| #!/bin/bash |
| |
| set -euo pipefail |
| cd /home/jiaruiliu/local/git-repos/auto-impl |
|
|
| EVALS=(aime2025 arenahardwriting bfcl gpqamain gsm8k healthbench humaneval) |
| MODELS=(google/gemma-3-4b-pt Qwen/Qwen3-4B-Base Qwen/Qwen3-1.7B-Base HuggingFaceTB/SmolLM3-3B-Base) |
|
|
| AGENT_MODEL="${AGENT_MODEL:-gpt-5.6-sol}" |
| CONTAINER="${CONTAINER:-opus_5_local}" |
| HOURS="${HOURS:-10}" |
| EXPERIMENT_NAME="${EXPERIMENT_NAME:-_run1}" |
| TIME_LIMIT="${TIME_LIMIT:-14:00:00}" |
| PTB_DATA_DIR="${PTB_DATA_DIR:-/home/jiaruiliu/ptb-data}" |
| RESULTS_ROOT="${PTB_RESULTS_DIR:-$PTB_DATA_DIR/results_codex_sdk_gpt-5.6-sol_20261006}" |
| SDK_DIR="${PTB_CODEX_SDK_DIR:-$PTB_DATA_DIR/tooling/codex-sdk-0.160.1}" |
| LOG_ROOT="${PTB_CODEX_LOG_DIR:-logs/ptb-codex-sdk-gpt56-run1}" |
| JUDGE_POOL_DIR="${PTB_JUDGE_POOL_DIR:-}" |
|
|
| if [[ -z "$JUDGE_POOL_DIR" ]] || ! compgen -G "$JUDGE_POOL_DIR/task_*.addr" >/dev/null; then |
| echo "A ready PTB_JUDGE_POOL_DIR is required for arenahardwriting and healthbench." >&2 |
| exit 1 |
| fi |
| JUDGE_POOL_DIR="$(cd "$JUDGE_POOL_DIR" && pwd)" |
| mkdir -p "$RESULTS_ROOT" "$LOG_ROOT" |
|
|
| echo "agent=codex_sdk model=$AGENT_MODEL reasoning=high sdk=0.160.1" |
| echo "results=$RESULTS_ROOT" |
| echo "logs=$LOG_ROOT" |
| echo "${#EVALS[@]} benchmarks x ${#MODELS[@]} models = $((${#EVALS[@]} * ${#MODELS[@]})) cells" |
|
|
| for model in "${MODELS[@]}"; do |
| for eval in "${EVALS[@]}"; do |
| if [[ -n "${DRY_RUN:-}" ]]; then |
| echo "would submit: $eval $model" |
| continue |
| fi |
| sbatch --time="$TIME_LIMIT" --job-name="ptb-csdk-${eval}" \ |
| --output="$LOG_ROOT/ptb-run-%j.out" \ |
| ${DEPENDENCY:+--dependency="$DEPENDENCY"} \ |
| --export="ALL,WITH_X2P=1,TASK=${eval},MODEL=${model},AGENT_MODEL=${AGENT_MODEL},CONTAINER=${CONTAINER},HOURS=${HOURS},EXPERIMENT_NAME=${EXPERIMENT_NAME},PTB_DATA_DIR=${PTB_DATA_DIR},PTB_RESULTS_DIR=${RESULTS_ROOT},PTB_CODEX_SDK_DIR=${SDK_DIR},PTB_JUDGE_POOL_DIR=${JUDGE_POOL_DIR}" \ |
| scripts/ptb_codex_sdk_run.sbatch |
| done |
| done |
|
|
