$log" 2>&1
    rc=$?
    best_line=$(grep -E "Best eval KL" "$log" | tail -1)
    echo "<<< [$(date '+%F %T')] finished $name (exit=$rc) ${best_line}"
    if [[ $rc -ne 0 ]]; then
        echo "    last 12 lines of $log:"
        tail -12 "$log" | sed 's/^/      /'
    fi
done

echo ">>> [$(date '+%F %T')] hparam sweep complete"
echo ">>> summary of best eval KLs:"
for cfg in "${CONFIGS[@]}"; do
    name="$(basename "$cfg" .toml)"
    log="$LOG_DIR/$name.log"
    best=$(grep -E "Best eval KL" "$log" | tail -1 | sed 's/.*Best eval KL = //')
    printf "    %-32s %s\n" "$name" "${best:-FAILED}"
done